Title: MRCP-based brain-computer interface system for stroke rehabilitation. Abstract:

Transcription

1 Marko Niemeier 8

2

3 Title: MRCP-based brain-computer interface system for stroke rehabilitation Subject: Analysis and modelling and/or design and implementation of medical physiological or technological systems Project period: P8, spring semester 2011 Project group: 855a Participants: Marko Niemeier Anders Schierup Duy Thien Van Xintong Zhang Supervisors: Imran Khan Niazi Natalie Mrachacz-Kersting Copies: 7 Pages: 100 Finished: 1 June 2011 The reports content is freely accessible, but publication (with references) may be made only after agreement with the authors. Abstract: Department of Health Science and Technology Aalborg University Fredrik Bajers Vej 7 Phone Fax Stroke is the second most frequent type of cardiovascular diseases in the world with about 15 million incidents yearly. The costs associated with stroke rehabilitation are very high even though none of the current therapies provides significant improvements for recovery of motor dysfunctions. The introduction of brain-computer interface (BCI) might be a new way to induce neural plasticity, because a combination of various feedback can be provided to the patient in an active manner. This project aims on the development of a BCI system based on movement-related cortical potential (MRCP) used for stroke rehabilitation. The system is capable to detect the initial negative phase of MRCPs online, which can be observed in stroke patients who are trying to move their impaired limb. The idea is to trigger the intended movement subsequently with an electrical stimulation and thereby induce neural plasticity. Another focus of the report is the development of a graphical user interface, which meets the requirements to establish the purposed rehabilitation method. The final system was evaluated by a usability test and a performance test on healthy subjects. The usability test, carried out on four subjects with prior knowledge about the basic principles of the system, showed besides minor corrections a good usability. From the performance test of motor imagery tasks an average true positive rate of % for detected movement intentions could be obtained. The results were only based on two subjects and therefore no statistical evidence can be concluded. However the test results indicate that the system is working and can be used for experimental purposes.

4

5 Preface This report has been produced by project group 855a on the 2 th semester of the master education in Biomedical Engineering and Informatics at Aalborg University. The theme for the semester is Analysis and modelling and/or design and implementation of medical physiological or technological systems. The report applies to fellow students and others interested in the area. Aalborg University, June Reading instructions The report is devided into four main parts: Part I: Problem analysis Part II: Problem solving Part III: Synthesis Part IV: Appendices References are indicated by the Harvard method, where the author s surname and the book s publication year is listed as the following example: [Surname, year]. In the report references are both indicated before and after a full stop. If a reference is indicated before a full stop it refers only to the sentence, whereas if it stands after a full stop it refers to the entire section. I

6

7 Contents 1 Introduction 1 I Problem analysis 3 2 Neurophysiological background Movement control Voluntary movement control Movement impairment in stroke patients Stroke rehabilitation approaches Physiotherapy Constraint-induced movement therapy Robotic therapy Functional electrical stimulation Paired associative stimulation Motor imagery Rehabilitation with brain-computer interface systems Brain-computer interface system Inducing neural plasticity by using brain-computer interface Brain-computer interface based on movement-related cortical potentials Problem formulation 21 II Problem solving 23 6 Requirements System description General requirements specifications Individual requirements specifications Design Design of the graphical user interface Implementation Implementation of the graphical user interface Testing Tests of the developing system Test of the finished system Test summary III Synthesis Discussion 57 III

8 CONTENTS 11 Conclusion 59 IV Appendices 69 A Hebbian principle 71 B Transcranial magnetic stimulation 73 C Electroencephalography 75 D Movement-related cortical potentials 79 E Experimental protocol for performance testing 83 F Usability test 87 IV CONTENTS

9 Chapter 1 Introduction The World Health Organization (WHO) estimates that approximately 15 million people yearly worldwide suffer a stroke, and of these 5 million die and another 5 million are left permanently disabled. Stroke is the second most frequent type of cardiovascular diseases and as a result, stroke is one of the main single causes of severe disabilities in the developed world. [Mackay and Mensah, 2004] According to the Danish National Patient Registry (DNPR), Denmark had about 12,500 cases of hospitalization from stroke in 2009 and about 9,500 cases of hospitalization from traumatic brain injury and other forms of acquired brain injury. Based on statements in the DNPR and an expert assessment, it is estimated that 4,500 to 8,000 persons yearly acquire brain injury and are in need of rehabilitation. [Sundhedsstyrelsen, 2011] The cost associated with rehabilitation is very high in Denmark. A person who was brain damaged in 2004 costed a total of 181,000 DKK the first year and 300,000 DKK in total over the next five years. The cost for the first year was distributed on hospital costs equivalent to approximately 42 % and a loss of productivity and municipal costs of roughly equal size. In 2008, when the health care costs include several more parameters, the overall costs were 110,000 DKK, including the incident year and the following year. The total costs of brain injury treatment and rehabilitation over the first two years, calculated from the 2008-numbers, were 270,000 DKK. None of these numbers cover all costs, but indicates that there are high and persistent high costs of brain injury and subsequent rehabilitation. [Sundhedsstyrelsen, 2011] Stroke is caused by disturbance of the blood supply to the brain and is classified in two categories; ischemic stroke and hemorrhagic stroke. Ischemic stroke is the most common type of stroke and is caused by decreasing blood supply to part of the brain due to circulatory blockage. Hemorrhagic stroke is caused by leakage of blood caused by a ruptured blood vessel or an abnormal vascular structure. Strokes lasting only seconds can cause neurological symptoms, while strokes lasting minutes can cause irreversible neuronal damage. If the primary motor cortex (M1) is affected by a disruption of blood supply the consequences are varying degrees of paralysis of the represented muscles. The consequence of a stroke can be fatal for the patient. After surviving a stroke the person may still suffer from a series of disabilities, most likely problems related to motor control. Stroke also affects the quality of life and changes daily life drastically at the workplace, at home and in the community. [Guyton and Hall, 2006] Treatment for stroke survivors can be divided in the two stages, acute stroke treatment and post stroke rehabilitation. The aim of acute stroke treatment is to limit the neurological damage of the stroke and to ensure the survival of the patient. Post stroke rehabilitation address the restoration or improvement of lost body functions, which will provide a better quality of life and help the patient to return back to a normal life. [Guyton and Hall, 2006] Different rehabilitation strategies have been attempted for recovery of motor dysfunctions, but none of the current therapies provides significant improvements and thus remain a therapeutic problem [Birbaumer et al., 2008]. Physiotherapy is a classical stroke rehabilitation approach which aims on providing the patient a physical basis by performing certain movements under the supervision of physiotherapist [Guyton and Hall, 2006]. A more modern method is rehabilitation with robotic devices, which have been widely used over the past 15 years [O Dell et al., 2009]. The method is used to assist exercise and 1

10 quantify movements in stroke rehabilitation, however it takes a long time to get familiar with the robot and the training path is restricted [Fasoli et al., 2004]. Another technique of rehabilitation is functional electrical stimulation (FES) of muscles. FES provides short coordinated bursts of electricity to weak muscles to facilitate functional movement. The design of the stimulation circuit and electrode is a crucial factor in the application of FES in order to acquire correct conduction from the stimulation current to the target tissue. [Rushton, 1997] Stroke rehabilitation through motor imagery (MI) is a method, where the patient executes the movement mentally. MI provides an active movement control by the patient and may include more natural and voluntary motions. The approach is based on the fact that an activation of the motor system can be seen during both motor imagery and actual execution of a movement. [Parsons, 2001], [Frak et al., 2001] Even though the use of new technological strategies to improve motor recovery after stroke has exploded over the past two decades, there are still questions about the effectiveness versus the cost. The use of robotic devices and FES are undoubtedly technological revolutions, but they will remain passive methods when applied alone, since they are not able to integrate the patients own will easily. However, this can be done with motor MI. [O Dell et al., 2009] In the Center for Sensory-Motor Interaction (SMI) at Aalborg University a new rehabilitation strategy is proposed based on a brain-computer interface (BCI). By using appropriate brain signals, movement intentions of the stroke patient can be predicted in order to trigger the execution of the actual movement. This study aims on the development of a user friendly operating interface for a BCI system, whereby the signal processing of the applied brain signals is not in the focus of the report Introduction

11 PART I PROBLEM ANALYSIS 3

12 4

13 Chapter 2 Neurophysiological background Patients exposed to stroke usually have an impairment in the motor-, cognitive- and emotional system. After acute stroke about 80 % of the patients have some form of motor impairment. [de Vries and Mulder, 2007] In order to help these patients regaining the lost motor function, neural and somatic structures which are involved in a movement control has to be identified. The following chapter will describe the road map for a thought of a movement in the brain to the desired action of the limb. Due to its high complexity, the level of explanation is limited to the relevant parts. 2.1 Movement control Skeletal muscles can be controlled from many places in the central nervous system (CNS), e.g. the spinal cord, the pons, the basal ganglia, the cerebellum and the motor cortex. Each part has its own unique role. The spinal cord is mainly seen as a channel passing signals from the brain to the periphery of the body and the other way around. In reality the structure is more complex as the spinal cord consists of several cord centers, which are commanded by the upper levels of the nervous system. These neuronal circuits in the cord are among others responsible for walking movements or different reflexes. Pons, basal ganglia and cerebellum belong to the lower brain (subcortical level) and control automatic, instantaneous muscle responses to sensory stimuli. These muscle activities are also called subconscious activities and include for instance the control of arterial pressure and respiration. The motor cortex (figure 2.1) is part of the higher brain (cortical level), which is in charge of complex movements that are controlled by thought processes. This part of the brain also functions as a storage of information for future control of motor activities. [Guyton and Hall, 2006] Figure 2.1: Overview of the cortical areas of the brain [Despopoulos and Silbernagl, 2003]. Almost all voluntary movements a person performs begin in the cerebral cortex and are based on movements the person has done numerous times before. When a person wants to perform a certain task, the cortical motor area starts a pattern already stored in deeper layers of brain stem, basal ganglia, cerebellum or spinal cord and combines the movement with the information that comes 5

14 2.2 Voluntary movement control from the somatosensory cortex (figure 2.2). The constant feedback from the somatosensory cortex enables adjustment of the voluntary movement before and during the execution. [Guyton and Hall, 2006], [Despopoulos and Silbernagl, 2003] As an important control center for motor function the cerebellum has afferent and efferent connections to the periphery and the cortex. When a new unknown movement is needed (motor learning), the cerebellum (figure 2.1) is responsible for motor adaptation to the new movement sequence but also associated in the planning, execution, controlling and refining of movements. [Despopoulos and Silbernagl, 2003] Figure 2.2: Events from decision to move to execution of movement [Despopoulos and Silbernagl, 2003]. 2.2 Voluntary movement control A motor command, coming from the CNS, is delivered by the somatic nervous system (SNS) which controls the voluntary contraction of skeletal muscles. The somatic motor pathway which enables voluntary skeleton muscle involves at least two types of motor neurons; lower motor neurons and upper motor neurons. The latter synapses on the lower motor neuron, resulting in a innervation of a single motor unit of a skeletal muscle. This means a destruction or damage of the lower motor neurons would eliminate voluntary control of the corresponding motor unit. The connection between the upper and the lower motor neurons is supported by the corticospinal pathway. When a voluntary movement is initiated, neurons in the M1 (figure 2.1) send commands to upper and lower motor neurons. All neurons in the M1 are responsible for one specific somatic location respectively and arranged in a somatotopic representation (figure 2.3). The leg components are in the middle, the face components are situated laterally and the toes are represented deeply in the cortex. The largest representations belong to muscles which imply the finest movements. [Guyton and Hall, 2006], [Despopoulos and Silbernagl, 2003], [Martini, 2006] 6 2. Neurophysiological background

15 2.2 Voluntary movement control Figure 2.3: Representation of the different muscles of the body in the motor cortex [Guyton and Hall, 2006]. The M1 cannot initiate a movement alone, but needs to be stimulated by neurons from the premotor cortex and the supplementary motor area (SMA), which support and coordinate the M1. One task of the premotor cortex is to provide sensory guidance of movement while the SMA is, among others, responsible for planning and coordination of more complex movements, e.g. movements that require two hands. [Martini, 2006] The premotor cortex and the SMA are able to receive information from different decisional centers within the brain. These brain areas interpret the information and act according to it by coordinating the execution commands, transferred to the M1, which subsequently sends the signal to the effectors. [Martini, 2006] If for instance sensory information arrives at the CNS, it is routed according to location and nature of the stimulus. The receiving brain center, e.g. the M1 tells that an object is collected by the arm, which will initiate the movement pattern. Thus, the movement is the consequence of triggering the pattern instead of stimulating each neuron separately, which means the subject is able to perform the movement more easily when it is repeated. [Martini, 2006], [Despopoulos and Silbernagl, 2003] Overview of the nervous system When the desired motor command is issued by the CNS, it is distributed to the destined limb by the peripheral nervous system. The peripheral nervous system is divided into an efferent division and an afferent division, where the efferent division carries motor commands from the central nervous system to the effectors. The efferent system is further divided into the SNS and an autonomic nervous system (ANS). The SNS controls voluntary skeletal muscle contractions, where the ANS is responsible for automatic regulation of e.g. smooth muscles and cardiac muscles. The afferent division brings sensory feedback from the peripheral tissues back to the CNS. In this way a closed loop system is established. [Martini, 2006] Sensory feedback coming from the periphery, enters the spinal cord through the dorsal roots of the spinal nerves. The signals from the skin and locomotor system are then carried through one of 2. Neurophysiological background 7

16 2.2 Voluntary movement control two alternative sensory pathways up to the brain, the dorsal column-medial lemniscal system or the anterolateral system. Both signal tracts are partially combined again in the thalamus. Figure 2.4: Transmission of sensory information through the dorsal column-medial lemniscal system or the anterolateral system to the somatosensory cortex (S1) [Despopoulos and Silbernagl, 2003]. The dorsal column-medial lemniscal system (figure 2.4) consists of large fibers that allow signal transmission with high velocities. Furthermore it has a high degree of spatial orientation of nerve fibers in relation to their origin. These properties makes the dorsal system appropriate for carrying signals which have to be transmitted rapidly and with temporal and spatial accuracy, for instance position sensations from the joints. On the contrary, sensory information that does not have to be transmitted rapidly or with high accuracy is carried mainly by the anterolateral system. However, the anterolateral system has the possibility to transmit a wide range of sensory modalities, such as pain, thermal sensations and crude tactile sensations. Similar to the motor cortex (figure 2.3), the different parts of the body are represented in the somatosensory area. [Despopoulos and Silbernagl, 2003], [Guyton and Hall, 2006] Skeleton muscles When the motor neuron reaches the skeleton muscle, it splits into collaterals with terminal branches, which supply muscle fibers. When a skeleton muscle receives an action potential (AP), the muscle will contract. If one AP arrives immediately after another, they will accumulate. In this manner the first stimulus facilitates the response to the second stimulus, which increases the muscle strength. [Despopoulos and Silbernagl, 2003] 8 2. Neurophysiological background

17 2.3 Movement impairment in stroke patients 2.3 Movement impairment in stroke patients Patients exposed to strokes usually have an impaired motor, cognitive and emotional system, due to a lesion in the corresponding cortex. Whereas 80 % of the patients have some form for motor impairment, 20 % regain a part of their motor control, leaving 60 % with a chronic motor disorder [de Vries and Mulder, 2007]. The movements are still decided in the prefrontal cortex and send to the motor cortex and the basal ganglia, but the movements cannot take place, due to cell death caused by the stroke. [Martini, 2006]. 2. Neurophysiological background 9

18 2.3 Movement impairment in stroke patients Neurophysiological background

19 Chapter 3 Stroke rehabilitation approaches This chapter focuses on different methods of stroke rehabilitation and explains the underlying active principles. The methods reach from physiotherapy and constraint-induced movement therapy which work without the help of machines, over to the application of robotic therapy and electrical stimulation. Finally motor imagery related to imagination and observation of movements is explained. 3.1 Physiotherapy Physiotherapy in stroke rehabilitation aims on the establishment of a certain physical basis on which future functionality can be build upon with the goal of independence in the far future. Thereby, physiotherapy mainly focuses on the performance of controlled and balanced movements in order to enhance strength and stamina and increase mobility. [of Malaysia, 2011] Physiotherapy treatment starts directly after a stroke occurred. First, the physiotherapist will evaluate the degree of disabilities in order to develop an appropriate rehabilitation plan. During the course of treatment the physiotherapist will assess to which extend recovery of the lost limb control can be reached in comparison to the state before stroke. If permanent impairment is conceivable it is the task of the physiotherapist to teach the patient methods how to manage daily life without the affected limb. [Hendersen, 2011] 3.2 Constraint-induced movement therapy In constraint-induced movement therapy (CIMT) the stroke patient is forced to use the affected side of the body by immobilization of the unaffected side. CIMT was developed by the American psychology professor Dr. Edward Taub. He observed that stroke patients evolve habits where they do not use the impaired limb, especially in the early phases after the event. Instead they learn techniques to compensate the disability by an increase use of the intact limb, which leads to prohibition of functional recovery in the impaired limb. [Grotta et al., 2004] When applying CIMT today, e.g. on a patient whose left arm is impaired, the therapist constrains his right arm in a sling. In this manner, the patient is forced to use his affected arm repetitively and intensively. A requirement for treatment according to CIMT is the patient s existing ability of wrist extension and arm and finger movements. It is also proven that CIMT enhances brain activity. [Taub et al., 2000] 3.3 Robotic therapy Frequent therapeutic rehabilitation training for stroke patients have shown to be effective in relearning of motor skills [Kwakkel et al., 1997]. However, due to finite physiotherapy resources, the patient can not invest enough time in concentrating on rehabilitation activities. This leads to the approach of amending traditional physical therapy with robot assisted therapy [Jackson et al., 2007]. In general, a robotic device supplies the intended movement of a patient by mimicking the role of the physiotherapist. For instance, for rehabilitation of functions of the arm, the patient is grasping a joystick-like handle which is connected to the computer monitor in front of him (figure 3.1). 11

20 3.4 Functional electrical stimulation A suggestive task could be to carry out a particular movement, e.g. moving the handle towards a fixed or moving target displayed on the monitor. If the patient moves the handle in the wrong direction the robotic device will take over control in order to correct the direction. Furthermore the device will try to initialize the move if no movement happens at all by slightly nudging the arm of the patient [Trafton, 2010]. Figure 3.1: Robotic therapy; the patient tries to follow with the handle a target which is displayed on the screen [Trafton, 2010]. 3.4 Functional electrical stimulation Functional electrical stimulation (FES) is the electrical stimulation of nerve pathways to produce a controlled contraction of muscles with the purpose of imitating normal voluntary movements and restoring neuroconduction function [Hewitt, 2011], [Raymond, 2006]. Cortical plasticity is a fundamental factor which influences motor relearning when FES is used related to stroke rehabilitation [Jenkins and Merzenich, 1987], [Alamancos et al., 1992], [Piero et al., 1992], [Chae and Yu, 1999], [Ridding and Rothwell, 1999]. When learning a motor task the cortical excitability increases. Further development happens when the movement is repeated and grows significantly during skillful training tasks [Muellbacher et al., 2001], [Nudo, 2003], [Nudo, 2007], [Perez et al., 2004]. Peripheral electrical stimulation (PAS) plays a dominant role in motor relearning in order to improve disabled functions [McDonnell and Ridding, 2006], [Powell et al., 1990]. A simple FES system (figure 3.2) consists of a main unit, a stimulator, a control sensor and stimulation electrodes. The stimulation electrodes are placed on the target muscle. The sensor is used to control the electrical pulse from the main unit and detect the intensity feedback of the patient. Generally the electrical pulse ranges from 150 µs to 300 µs, while the stimulation amplitude ranges from 10 ma to 150 ma and the stimulation frequency between 20 Hz and 40 Hz [Raymond, 2006] Stroke rehabilitation approaches

21 3.5 Paired associative stimulation Figure 3.2: The components of a simple FES system [Peckham and Knutson, 2005]. Three types of training were designed in a finger flexion/extension experiment to test the best way for motor relearning; 1) FES alone, 2) voluntary muscle activity alone, 3) therapeutic FES (TFES) - the combination of voluntary movement and FES. It was concluded that TFES is better than the other two methods because more obvious changes were produced on motor evoked potential (MEP) area. Besides TFES had a great potential to restore neuroconduction function efficiently. [Barsi et al., 2008] 3.5 Paired associative stimulation Paired associative stimulation (PAS) is a protocol that uses the Hebbian principle (appendix A) by stimulating the median nerve with a slow-rate repetitive low frequency followed by a transcranial magnetic stimulation (TMS) (appendix B) on the contralateral motor cortex. Due to the theory of Hebbian, the synaptic efficacy induced by associative activity is dependent on the order of activity from the pre- and postsynaptic neuron. When the postsynaptic cell is depolarized synchronously with the afferent input from another cell this will result in a Hebbian long-term potentiation. In contrast if the postsynaptic neuron spiked before excitatory postsynaptic potential was induced by afferent stimulus, it will result in a long-term depression which is also called asymmetric Hebbian rule. TMS is a noninvasively technique to stimulate peripheral nerves and brain tissue by using the principle of electromagnetic induction. When timed appropriately, PAS induces a persistent increase in corticospinal excitability which could be demonstrated as an indicator of plasticity within M1 [Cirillo et al., 2009], [Classen et al., 2004]. It is possible to induce an increase in amplitude of the MEP when TMS is transferred at the intensity below the demand of evoking a response in a resting target muscle by using PAS. The increase is conditionally dependent on the timing between the afferent and the magnetic stimulation which needs to be synchronous at the level of the motor cortex. A study showed that the plasticity changed rapidly within 30 minutes by inducing PAS and had a duration of minimum 60 minutes [Stefan et al., 2000], [Jayaram et al., 2007]. Studies have shown that stimulation of peripheral nerves, peripheral nerve and motor cortex or muscle motor point and motor cortex is capable of inducing reorganizations within the human motor cortex in association with skill acquisition [Pascual-Leone et al., 1994], [Stefan et al., 2000], [McKay et al., 2002], [Ridding and Uy, 2003]. The dominate theorem is that the organizational changes induced by modifying afferent input are 3. Stroke rehabilitation approaches 13

22 3.6 Motor imagery similar in nature to those seen during motor learning. This is already used in order to induce therapeutically relevant plasticity [Conforto et al., 2002], [Fraser et al., 2002]. In relation to stroke patient PAS can induces a modulation in the corresponding non-paretic limb and reduce the asymmetry between the the healthy and the impaired limb. This means corticomotor excitability decreases in non-lesioned motor system but increases in lesioned motor system between hemispheres with inhibitory PAS in imperative or voluntary movement and in rest [Stinear and Hornby, 2005], [Rogers et al., 2010]. 3.6 Motor imagery Improvements of stroke rehabilitation is among others dependent on the amount of sensory feedback given by motor activity [Kwakkel et al., 2004]. There are five commonly known sources of sensory information [de Vries and Mulder, 2007]: Tactile information Vestibular information Proprioceptive information Visual information Auditory information Recent results suggest that imagination and observation of movements could produce a further source of information that could be appropriate in stroke rehabilitation. A study by Feltz and Landers stated that subjects who where mentally practicing a particular task showed better results in the performance than subjects who did not train at all. However less improvements were seen in comparison to physical practice. [Feltz and Landers, 1983] The term MI can be defined as the mental process of imagining a certain body movement without moving it in reality [Jeannerod, 2001], [Mulder et al., 2005]. Two different perspectives of MI can be distinguished. In the first person perspective, the person thinks about the feeling of the movement itself. In comparison, the third person perspective describes the situation when the person is imagining another person doing the movement. The latter is more a visual imagination than feeling. To enhance relearning of a new skill, the first person perspective is appropriate while involving in learing coordinate and timing of motor skill and the third person perspective could play an important role in motor tasks focusing on the form of movement.[de Vries and Mulder, 2007] Evidence for the use in rehabilitation In the field of sports sciences it was shown that learning of a movement is supported by the repeated first person perspective imagery [Feltz and Landers, 1983], [Driskell et al., 1994]. An important condition for the application of MI is that it is only appropriate if a representation of the movement exist, meaning that the subject must somehow have the ability to act. However, there are only few studies dealing with MI in terms of rehabilitation of stroke patient. The first arbitrary controlled study was made by Page et al. with early stroke patient (2 to 11 months post-stroke). It could be proven that the patient who besides physical therapy also received motor imagery training showed a significant higher improvement in comparison to the control group. The control group got some general information about stroke instead. [Page et al., 2001] In a second study Page et al. found the same result even for chronic stroke patient (one year post-stroke)[page et al., 2005] Stroke rehabilitation approaches

23 3.6 Motor imagery Evidence for similarities in cortical activation The approach of stroke rehabilitation through motor imagery is based on substantial similarities between motor imagery and the actual execution of a movement. Studies by Parsons and Frak et al. showed that the duration for mentally executing a movement is quite similar to the time which is needed to carry out the actual movement [Parsons, 2001], [Frak et al., 2001]. This observation is called mental isochrony. A further evidence for similarity is based on studies using brain-mapping techniques [Jackson et al., 2001], [Jeannerod, 2001]. It was found that the same brain areas are activated during imagination and the real action. In particular these are the parts of the neural system which are associated with preparing and commanding of movements: the premotor cortex, the dorsolateral prefrontal cortex, the inferior frontal cortex, the posterior parietal cortex, the cerebellum and the basal ganglia. The activation of the M1 during imaginary movements is still unclear because some studies found neural activation and others not. This is however, a crucial question due to the conclusion of several studies that neural reorganization concerning motor re-learning happens in the M1. Especially the latter mentioned correlation is an important observation as it states that the point in time can be identified when the impaired subject wants to execute a movement. [Hlustik and Mayer, 2006] Mirror neurons A further considerable indicator for possible offline activation of the motor system was the discovery of mirror neurons, which become active, e.g. when a person is observing another person performing a task [Rizzolatti, 2005]. Neurones related to goal-directed motor control were first found by Rizzolatti [Rizzolatti et al., 1987] in Marque Monkeys. Later [Fadiga et al., 1995] concluded that these kind of neurones must somehow also exist in the human motor cortex. During their experiment they could see an increase in MEP when the subject observed another person grasping a object. Furthermore, the shape of the recorded MEP reflect the shape of MEP, which was recorded when the subject carried out the movement itself. Due to these findings researchers consider mirror neurons to play potentially a role in the re-learning of motor control [Gallese, 2005]. However, the function of mirror neurons is not completely understood yet, which may be one reason why there are no systematic studies in this subject area of neurological rehabilitation [de Vries and Mulder, 2007]. 3. Stroke rehabilitation approaches 15

24 3.6 Motor imagery Stroke rehabilitation approaches

25 Chapter 4 Rehabilitation with brain-computer interface systems Brain-computer interface technologies are frequently mentioned in relation with therapeutic purposes. The use of brain-computer interface for inducing neural plasticity in order to rehabilitate stroke patient has gained more attention. The opportunity of using this method will be explained in this chapter. 4.1 Brain-computer interface system A brain-computer interface (BCI) provides a connection between humans and external devices or applications by using neurophysiological signals obtained from the brain. There are two different types of BCI systems, invasive- and non-invasive. Invasive BCI is directly implanted in the brain tissue by using electrodes, while non-invasive BCI makes use of electrophysiological recordings. Electroencephalography (EEG) (appendix C) is commonly used as a recording technique for noninvasive BCI systems. [Birbaumer and Cohen, 2007] BCI systems let the user convert thoughts into actions, which do not involve voluntary muscle movement. The system offers new means of communication for people with paralysis or severe neuromuscular disorders and can be useful for patients whose natural output pathways of peripheral nerves and muscles are out of function e.g. caused by stroke. [Dornhege et al., 2007] There are two separate approaches for control of BCI: The human learns to adjust a certain brain activity by getting a neurophysiological feedback. During training sessions the subject improves the ability to activate voluntarily different areas in the brain, which can be utilized later as control commands. The machine learns to interpret the brain activity from the human using a learning algorithm. This will require adjustment (calibration) of the algorithm by asking the individual user to generate a specific brain state several times. Based on these examples the machine derives later the statistical structure of specific brain activity. In fact, in practice it is not possible to see the mentioned interactive learning algorithm completely autonomous. There will always be a mutual influence between the learning user and algorithm whereby it is still an open question how they can be combined ideally. [Dornhege et al., 2007] 4.2 Inducing neural plasticity by using brain-computer interface Research in BCI has been mainly concentrated on providing alternative communication devices for the past decade. This has changed in the recent years due to a increasing interest in extending the application range of BCI technology. One of the most interesting new approaches is the use of BCI technology to induce neural plasticity, and thereby restore damaged brain functions. [Grosse- Wentrup et al., 2011] 17

26 4.3 Brain-computer interface based on movement-related cortical potentials Studies on attention deficit hyperactivity disorder (ADHD) and epilepsy has provided proof-ofprinciples for a positive impact of BCI technology in inducing cortical reorganization, and stroke patient have been shown capable of operating a BCI based on magnetoencephalography (MEG) and electrocorticography (ECoG). [Grosse-Wentrup et al., 2011] BCI can complement the existing rehabilitation methods by detecting specified actions in the brain signal, and thereby provide feedback to the patient. This can induce cortical reorganization due to neural plasticity and cause functional recovery. A well-known approach of this principle is based on the theory of Hebbian (appendix A). Assuming the case of patient which is suffering from a sub-cortical stroke, the connection between peripheral muscles and the sensorimotor cortex has been corrupted. If it would be possible to activate sensory feedback loops and M1 synchronously, formerly passive cortical connections can be strengthened by Hebbian plasticity. For this particular case the task of the BCI system would be to detect M1 activation e.g. for an intended movement. When M1 activation is detected the BCI provides coordinated sensory stimulation. The sensory stimulation can be optimized to provide feedback on those neural states that are optimal for inducing neural plasticity. [Grosse-Wentrup et al., 2011] 4.3 Brain-computer interface based on movement-related cortical potentials BCI technology might increase the efficacy of a rehabilitation protocol and thus improve the muscle control for stroke patients and others suffering from lost motor control. This can be implemented by detecting the planning phase of a movement in the brain activity and supplement the patient s impaired muscle control, e.g. send trigger (figure 4.1). The planning phase of a movement can already be measured as part of MRCP (appendix D) which are correlated to voluntary muscle contraction and changes in the cortical activity over the area of the motor cortex. [Daly and Wolpaw, 2008] Figure 4.1: BCI system which is able to detect MRCP online from asynchronous (spontaneous) imaginary movements. The intended movement will be triggered by an electrical stimulation Rehabilitation with brain-computer interface systems

27 4.3 Brain-computer interface based on movement-related cortical potentials MRCP represent averaged EEG activity in the motor cortex off the brain related to a voluntary movement [Yom-Tov and Inbar, 2003]. The signal is produced by firing of the neurons within the brain, here especially in the motor cortex, which produces an electrical activity that can be recorded as EEG from the scalp. It is not possible to detect the electrical potential of one single firing neuron due to the low amplitude of the discharge of the neuron. Therefore the EEG that gets picked up on the scalp is a summation of thousands synchronous firing neurons. [Misulis and Head, 2003] MRCP may be a suitable approach for rehabilitation of stroke patients, because its morphology has been investigated under several different conditions. The phases of MRCP consist of an initial negative phase (INP), also known as the early and late Bereitschaftspotential, and a movementmonitoring potential. The INP is considered to represent the planning phase of a movement and is believed to occur even if the movement is imagined. do Nascimento investigated the influence of the rate of torque development and torque amplitude on MRCP during imaginary and real isometric plantar flexions in two different studies [do Nascimento et al., 2005] [do Nascimento et al., 2006]. In both cases it was concluded that information about these two factors are encoded in the MRCP. The comparison of MRCP from real movements and imagined movements showed that the latter was mainly characterized by a reduced amplitude. Furthermore, although the known movement related parameters can be detected before the movement onset, the shape of the MRCP during imagination looked different. This applies especially to the part after the movements onset which is probably due to the disparity in efferent feedback. Due to the similarities between imaginary and real movements a BCI system based on MRCP might be useful in order to detect intended movements [Cunnington et al., 1996]. 4. Rehabilitation with brain-computer interface systems 19

28 4.3 Brain-computer interface based on movement-related cortical potentials Rehabilitation with brain-computer interface systems

29 Chapter 5 Problem formulation An increasing number of individuals suffer from motor impairment caused by stroke. While different rehabilitation strategies have been attempted for recovery of motor functions none of them have provided significant improvements. [Birbaumer et al., 2008] MI has been found to activate the same cortical areas as they are activated in real movements except for the somatomotor cortex. This could be interpreted in this manner that both motor execution and MI have a similar movement preparatory phase, whereby the latter is not followed by execution commands from the motor neurons in the M1. MRCP opens important perspectives in the development of new rehabilitation technology, because they allow to distinguish movement intentions, which can be used to control an output variable in BCI. The current BCI system from the Center for Sensory-Motor Interaction (SMI) at Aalborg University is able to determine asynchronous movements onsets offline. A true positive rate (TPR) for the detection of INP was 63.3 % for healthy subjects during imaginary movements (n = 7; 63.3 % ± 9.8 %) [Niazi et al.]. When combining the BCI system with peripheral electrical stimulation the system has to be able to detect asynchronous movements online. Thereby one of the biggest challenges is the fast and accurate detection of INP, which appears prior the actual movement onset. In order to do so, this will require a detection algorithm with minimal processing time to reduce feedback delay, which is the time difference between the measurement of neural state and the response of the system. Further for the detection algorithm advanced signal processing techniques is required in order to obtain a maximal TPR of detection. The acquisition of an accurate feedback is important to induce neural plasticity. The latency between detection of INP until the triggering of the intended movement by electrical stimulation has to match the time that it takes from the onset of an INP until the actual movement is carried out by a healthy subject. Neural plasticity based on the Hebbian principle can be induced with a maximum latency of about 200 to 300 ms [Grosse-Wentrup et al., 2011]. The detection of a motor preparation phase such as INP will allow to predict the intention to move in order to establish such a causal association. [Niazi et al.] A solution of these challenges, implemented in a rehabilitation program for stroke patients, will give the patient a feeling of free will in aspect of movement control, because of the immediate electrical stimulation after the intention of a movement. This approach might be an improved rehabilitation method compared to established rehabilitation therapies. As a result of the problem analysis the following problem can be stated: How can a BCI system, based on online detection of INP, be designed and implemented so that it can be used for rehabilitation purposes in order to help stroke patients regaining their basic motor functions? 21

30 22 5. Problem formulation

31 PART II PROBLEM SOLVING 23

32 24

33 Chapter 6 Requirements Based on the problem formulation this chapter contains an overall description of the system with the corresponding functional and non-functional requirements. The specific functional requirements for the login screen, training mode, detection algorithm and testing mode are described individually. 6.1 System description The system under development is the initial approach of making BCI work in a rehabilitation setting for stroke patients in order to help them regaining their basic motor functions. The system needs to be an online BCI which operate by trying to build the best possible detector in order to identify INP from the subject s EEG. Because each subject is unique, the detector has to be build based on individual recorded training data. When the detector has been found for the subject, it would be used in a future rehabilitation program to give electrical stimulation to the impaired extremity. it Start system (Login) Training mode Detection algorithm Testing mode Trigger (stimulation) Figure 6.1: The BCI system can be mainly be divided into training and testing mode. A detector is build from recorded training sessions. In the testing mode the detector is used to detect INP which can trigger an electrical stimulation. The BCI system can be divided into two parts; a training mode and a testing mode (figure 6.1). In the training mode brain activity will be recorded from multiple EEG channels together with EOG and EEG from the tibialis anterior muscle. The parameters that has to be set before recording can either be synchronous or asynchronous and real or imaginary movements. The stored recordings will be used for building a detector which can detect INP online. In order to optimize and adapt the detector, different training sessions from the database can be selected, which do not necessarily have to origin from the same subject. In the testing mode the detector is tested on imaginary synchronous or asynchronous movements. Every time an INP is detected, the intended movement of the subject will be triggered by an electrical stimulation. The user is able to see the detection performance online through feedback from the subject coded in finger movements. 25

34 6.2 General requirements specifications 6.2 General requirements specifications The requirements are divided into functional and non-functional requirements. The functional requirements describes directly what the system should do, while the non-functional requirements describes the limitations to be respected by the system Functional requirements The system will record nine channel EEG measured from the scalp. The system will record one channel EOG measured from the eye. The system will record three channel EMG (one channel from the tibialis anterior muscle and two channels from the fingers). The system will store all recordings locally. The system will display EEG and EMG signals online. The user will have the possibility to display one of nine recorded EEG channels at a time. The system will give visual feedback to the patient about movement performance. The system will send a trigger when an INP is detected Non-functional requirements The system must be programmed in MATLAB. The system must be easy to use and the graphical interface must have a user friendly design. The system will use hardware components, e.g. DAQ card, in order to acquire signals. 6.3 Individual requirements specifications The system consist of a login window, a training and a testing part with a menu for building the detector. Each part has its own requirements which have to be compiled in order to run the system Login window The user have to enter the name of the subject in order to gain access to the training and testing mode. The user will have to enter a subject name before starting a session. The system will check if the entered subject name exists in the database. The system will continue to add new training and testing sessions to the existing subject if the name exists Training mode In the training mode the user can choose to record sessions based on four different movement parameters. The system will have following parameters; synchronous or asynchronous and imaginary or real. The system will show two windows containing EEG and EMG signals online Requirements

35 6.3 Individual requirements specifications The system will provide visual feedback to the subject by showing a subject window on an external screen containing the force of EMG movement and other relevant information specified by the user. The system will save the selected parameters and recorded data Detector menu The recorded sessions in the training mode will be used to build a detector that is able to detect INP online in testing mode. The system will build a detection algorithm based on specific training sessions, which are selected by the user. The algorithm will be able to detect INP online in testing mode. The accuracy will be based on a threshold of TP and FP selected by the user. The system will store the detection algorithm until a new is build Testing mode The testing mode will use the detection algorithm for detecting INP online. The system will have following parameters; synchronous or asynchronous. The system will show two windows containing EEG and EMG signals online. The user will have the possibility to turn on/off the electrical trigger. The performance of the used detector will be displayed during the testing mode. The system will provide visual feedback to the subject by showing a subject window on an external screen containing the performance of the test and other relevant information specified by the user. The system will save the selected parameters and recorded data. 6. Requirements 27

36 6.3 Individual requirements specifications Requirements

37 i Chapter 7 Design This chapter explains the design of the system based on activity diagrams, which is also known as object-oriented flowcharts. The purpose of an activity diagram is to give a graphical representation of where and at which time the stepwise actions and activities in a program are taking place. This is visualized in a flowchart. 7.1 Design of the graphical user interface The system is divided into four main components; login window, training mode, detector menu and testing mode Login window User System Start program [Name is not entered] Type name Error message Confirm name Validate name Choose mode Enable training mode and testing mode [Name entered or already exist] [Testing mode is choosen] [Training mode is choosen] Open training mode Open testing mode Figure 7.1: Activity diagram for the login window. When the program starts, the user first has to login the name of the subject (figure 7.1). The input will be confirmed by the user and subsequently checked by the system. If the entered name is valid, the system will allow the user to select between training mode and testing mode. Otherwise, if the input is invalid the system will return an error message prompting the user to repeat the process. 29

38 u 7.1 Design of the graphical user interface Training mode User System Check subject name in database [Build detector is choosen] [Preparing training session is choosen] Build detector Choose training pattern Set subject window Press start Save settings, initialize signal acquisition and open subject window Start recording Recording Stop recording Validate session [Discard session is choosen] [Save session is choosen] Save recorded session Discard recorded session Figure 7.2: Activity diagram for the training mode. The training mode (figure 7.2) will start, when the user has selected it after login of the subject. Before starting a training session, the user has to choose between four different movement parameters; real or imaginary movements and synchronous or asynchronous movements. Furthermore, the user can decide which information should be presented to the subject as visual feedback. These elements are; a bar which reflects the EMG signal, duration of the current recording and a countdown timer which is used for synchronous movements. After all settings are made the user starts the data acquisition. The incoming signals from EMG and EEG channels are visualized to the user. If the user decides that the signals are up to the standard a recording can be started. When the user stops the recording the system will prompt to decide whether the current session should be saved in the database. If the user does not want to save the data, the current session will Design

39 7.1 Design of the graphical user interface be discarded. Otherwise the data will be saved. At this point one training session is finished and the user has the possibility either to build the detector or to start a new training session. At all times when the user has the control, it is possible to finish the training mode and switch back to login. Therefore this activity is neglected in the activity diagram Detector menu m User System Select subject Select sessions and enter used gain Start building detector Extract MRCP- template Enter threshold for EOG activity Building ROC- curve Enter TP and FP Finishing detector Figure 7.3: Activity diagram for the detector menu. The user can build the detection algorithm (figure 7.3) either after finishing training mode or before starting testing mode. At the beginning the user has to select the training sessions to be included for the composition of the detection algorithm. When the training sessions are selected the user has to state the gain of the EEG signals used for the recording. Then the preprocessing of building the detection algorithm can be started. In the next step the user has to state the threshold for the eye electrode (EOG channel) based on a graph which shows the energy at the eye electrode in each epoch of the selected sessions. If the energy is too high it will disturb the MRCP signal in this epoch and thereby also influence the detection algorithm. By setting a threshold these epochs will be excluded. Finally the user confirms the threshold and a receiver operating characteristic (ROC) curve is obtained. Based on this curve the user states a number for accepted true positive and false positive 7. Design 31

40 ] 7.1 Design of the graphical user interface values. After confirming the input again the system pops up with a message, that the build of the detector is completed Testing mode User System Choose training pattern Set subject window Set trigger for stimulation Press start Save settings, initialize signal acquisition and open subject window Start recording Recording (send trigger for stimulation) Stop recording Validate session [Discard session is choosen] [Save session is choosen] Save recorded session Discard recorded session Figure 7.4: Activity diagram for the testing mode. The testing mode (figure 7.4) will start when the user has build the detector. Before starting a testing session, the user has to choose between two different movement parameters; synchronous or asynchronous imaginary movements. Furthermore, the user can decide which information should be presented to the subject as visual feedback. These elements are; a bar which reflects the EEG signal from the selected channel, duration of the current recording and a countdown timer which is used for synchronous movements. Unlike the training mode, the testing mode will use the detector for detecting the INP of MRCP online. Whenever the INP is detected and the energy on the EOG channel is below the threshold, Design

41 7.1 Design of the graphical user interface the intended movement of the subject will be supported by an electrical stimulation with low intensity. This stimulation is possible to toggle on and off. After all settings are made the user starts the data acquisition. The incoming signals from EMG and EEG channels are visualized to the user. If the user decides that the signals are up to the standard a recording can be started. When the user stops the recording the system will prompt to decide whether the current session should be saved in the database. If the user does not want to save the data, the current session will be discarded. Otherwise the data will be saved. At this point one testing session is finished and the user has the possibility either to build the detector or to start a new testing session. At all times when the user has the control, it is possible to finish the testing mode and switch back to login. Therefore this activity is neglected in the activity diagram. 7. Design 33

42 7.1 Design of the graphical user interface Design

43 Chapter 8 Implementation This chapter contains an in-depth explanation on how the system is developed and the functions which the system is build up upon. The functions will be explained with respect to the corresponding GUI. 8.1 Implementation of the graphical user interface The system is divided into four main components; login window, training mode, detector menu and testing mode Login window Figure 8.1: Screenshot for the login window Features: Enter subject name Verification of subject name Select training/ testing Quit the program After the user has started the program a login screen will be presented prompting the user to enter the name of the subject in a text box (figure 8.1). The buttons for training and testing appear disabled at the beginning. After pressing the OK button the system will check the input. In case the entered name is valid the system will allow the user to select between training or testing. If the OK button is pressed without entering a name, an error message will pop up. 35

44 8.1 Implementation of the graphical user interface Training mode Figure 8.2: Screenshot for the training mode Features: Display of subject name and training session number Display of EMG and EEG signal Display of recording time Set training patterns (movement type) Set visibility of elements in subject window Control of recordings After the user has selected training from the login screen a new window will open representing the training mode (figure 8.2). The subject name is displayed together with the session number in the upper left corner of the window. If the entered name does already exist in the database, the system will automatically continue counting the training sessions according to the existing number of previous saved sessions. In the upper right corner of the window the user can make the settings for the upcoming recording. The elements displayed in the subject window (figure 8.3) can be activated/ deactivated simply by pressing check buttons. The training patterns and the displayed EEG channel can be selected by dropdown menus. EMG and EEG are displayed in the center of this window Implementation

45 8.1 Implementation of the graphical user interface Figure 8.3: Screenshot for the subject window By pressing the Start/Stop button data acquisition from the DAQ card is controlled. After Start is pressed the system assigns space for incoming data which is then continuously acquired from the input channels. The input is stored temporarily in a stack structure. Besides, the subject window pops up at the second screen. By pressing the Record button the acquired EEG and EMG signals are recorded and saved in a fixed path. Moreover the subject window is continuously updated with the EMG signal and the timer starts to count up in both the training and the subject window. During the recording the user is able to change the displayed EEG channel. When the recording is stopped the subject windows closes and the user is asked by the program if the current session should be saved. The session can either be discarded or saved in a separate folder for training sessions. The latter case entails to increment the training session number by one. The Detector button directs the user to the detector menu. When pressing Finish training the program goes back to the login screen. In the subject window (figure 8.3), the visual feedback reflects the force of EMG movement in the training session, but in the testing session, it presents the output of matched filter Detection algorithm Figure 8.4: Screenshot for the detector menu 8. Implementation 37

46 8.1 Implementation of the graphical user interface Features: Specification of subject selection Listing of training sessions Set amplifier gain Display MRCP templates and EOG activity Set threshold for EOG activity In the detector menu the user selects the training sessions to be included for the composition of the detection algorithm (figure 8.4). In order to simplify the search the user first can make a preselection, deciding whether to choose sessions only from the logged in subject or from all subjects. After training sessions are selected from the listbox the user has to enter the gain used for the recorded signals. The first graph is to calculate the averaged signal from channel Cz, the second one is the template which is illustrated detailed in the following paragraphs and the third one is the power of EOG in each EEG 3 s epoch. The process of finding the best possible detection algorithm can be divided into two steps. In the first step a spatial filter is applied to the training data to obtain a surrogate channel from the EEG channels recorded. From the surrogate channel MRCP template is extracted. In the second step this template is used to detect movement onsets the training data by a matched filter algorithm. The first step is started by pressing Build detector. The signal processing behind is described below. EEG signals are loaded, bandpass filtered from 0.05 Hz to 10 Hz and downsampled from 500 Hz sampling frequency to 20 Hz. Then a large Laplacian spatial filter (LLSF) is applied in order to obtain a surrogate channel (linear combination of the nine channels) from the EEG channels because LLSF provides a global linear filter resulting in a more concentration on the signal which contributes the most [Mellinger et al., 2007]. The channel coefficients are calculated as the following: 1, i = 1 x k = 1 (N ch 1), i = 1 (8.1) where N ch is the number of channels. The Cz-channel corresponds to the highest weight one and the other channels have a weight of 1/8, so that the sum of all channels is equal to zero. This approach blocks direct current interference effectively. The surrogate channel now contains the recorded data. In the next phase the 3 s epoch of signals are abstracted and divided into 2 s before and 1 s after each event which is defined as the indication of movement occurence implying the power of EMG is above the 10 % of the maximal value. In case of synchronous imaginary movements the epochs corresponds to the periodic cue and for asynchronous imaginary movements the subject is asked to give an indication by pressing a button 2 s after the event. For each 3 s epoch, in order to increase the SNR of the EEG signals, an optimized version of the LLSF called optimized spatial filter (OSF) is applied. OSF uses the coefficients obtained from LLSF and aims on finding the optimal combination of EEG channels that maximizes MRCP energy while minimizing the noise energy. Since the used filter technique is a data driven approach, Implementation

47 8.1 Implementation of the graphical user interface EEG channels that contain the more signal information have to have a higher weight than others. By getting an optimal set of x, while the sum of the coefficients is still zero, SNR will be maximized. The average algorithm is applied for all the epochs to get the 2 s long template. Eye blinking during a session produces large artifacts in the EEG channels which influences the MRCP template. For that reason electrooculography (EOG) activity is calculated for each epoch. The results are plotted as power of EOG in the third graph of the detector menu. According to the literature an epoch with EOG activity above 125 µv should be rejected. However, in the detector menu the user can choose the threshold manually according to the individual requirements. By pressing Cross validation the second step of finding the optimal detector is started. The extracted template is used to detect movement onsets in the training data. A detection decision underlies the likelihood ratio method and is computed (2 s sliding window with 200 ms shift) between the surrogate channel based on the training data and the obtained MRCP template. From this cross validation the user will be presented to a ROC curve in a separate window. The ROC curve represents a graphical plot of the sensitivity, given by the truth positive rate vs. false positive rate. The user picks a percentage value, by aiming for the best relation between a high true positive value and low false positive value, and input these values. Normally, the middle point of turning phase in the ROC curve is adoptive Testing mode Figure 8.5: Testing mode Features: Display of subject name and testing session number Display of EMG and EEG signal Display of recording time Set visibility of elements in subject window Set testing patterns (movement type) Set electrical stimulation on/off Control of recordings 8. Implementation 39

48 8.1 Implementation of the graphical user interface In order to attain the testing mode the user has to build a detection algorithm beforehand. Subsequently the window, called testing mode will open (figure 8.5). It can be seen that the window is very similar to training mode(figure 8.2). The differences or enhancements are explained in the following. In the upper right side of the window the user can decide if the recorded movements will be synchronous or asynchronous. Furthermore the electrical stimulation can be activated/ deactivated simply by pressing radio buttons. By pressing the Start/Stop button data acquisition from the DAQ card is controlled. After Start is pressed the system assigns space for incoming data which is then continuously acquired from the input channels. The input is stored temporarily in a stack structure. Besides, the subject window pops up at the second screen. By pressing the Record button the acquired EEG and EMG signals are recorded and saved in a fixed path. Moreover, the subject window is continuously updated by the EEG signal from the channel and the timer starts to count up in both, the testing and the subject window. During the recording the user is able to change the displayed EEG channel. A detection decision underlies the likelihood ratio method and is computed (2 s sliding window with 200 ms shift) between the surrogate channel from the recorded testing data and the MRCP template. A movement is detected online when two consecutive windows cross the threshold obtained from the ROC curve. If, in addition, the power of EOG activity lies in the allowed range of µv a command for electrical stimulation is send. When the recording is stopped the subject windows closes and the user is asked by the program if the current session should be saved. The session can either be discarded or saved in a separate folder for testing sessions. The latter case initiate an increment for the testing session number by one. The Detector button directs the user to the detector menu. When pressing Finish testing the program goes back to the login window. The performance of the current detector is displayed by a truth positive rate (TPR), a false positive rate (FPR) and a false negative rate (FNR). The performance is based on a feedback given by finger movements from the subject. The three different cases are detected in the following way: Case 1: Truth Positive The subject is instructed to extend its right index finger when a peripheral electrical stimulation is given right after a movement is executed (stroke patient) or imagined (healthy subject). This will mean that the detector has detected the INP in the EEG virtual channel. Case 2: False Positive The subject gets an peripheral electrical stimulation although no movement is executed (stroke patient) or imagined (healthy subject). This will mean that the detector has mistakenly detected a INP in the EEG surrogate channel which did not occur. Case 3: False Negative The subject is instructed to extend its left index finger when a peripheral electrical stimulation is not given right after a movement is executed (stroke patient) or imagined (healthy subject). This will mean that the detector has not detected the INP in the virtual channel. The finger movements are detected by measuring EMG on first dorsal interosseus muscle from index finger of the left and the right hand. Two to three seconds after starting data acquisition Implementation

49 8.1 Implementation of the graphical user interface the user has to press Calibrate. This will prompt the calculation of the threshold used for finger movement onset. During the recording the EMG signals are bandpass filtered from 30 Hz to 250 Hz, rectified and downsampled to 20 Hz. The occurrence of a finger movement is defined as the exceedance of the threshold by five consecutive samples. 8. Implementation 41

50 8.1 Implementation of the graphical user interface Implementation

51 Chapter 9 Testing Based on the developed software this chapter contains the tests that are performed during the design and implementation of the system and the subsequent tests with the finished system. The chapter is divided in two parts, where the first part will concentrate on the testing that was done during the development of the program, while the last part will focus on the testing of the finished program. 9.1 Tests of the developing system Continuous testing of the system is a natural way of developing a program. In this manner it is possible to locate and mend errors to make the program run correct. The test of the system was performed according to Unified Process, which contains following tests [Arlow and Neustadt, 2005]: Unit test Regression test Integration test System test A usability test was also performed on a low fidelity-prototype (LoFi) during the development of the system. The mentioned tests are subsequently explained Unit test The unit test tests a single function or a group of functions and is used in order to verify if the function is working correctly. This is done continuously throughout the system development. Functions usually contain several internal variables, and the execution of the unit test is to test if the values of these internal variables are correct. This is done by writing the value of the variable in question to the console window. Another way is to debug. The debug mode makes it possible to put breakpoints in the code execution and read the different variables that are processed step by step. This test is done continuously throughout the system development and will therefore not be further documented in the report Regression test The regression test is a method, which is used to handle errors that appear when functions are changed and inflicts errors on other functions in the system. This test is an integrated part of the iterative process of system development and will therefore not be further documented in the report Integration test The integration test is done by testing larger parts of the system together. It is investigated if the different components can work together as it was intended. The integration test is usually carried out as a black box test, where the focus is on the input and output of the system, rather than the execution of the internal commands. 43

52 9.1 Tests of the developing system Aim The aim of the integration test is to find errors in the functionality and thereby investigating if the input gives the expected result. Method A systematic test of all possible scenarios will be executed. This includes a test of the main flow of the system and possible alternative flows. An integration test will accentuate problems that can occur by regular use of the system. In the scenarios triggering and stimulation a healthy subject is used. As with the other test types used in the development phase, this test will be performed continuously while developing the system. Thus all software errors are corrected. The scenarios are following the flow of the system with the following order: Login Entering training/testing mode Training/testing mode: Data acquisition Training/testing mode: Data saving Detector menu: INP template extraction Detector menu: Set thresholds and build detector Testing mode: Stimulation Testing mode: Subject feedback Login Aim: It should be tested if it is possible to login a subject and subsequently activate the Training and the Testing button. Method: The user should type the name of the subject in the text field of the login screen and then presses OK Input: a) A name is entered in the text field and the user subsequently presses OK. b) A name is not entered and the user presses OK. Expected output: a) Training and Testing button are activated. b) An error message will appear, warning the user that a subject name is needed. Result: a) The subject is logged in. b) A warning message appears when no name is entered. Table 9.1: Result from the integration test of the scenario Login Testing

53 9.1 Tests of the developing system Entering training/testing mode. Aim: It should be tested if it is possible to achieve the name from the login screen and get the current training/testing session number while possible previous recorded sessions are considered. Method: After the user has entered the subject name in the login screen, the button Training is pressed. In the training/ testing window it has to be controlled if the entered subject name is displayed and if the session number is correct. Input: a) An already existing subject name is entered in the text field and the user presses OK and subsequently the Training button. b) A new subject name is entered in the text field and the user presses OK and subsequently the Training button. Expected output: a1) The correct name and session number are displayed. a2) An incorrect name and session number are displayed. b1) The correct name and session number are displayed. b2) An incorrect name and session number are displayed. Result: The name and session number are displayed as described in case a1) and b1). Table 9.2: Result from the integration test of the scenario Entering training/testing mode. Training/testing mode: Data acquisition. Aim: It should be tested if it is possible to achieve online EEG and EMG data in training/testing mode and display these in the designated graphs. Method: The user presses Start. This should initiate the data acquisition for the graphs. Input: a) EEG and EMG electrodes are connected correctly to the system. b) EEG and EMG electrodes are incorrectly connected to the system. Expected output: a) Both EEG and EMG signals are displayed correctly. b) EEG and/or EMG signals are not present. c) EEG and/or EMG signals are displaying noise. Result: Both EEG and EMG signals are functioning as expected. Table 9.3: Result from the integration test of the scenario Training/testing mode: Data acquisition. 9. Testing 45

54 9.1 Tests of the developing system Training/testing mode: Data saving Aim: It should be tested if it is possible to save the recorded EMG and EEG data in the training/testing mode with the corresponding name and session number, when the user presses Save. Method: After starting a recording the user presses Stop. A dialog window will pop up prompting the user to decide whether the recorded data should be saved or not. Input: a) EEG and EMG signals, subject name and current training/testing session number. Expected output: The data from both EEG and EMG signals are saved correctly in a file, which carries the name of the subject, the training type and the session number. Result: The saving function is working as expected. Table 9.4: Result from the integration test of the scenario Training/testing mode: Data saving. Detector menu: INP template extraction. Aim: It should be tested if it is possible to extract INP template from one or multiple training sessions. Method: The user can access the detector builder in two ways. He can select to record training sessions first and build the detector afterwards, or he can go to the detector building directly from the login screen by pressing Testing. When the user accesses the detector builder menu he can select several training sessions by holding in the ALT-key on his keyboard. When the wanted training sessions are selected the user has to enter the used gain and press Build detector menu. Input: One or multiple training sessions. Expected output: Three graphs are displayed in the detector menu, showing average INP, INP template and standard deviation and power on the EOG channel for each epoch. Result: INP template is extracted and the three graphs are displayed. Table 9.5: Result from the integration test of the scenario Detector menu: INP template extraction from multiple files Testing

55 9.1 Tests of the developing system Detector menu: Set thresholds and build detector. Aim: It should be tested if it is possible to enter a threshold value for the power in the EOG channel (FP1) and a threshold of true positive (TP) and false positive (FP) values based on a ROC-curve. Method: The user is entering a threshold value and is directed to the next window which contains a ROC-curve. Here the user selects a data point based on the ROC curve, representing a false positive and true positive value, and enter this value in the corresponding fields. If the user pushes the OK button, the system will prompt that the detector is build. Input: a) The user will enter a threshold value for FP1 and a valid data point of TP and FP values. b) The user will enter a threshold value for FP1 and an invalid data point of TP and FP values. Expected output: a) The detector is build if the input values are within the allowed limit. b) If the input values are out of bounce, an error message will be prompted. Result: The detector is build successfully when valid values are entered. If invalid numbers are entered, the system will prompt an error message and the user has to enter new value. Table 9.6: Result from the integration test of the scenario Detector menu: Set thresholds and build detector. Testing: Stimulation. Aim: It should be tested if it is possible to send a trigger, based on a detected INP and toggle the stimulation mode on and off. Method: The subject will have the EEG cap mounted on the head and is instructed to imagine a movement. EEG signals are recorded in the testing mode. To identify if a trigger command is send, the electrical stimulator is connected to an oscilloscope. The same setup is used when the radio button for the stimulation mode is enabled and disabled. Input: The EEG signal of the subject. Expected output: a) Stimulation enabled: a trigger command is send when an INP is detected, the stimulation signal can be seen on the oscilloscope. b) Stimulation disabled: a trigger command is send when an INP is detected, no stimulation signal can be seen at on the oscilloscope. Result: The stimulation was given, visualized on the oscilloscope, when an INP was detected with the stimulation mode enabled. Table 9.7: Result from the integration test of the scenario Testing Mode: Stimulation. 9. Testing 47

56 9.1 Tests of the developing system Testing: Subject feedback. Aim: It should be tested if it is possible to detect finger movements from the subject to indicate when a stimulation has happened. Method: EMG signals are recorded from the index finger of the left and the right hand. a) The subject should extend his right finger, when he was thinking of a movement and got a stimulation, which will make the true positive (TP) count one up. b) The subject should move his left index finger, when he was thinking of a movement, but got no stimulation, which will count false negative (FN) one up. c) The subject should not move any of his fingers, when he was not thinking of a movement, but got a stimulation, which will count false positive (FP) one up. Input: EMG signal from the right and left index finger. Expected output: The counters from TP, FP and FN will be controlled from the two fingers respectively. Result: It was possible to control all counters with the two fingers. Table 9.8: Result from the integration test of the scenario Testing: Triggering System test The system test verifies if the requirements stated in chapter 6 are fulfilled by the system. The test will be done on the prototype of the system and is tested as a black box test, where only input to the system is known. Aim The aim for the system test is to document that the functionality of the program fulfills the requirements. Method The system test evaluates every single entry one by one. If any mismatches are found between the requirements and the system under development, it has to be corrected. This documentation is from the last iteration of the system and will therefore fulfill all requirements Testing

57 9.1 Tests of the developing system Results Requirements The system will record nine channel EEG measured from the scalp. The system will record one channel EOG measured from the eye. The system will record three channel EMG measured from the tibialis anterior muscle and the right and left index finger. The system will allow the user to make recordings with different movement parameters. The system will store all recordings locally. The system will display EEG and EMG signals online. The user will have the possibility to display one of nine recorded EEG channels at a time. The system will give visual feedback to the subject about movement performance. The system will send a trigger signal when a INP is detected in order to stimulate. Results It is possible to record from all EEG channels connected to the system, so the requirement is fulfilled. It is possible to record from the EOG channel connected to the system, so the requirement is fulfilled. It is possible to record from all the EMG channels connected to the system, so the requirement is fulfilled. This requirement is fulfilled with the two dropdown menus in the training window, where the user can select between imaginary and real movements and asynchronous and synchronous movements. All data is saved on the computer which run the system, so this requirement is fulfilled. This requirement is fulfilled. It is possible to show all EEG signals by selecting the channel of interest in the dropdown menu and it is possible to show the EMG from the tibialis anterior muscle, which is of interest to visualize during the experiment, unlike the EMGs from the index fingers. As stated in one of the previous requirements, this demand is fulfilled. The subject will get a subject window displayed which contains a bar, showing a live image of the movement performance, and a counter, thus this requirement is fulfilled. An electrical stimulation is sent when a INP is detected, so this requirement are fulfilled. Table 9.9: Results from the system test of the system LoFi usability test The following test was performed in the initial phase of the system on a paper-based LoFi prototype. The results from this test were discussed within the group and solutions were made, where it was seen necessary. Aim The intention of this usability test is to see if the test user is able to move logically in and between the Training and the Testing mode. Furthermore, it would be investigated if the placement of the different setting buttons can be improved in order to give the user a better overview. Finally the test should help to detect redundant or missing functions. 9. Testing 49

58 9.1 Tests of the developing system Experimental procedure The following usability test will include 4 test users in the age of 22 to 26. The test is performed in a usability laboratory at Aalborg University. During the performance, the test will be videotaped together with sound. The test user will sit in front of a table where a printed version of the LoFi prototype is presented. A task list will be given to the test user. One group member will sit next to the test subject reading out the introductions F.1 and guide the test user through the test course. Whenever necessary it is allowed to help the test user. Another group member is taking notes. The test user is asked to filled out a questionnaire to evaluate the program. The answers of the anonymous questionnaires can be found in the appendix F.1.3. A summary of the answers is given in the result part. The task list: 1. Start the system using the short cut on the desktop. 2. Choose the mode Testing for a subject named Hans Jenson. 3. Before starting the recording make the following settings: Change the EEG channel to e.g. Cz. Choose Sync as BCI movement. 4. Make a recording in the length of 30 seconds. 5. During the recording use the zoom function for both the graphs EMG and EEG. 6. Make two more recordings and save only one of them. 7. Press Build detector in order to look at the performance of the current algorithm. 8. Change to mode Testing and make the following setting: Change the visibility settings for the subject window to Time disabled. 9. Save the recording. 10. Leave the program. Results The test is evaluated by analysis of the video tapes and the questionnaire evaluation, where each task was evaluated with a score between 0-10 where 10 represented the best possible score and 0 the worst, and a written comment. The following graph (figure 9.1) shows the average score of each question, where the average test score was 7.6. Figure 9.1: Average score for each question of the questionnaire. The red line indicates the mean score of the test users for all questions Testing

59 9.2 Test of the finished system Following is a summary over the key points said during the test and written in the questionnaires: The login window is clear and easy to understand. Regarding to the changeable options, for example the channels and different training patterns, it is confusing for the subject to decide when it is appropriate to change the options. Most of the options are understandable except the Build detector button. Hence, it was suggested by the test subjects that the explanations of the some of the options should be more detailed. The interface was in general rated user friendly even though the whole system seemed to become more clear and familiar after using it several times. The saving option is simple and secure, but for the proficient user, it may be annoying. The structure of window is reasonable, however, it would be better if the EEG and EMG display windows can be adjusted to a bigger size. Based on the usability test, it can be concluded that the LoFi imitation of system basically is feasible and functional and can be used of rehabilitation of stroke patients. The dominant problems are the size of graphs and position of buttons. These problems are resolved during the implementation iterations. 9.2 Test of the finished system The testing of the finished system is just as important as testing of the developing system. The finished system has to be tested with different goals in mind, compared to the tests performed in the development of the system. While the test of the developing system is focusing on debugging, the following tests will investigate how the finished system is performing as a scientific tool and how it is controlled. The following tests are performed on the finished system: Performance test Usability test There will not be made any attempts to improve the system based on the test results Performance test To investigate the performance of the system, a performance test has to be done. The performance test measures the systems capability of detecting INP and give the trigger for the external device. The full experiment protocol can be found in appendix E. Subjects Four healthy subjects participated in the experiments. All subjects were from the work group and gave their informed consent before participating and were allowed to withdraw from the experiment at any time. None of the subjects had any known psychological disorders, sensory-motor deficits or neurologic disorders. 9. Testing 51

60 9.2 Test of the finished system Experimental procedure When the subject was equipped with the equipment, the subject was instructed about the experiment procedure. In order to get familiar with the setup the subject was asked to practice make a ballistic dorsal finger flexion and a forceful ballistic dorsal foot flexion in the range of % of maximum voluntary contraction. The experiment was divided into two parts. First, asynchronous real movements were recorded in the training mode. After a detection algorithm was build from the training sessions, it was tested on imaginary asynchronous movements in the testing mode. In the training mode a dorsal flexion had to be done at random intervals. The subject was asked to focus on the task just prior the movement. The dorsal flexion had to be quick and powerful and the subjects were told to maintain the same force and rate during the session. In order to do so visual feedback was provided to the subjects on a computer screen. Two training sessions were recorded with a duration of 5 minutes respectively For the asynchronous imagined movements in the testing mode, the subjects were asked to imagine the perception of doing a dorsal flexion. Due to the prematurely state of the system, the stimulation was not given to the subject, but to an oscilloscope, which displayed the stimulation. The subject should indicate with his right and left index finger whether the stimulation was initiated by a subject generated INP. These values for TP, TN and FP were counted automatically by the system and the performance of the online stimulation was read directly from these results. The subjects were allowed to do real movements five seconds after having thought of a movement to help them imagining the movement. This phase continued until 50 TP values had been reached. Results Recordings from two subjects were excluded, due to an amplifier overload during the test. This resulted in the signal to get saturated, which means that no appropriate template could be derived from the training session. Due to time constraints no new recordings were made. The performance test of the online stimulation gave the following results, based on the two training sessions that succeeded: Subject Detection FP/min Duration of testing Total number of TPR % session tasks attempted Table 9.10: Performance test of the online detection of INP Usability test Unlike the LoFi usability test, this test has the purpose to investigate how well a potential end-user interact with the system. The following test was performed with the final version of the system. Aim The purpose of the usability test is get in-depth opinions from an experience group and to determine whether the user is able to use the GUI for the rehabilitation of stroke patients Testing

61 9.3 Test summary Experimental procedure. The test users selected for this usability test, were users who already were familiar with similar systems for recording EEGs. Due to their preknowledge from similar systems, the users could give expert comments during the test. This test was performed in an informal manner in the group room. A script located in appendix F.1, was read out to the test user in order to give an understanding of the background of the program. The test was carried out by giving an experienced user a task list, which the user has to perform. The task list: 1. Start the system using the short cut on the desktop. 2. Choose the mode Testing for a subject named Hans Jenson. 3. Before starting the recording make the following settings: Change the EEG channel to e.g. Cz. Choose Sync as BCI movement. 4. Make a recording in the length of 30 seconds. 5. Make two more recordings and save only one of them. 6. Press Build detector in order to look at the performance of the current algorithm. 7. Change to mode Testing and make the following setting: Change the visibility settings for the subject window to Time disabled. 8. Save the recording. 9. Leave the program. The intention with this test was to discuss problems and get comments on the flow and appearance on the system. While the test user was performing the mentioned tasks, one group member was responsible for making notes about any particular events or comments from the test user. These notes are in appendix F.3. The key points are included in the result section. Results The expert group were in general very content with the system, although there were both minor and major things they mentioned should be replaced or added. The key points are explained below. A majority of the expert group stated that the EMG and EEG window should be labeled according to which signal they displayed. A majority also mentioned that a help guide added in the system, will help clarify questions, which the user encounters. Two users claimed that the system opens too many windows, which will be confusing to operate with. Further comments were concerning the layout and buttons where alternatives were suggested. 9.3 Test summary The system was tested in several ways both during the development of the system and when the system was finalized. Continuously during the implementation of the system, there has been made unit test and integration test so bugs could be mended as they appeared, thus no documentation is available for these tests. The integration test was done after the basic implementation to validate if consecutive parts could fit together. The integration test resulted in a positive result, since all expected outcomes were fulfilled. The system test confirmed that the requirements were respected. The usability test in the development of the test was performed on non-experts and showed that the system is understandable, especially when the user gets acquainted with the system. In a scale 9. Testing 53

62 9.3 Test summary between 0 to 10, the system under development got a score on 7.6. The test users came with several suggestions for improvements of the system, which is implemented in later versions of the system. Where the testing of the system in development concentrated on how the system works, the tests of the finished system focused more on how good the system works. The performance test measured the systems capability of detecting INP and give the stimulation, based on the results from the training session. Based on two subjects, the average true positive rate were %. The second usability test was done on a working finished system with experienced users, and showed that the users were satisfied with the system. Although the users had some minor and major changes as suggestion Testing

63 PART III SYNTHESIS 55

64 56

65 Chapter 10 Discussion The problem analysis contains several considerations, all related to the rehabilitation of stroke. The most important considerations are discussed in the first section of this chapter. Afterwards the developed BCI system is discussed. Current rehabilitation strategies like physiotherapy and robotic therapy share two characteristics. On one hand they are passive methods, meaning that the subjects initiation of movement cannot be respected. On the other hand they are methods based on repetitive movements with a constrained movement orientation. Unlike these a BCI system used for rehabilitation purposes has a different approach, since brain signals can be detected in order to provide feedback to the subject. The developed BCI system, this report refers to, can detect INP and send a trigger. In the implemented system the trigger is used in form of visual feedback. Furthermore, the trigger can be used by the patient to control a variety of external devices (e.g. FES or robotic systems), which can support the movement execution. While treatments like CIMT expect that the patient has already a certain ability to move the impaired limb, BCI systems only expect a certain level of cognitive awareness in order to adjust the system to the subject. Our BCI system uses the INP of MRCP due to the fact, that INP occurs before the onset of the movement, and thereby can be used to predict a movement. Other indicators, which can be used as control signals for BCI, could be µ and β rhythms and P300 waves, all found in an EEG. MRCP was chosen as a control signal, due to previous research at Center for Sensory-Motor Interaction (SMI) at Aalborg University. Unlike the current system, the previous system developed at SMI, worked with an offline detection of INP, where the system in hand is working with an online detection of INP. We managed to build a working system which is capable of detecting INP with an average TPR of %. However to obtain a statistical significant result on the performance of the system more subjects are needed. It can be expected that the result would get better when more training trials are included. Only results from 50 % of our subjects were acceptable for further analysis, due to problems with the used amplifiers. These problems can be caused either by complications with the equipment or the setup, e.g. high impedance. A solution for these problems could be to revise the experimental protocol and use more reliable equipment. The used detection algorithm also have an influence on the performance, but it was not in the scope of this project to improve this part of the system. The test results from the offline version of the system had a TPR of 63.3 %. With the online result of % it can be assumed that the two systems are performing similar. It can be argued that the performance test should be tested on rehabilitation subjects, due to the difference in the morphology of MRCPs compared to healthy subjects. Since the system is in the development phase only a proof of concept was required and therefore no tests were performed on stroke patients. Two usability tests were performed on the system. The first usability test was done on a LoFi prototype in the initial phase of the system development, and served as a guide for how the system should be designed. The LoFi prototype obtained a score of 7.4. This test was performed on regular users in order to test the logic flow of the system. A second usability test was performed on the finished system in order to test the usability and improve the GUI. However, due to its high speciality the test was only carried out with users having a background knowledge regarding BCI and EEG signals. The comments implied that the system 57

66 was user friendly and easy to navigate through, but had some flaws. The current system is focusing on research purposes. When this system should be used for rehabilitation in for instance clinics, another usability test has to be performed on the end-user segment before the system is released Discussion

67 Chapter 11 Conclusion In Denmark it is estimated that there are incidents of stroke every year of which 50 % are in need of rehabilitation. The post stroke rehabilitation is a very time consuming process and expensive for the healthcare system. The current rehabilitation methods do not provide significant improvements concerning recovery of motor dysfunction. A new approach might be an introduction of a BCI system for rehabilitation purposes based on MRCP. The goal for this project was to develop an online BCI system that can detect INP. The integration test showed that the system was capable of detecting INP and provide a trigger, while the system test proved that all the functional requirements were fulfilled. The performance test showed a TPR of % for the online stimulation. However, the results were only based on two healthy subjects and therefore no statistical evidence can be concluded. In order to obtain a more valid result a higher number of subjects is required. The usability test performed with experienced users supported the positive feedback from the LoFi usability test, but also revealed some flaws. For instance, an implemented help function as well as a FAQ would be of great benefit for the usability. Furthermore, the number of windows, presented to the user, should be reduced in order to give a better overview and enhance the usability. As the system evolves and more features will be implemented, a code optimizing has to be done to keep the feedback latency below 300 ms, which is necessary to induce neural plasticity. Like most studies concerning BCI for therapeutic purposes, the tests for this system were carried out only with healthy subjects. Therefore the obtained results cannot be generalized for patients who are actually in need of BCI. Prior the prospective use of the system for rehabilitation, it has to be tested on patients suffering from cerebrovascular disease. 59

68 Conclusion

69 Bibliography Alamancos, Segura, and Borrell, M. C. Alamancos, L. G. Segura, and J. Borrell. Transfer of function to a specific area of the cortex after induced recovery from brain damage. Eur J Neurosci, 4, , Arlow and Neustadt, Jim Arlow and Ila Neustadt. UML 2 and the Unified Process, Second Edition. Pearson Education, Barsi, Popovic, M.Tarkka, Sinkjær, and Grey, G. I. Barsi, D. B. Popovic, I. M.Tarkka, T. Sinkjær, and M. J. Grey. Cortical excitability changes following grasping exercise augmented with electrical stimulation. Exp Brain Res, 191, 57 66, Birbaumer and Cohen, Niels Birbaumer and Leonardo G. Cohen. Brain computer interfaces: communication and restoration of movement in paralysis. J Physiol, 579.3, , Birbaumer, Murguialday, and Cohen, Niels Birbaumer, Ander Ramos Murguialday, and Leonardo Cohen. Brain computer interface in paralysis. Neurology, 21, , Caspers and Speckmann, H. Caspers and E.J. Speckmann. Handbook of Electroencephalography and Clinical Neurophysiology, volume 10A. Elsevier, Chae and Yu, J. Chae and D. Yu. Neuromuscular stimulation for motor relearning inhemiplegia. Crit Rev Phys Med Rehabil Med, 11, , Cirillo, Lavender, Ridding, and Semmler, J. Cirillo, A. P. Lavender, M. C. Ridding, and J. G. Semmler. Motor cortex plasticity induced by paired associative stimulation is enhanced in physically active individuals. Physiology, , , Classen, Wolters, Stefan, Wycislo, Sandbrink, Schmidt, and Kunesch, Joseph Classen, Alexander Wolters, Katja Stefan, Matthias Wycislo, Friedhelm Sandbrink, Arne Schmidt, and Erwin Kunesch. Paired associative stimulation. Advances in Clinical Neurophysiology, 57, 7, Conforto, Kaelin, and Cohen, AB Conforto, L Kaelin, and LG Cohen. Increase in hand muscle strength of stroke patients after somatosensory stimulation. Ann Neurol, 51, , Cunnington, Iansek, Bradshaw, and Phillips, R. Cunnington, R. Iansek, J. L. Bradshaw, and J. G. Phillips. Movement-related potentials associated with movement preparation and motor imagery. Exp Brain Res, 111, , Daly and Wolpaw, Janis J. Daly and Jonathan R. Wolpaw. Brain-computer interfaces in neurological rehabilitation. The Lancet Neurology, 7, Issue 11, , Vries and Mulder, Sjoerd de Vries and Theo Mulder. Motor imagery and stroke rehablitation: A critical discussion. J Rehabil Med, 39, 5 13, Deecke, Kornhuber, Lang, and Schreiber, L. Deecke, H.H. Kornhuber, W. Lang, and H. Schreiber. Timing function of the frontal cortex in sequential motor and learning tasks. Human Neurobiol, 4, ,

70 BIBLIOGRAPHY Despopoulos and Silbernagl, Agamemnon Despopoulos and Stefan Silbernagl. Color Atlas of Physiology. Thieme, Nascimento, Nielsen, and Voigt, O. F. do Nascimento, K. Dremstrup Nielsen, and M. Voigt. Relationship between plantar-flexor torque generation and the magnitude of the movement-related potentials. Exp Brain Res, 160, , Nascimento, Nielsen, and Voigt, O. F. do Nascimento, K. Dremstrup Nielsen, and M. Voigt. Movement-related parameters modulate cortical activity during imaginary isometric plantar-flexions. Exp Brain Res, 171, 78 90, Dornhege, Millán, Hinterberger, McFarland, and Müller, G. Dornhege, J. R. Millán, T. Hinterberger, D. J. McFarland, and K. R. Müller. Toward Brain-Computer Interfacing. The MIT Press, Driskell, Copper, and Moran, J. E. Driskell, C. Copper, and A. Moran. Does mental practice enhance performance. J Sport Psychol, 79, , Fadiga, Fogassi, Pavesi, and Rizzolatti, L. Fadiga, L. Fogassi, G. Pavesi, and G. Rizzolatti. Motor facilitation during action observation: a magnetic stimulation study. J Neurophysiol, 73, , Fasoli, Krebs, and Hogan, S.E. Fasoli, H.I. Krebs, and N. Hogan. Robotic Technology and Stroke Rehabilitation: Translating Research into Practice. TOPICS IN STROKE REHABILITATION, 11, 11 19, Feltz and Landers, D. L. Feltz and D. M. Landers. The effects of mental practice on motor skill learning and performance: a meta-analysis. J Sport Psychol, 5, 25 57, Frak, Paulignan, and Jeannerod, V Frak, Y Paulignan, and M Jeannerod. Orientation of the opposition axis in mentally simulated grasping. Exp Brain Res, 136, , Fraser, Power, Hamdy, Rothwell, Hobday, Hollander, Tyrell, Hobson, Williams, and Thompson, C. Fraser, M. Power, S. Hamdy, J. Rothwell, D. Hobday, I. Hollander, P. Tyrell, A. Hobson, S. Williams, and D. Thompson. Driving plasticity in human adult motor cortex is associated with improved motor function after brain injury. Neuron, 34, 5 40, Gallese, V. Gallese. Embodied simulation: from neurons to phenomenal experience. Phenomenol Cognitive Sci, 4, 23 38, Grosse-Wentrup, Mattia, and Oweis, Moritz Grosse-Wentrup, Donatella Mattia, and Karim Oweis. Using brain computer interfaces to induce neural plasticity and restore function. Journal of Neural Engineering, 8 (2), 1 5, Grotta, Noser, Ro, Boake, Levin, Aronowski, and Schallert, J. C. Grotta, E. A. Noser, T. Ro, C. Boake, H. Levin, J. Aronowski, and T. Schallert. Constraint-Induced Movement Therapy. Stroke, 35, , Guyton and Hall, Arthur C. Guyton and John E. Hall. Textbook of Medical Physiology, 11th Edition. Saunders/Elsevier, Hebb, D.O. Hebb. The organization of behavior. A neuropsychological theory. New York: Wiley, Hendersen, Steven Hendersen. Why physiotherapy is so important in stroke rehabilitation BIBLIOGRAPHY

71 BIBLIOGRAPHY Hewitt, P. Hewitt. What is Functional Electrical Stimulation. Worldwide Health, Hlustik and Mayer, P. Hlustik and M. Mayer. Paretic hand in stroke: from motor cortical plasticity research to rehabilitation. Cogn Behav Neurol, 19, 34 40, National Instruments, National Instruments. NI 6023E/6024E/6025E Family Specifications, A. Jackson, P. Culmer, S. Makower, M. Levesley, R. Richardson, A. Cozens, M. M. Williams, and B. Bhakta, A. Jackson, P. Culmer, S. Makower, M. Levesley, R. Richardson, A. Cozens, M. M. Williams, and B. Bhakta. Initial patient testing of ipam - a robotic system for Stroke rehabilitation. In IEEE 10th International Conference - Rehabilitation Robotics, pages , Jackson, Lafleur, Malouin, Richards, and Doyon, P. L. Jackson, M. F. Lafleur, F. Malouin, C. Richards, and J. Doyon. Potential role of mental practice using motor imagery in neurologic rehabilitation. Arch Phys Med Rehabil, 82, , Jahanshahi and Hallett, Marjan Jahanshahi and Mark Hallett. The Bereitschaftpotential - Movement-Related Cortical Potentials. Kluwer Academic/Plenum Publishers, New York, Jayaram, Santos, and Stinear, G. Jayaram, L. Santos, and J.W. Stinear. Spike-timing-dependent plasiticity induced in resting lower limb cortex persists during subsequent walking. Brain Research, 1153, 92 97, Jeannerod, M Jeannerod. Neural simulation of action: a unifying mechanism for motor cognition. Neuroimage, 14, , Jenkins and Merzenich, W. M. Jenkins and M. M. Merzenich. Reorganisation of neocortical representations after brain injury: an europhysiological model of the bases of recovery from stroke. Progress in brain research, 71, 249, Kitamura, Shibasaki, and Kondo, J. Kitamura, H. Shibasaki, and T. Kondo. A cortical slow potential is larger before an isolated movement of a single finger than simultaneous movement of two fingers. Electroenceph Clin Neurophysiol, 86, 252 8, Kornhuber and Deecke, H.H. Kornhuber and L. Deecke. Hirnpotentialänderungen beim Menschen vor und nach Willkürbewegungen, dargestellt mit Magnetband-Speicherung und Rückwärtsanalyse. Pflügers Arch, 281, 52, Kristeva, Cheyne, and Deecke, R. Kristeva, D. Cheyne, and L. Deecke. Neuromagnetic fields accompanying unilateral and bilateral voluntary movements: topography and analysis of cortical sources. EEG Clin Neurophysiol, 81, , Kwakkel, Wagenaar, Koelman, Lankhorst, and Koetsier, G. Kwakkel, R.C. Wagenaar, T. W. Koelman, G. J. Lankhorst, and J. C. Koetsier. Effects of Intensity of Rehabilitation After Stroke. Stroke, 28, , Kwakkel, Peppen, Wagenaar, Wood, Richards, Ashburn, and al., G. Kwakkel, R. van Peppen, R. C. Wagenaar, D. D. Wood, C. Richards, A. Ashburn, and et al. Effects of augmented exercise therapy time after stroke: a meta-analysis. Stroke, 35, , Libet, Gleason, Wright, and Pearl, B. Libet, C. A. Gleason, E.W. Wright, and D.K. Pearl. Time of conscious intention to act in relation to onset of cerebral activity. Brain, 106, , BIBLIOGRAPHY 63

72 BIBLIOGRAPHY Mackay and Mensah, Judith Mackay and George A. Mensah. The Atlas of Heart Disease and Stroke. World Health Organization, Martini, Frederic H. Martini. Fundamentals of Anatomy & Physiology. Pearson Education, Masaki, Takasawa, and Yamazaki, H. Masaki, N. Takasawa, and K. Yamazaki. Enhanced negative slope of the readiness potential preceding a target force production task. Electroenceph Clin Neurophysiol, 95, 390 7, McCallum and Curry, W.C. McCallum and S.H. Curry. Slow potential changes in the human brain. Plenum Press, New York, McDonnell and Ridding, M.N. McDonnell and M.C. Ridding. Afferent stimulation facilitates performance on a novel motor task. Exp Brain Res, 170, , McKay, Ridding, Thompson, and Miles, DR McKay, MC Ridding, PD Thompson, and TS Miles. Induction of persistent changes in the organisation of the human motor cortex. Exp Brain Res, 143, , Mellinger, Schalk, Braun, Preissl, Rosenstiel, Birbaumer, and Kübler, J. Mellinger, G. Schalk, C. Braun, H. Preissl, W. Rosenstiel, N. Birbaumer, and A. Kübler. An MEG-based brain computer interface (BCI). NeuroImage, 36, , Misulis and Head, Karl E. Misulis and Thomas C. Head. Essentials of Clinical Neurophysiology, Third Edition. Butterworth-Heinemann, Muellbacher, Ziemann, Boroojerdi, Cohen, and Hallett, W. Muellbacher, U. Ziemann, B. Boroojerdi, L. Cohen, and M. Hallett. Role of the human motor cortex in rapid motor learning. Exp Brain Res, 136, , Mulder, Vries, and Zijlstra, T. Mulder, S de Vries, and S. Zijlstra. Observation, imagination and execution of an effortful movement: more evidence for a central explanation of motor imagery. Experimental Brain Research, 163, , Neshige, Luders, and Shibasaki, R. Neshige, H. Luders, and H. Shibasaki. Recording of movement-related potentials from scalp and cortex in man. Brain, 111, , I. K. Niazi, N. Jiang, O. Tiberghien, J. F. Nielsen, K. Dremstrup, and D. Farina. Detection of movement intention from movement-related cortical potentials. Nudo, R. J. Nudo. Functional and structural plasticity in motorcortex: implication for stroke recovery. Phys Med Rehabil Clin N Am, 14, , Nudo, R. J. Nudo. Postinfarct cortical plasticity and behavioral recovery. Stroke, 38, , Malaysia, NASAM National Stroke Association of Malaysia. Physiotherapy Orkano and Tanji, K. Orkano and J. Tanji. Neuronal activities in the primate motor fields of the agranular frontal cortex preceding visually triggered and self-paceed movement. Experimental Brain Research, 66, , O Dell, Lin, and Harrison, Michael W. O Dell, Chi-Chang David Lin, and Victoria Harrison. Stroke Rehabilitation: Strategies to Enhance Motor Recovery. The Annual Review of Medicine, 60, 55 68, BIBLIOGRAPHY

73 BIBLIOGRAPHY Page, Sisto, and Johnston, S. J. Page, P. Levine ANDS. Sisto, and M. V. Johnston. A randomized efficacy and feasibility study of imagery in acute stroke. Clin Rehabil, 15, , Page, Levine, and Leonard, S. J. Page, P. Levine, and A. C. Leonard. Effects of mental practice on affected limb use and function in chronic stroke. Arch Phys Med Rehabil, 86, , Parsons, LM Parsons. Integrating cognitive psychology, neurology and neuroimaging. Acta Psychol (Amst), 107, , Pascual-Leone, Grafman, and Hallett, A Pascual-Leone, J Grafman, and M. Hallett. Modulation of cortical motor output maps during development of implicit and explicit knowledge. Science, 263, , Paulsen and Sejnowski, O. Paulsen and T.J. Sejnowski. Natural patterns of activity and long-term synaptic plasticity. Curr. Opin. Neurobiol., 10(2), , Peckham and Knutson, P.H. Peckham and J.S. Knutson. FUNCTIONAL ELECTRICAL STIMULATION FOR NEUROMUSCULAR APPLICATIONS. Biomed.Eng., 7, , Perez, Lungholt, Nyborg, and Nielsen, M. A. Perez, B. K. S. Lungholt, K. Nyborg, and J. B. Nielsen. Motor skill training induces changes in the excitability of the leg cortical area in healthy humans. Exp Brain Res, 159, , Piero, Chollet, Carthy, Lenzi, and Frackowiak, V.D. Piero, F. Chollet, P.M. Carthy, G. Lenzi, and R. Frackowiak. Motor recovery after acute ischaemic stroke: a metabolic study. Neurol Neurosurg Psychiatr, 55, , Powell, Pandyan, Granat, M.Cameron, and Stott, J. Powell, D. Pandyan, M. Granat, M.Cameron, and D. J. Stott. Electrical stimulation of wrist extensors inpost stroke hemiplegia. Stroke, 30, , Praamstra, Stegeman, and M. W. I. M. Horstink, C. H. M. Brunia, P. Praamstra, D. F. Stegeman, and A. R. Cools M. W. I. M. Horstink, C. H. M. Brunia. Movement-related potentials preceding voluntary movement are modulated by the mode of movement selection. Experimental Brain Research (Springer), 103, , Raymond, T. K. Y Raymond. Wiley Encyclopedia of Biomedical Engineering. John Wiley & Sons, Inc., Ridding and Rothwell, M. C. Ridding and J. C. Rothwell. Afferent input and cortical organisation: a study with magnetic stimulation. Exp Brain Res, 126, , Ridding and Uy, M.C. Ridding and J. Uy. Changes in motor cortical excitability induced by paired associative stimulation. Clinical Neurophysiology, 114, 8, Rizzolatti, G. Rizzolatti. The mirror neuron system and imitation. Perspectives on Imitation: From Neuroscience to Social Science, 1, 55 77, Rizzolatti, Gentilucci, Fogassi, Matelli, and Ponzoni-Maggi, G. Rizzolatti, M. Gentilucci, L. Fogassi, G. Luppino M. Matelli, and S. Ponzoni-Maggi. Neurons related to goal-directed motor acts in inferior area 6 of the macaque monkey. Exp Brain Res, 67, , BIBLIOGRAPHY 65

74 BIBLIOGRAPHY Rogers, Brown, and Stinear, L.M. Rogers, D.A. Brown, and J.W. Stinear. The effect of paired associative stimulation on knee extensor motor excitability of individuals post-stroke: A pilot study. Clinical Neurophysiology, Rossini and Rossi, Paolo M. Rossini and Simone Rossi. Transcranial magnetic stimulation: Diagnostic, therapeutic, and research potential. Neurology, 68 (7), , Rushton, D.N. Rushton. Functional electrical stimulation. Physiol, 18, , Shibasaki and Hallett, Hiroshi Shibasaki and Mark Hallett. What is the Bereitschaftspotential. Clinical Neurophysiology, 117, , Skinner and Yingling, J.E. Skinner and C.D. Yingling. Regulation of slow potential shifts in nucleus reticularis thalami by the mesencepahic reticular formation and the frontal granular cortex. EEG & Clinical Neurophysiology, 40, , Stefan, Kunesch, Cohen, Benecke, and Classen, K. Stefan, E. Kunesch, L.G. Cohen, R. Benecke, and J. Classen. Induction of plasticity in the human motor cortex by paired associative stimulation. Brain, 123, , Stinear and Hornby, J.W. Stinear and T. G. Hornby. Stimulation-induced changes in lower limb corticomotor excitability during treadmill walking in humans. The Physiological Society, 567.2, , Sundhedsstyrelsen, Sundhedsdokumentation Sundhedsstyrelsen. Hjerneskaderehabilitering en medicinsk teknologivurdering. Sundhedsstyrelsen, Taub, Uswatte, Lee, Lankhorst, Bouter, and Wagenaar, E. Taub, G. Uswatte, J.H. van der Lee, G. J. Lankhorst, L. M. Bouter, and R. C. Wagenaar. Constraint-Induced Movement Therapy and Massed Practice Response. Stroke, 31, , Taylor, Margo J. Taylor. Bereitschaftspotential during the acquisition of a skilled motor task. Electroencephalography and Clinical Neurophysiology, 45, , Trafton, Anne Trafton. Robotic therapy helps stroke patients regain function. MIT news, Wassermann and Lisanby, E. M. Wassermann and S. H. Lisanby. Therapeutic application of repetitive transcranial magnetic stimulation: a review. Clinical Neurophysiology, 112, , Wessel, Verleger, Nazarenus, Vieregge, and Koempf, K. Wessel, R. Verleger, D. Nazarenus, P. Vieregge, and D. Koempf. Movement-related cortical potentials preceding sequential and goal-directed finger and arm movements in patients with cerebellar atrophy. Electroenceph Clin Neurophysiol, 92, , Yom-Tov and Inbar, E. Yom-Tov and G.F. Inbar. Detection of movement-related potentials from the electro-encephalogram for possible use in a brain-computer interface. Med. Biol. Eng. Comput., 41, 85 93, Ziemann, U. Ziemann. Transcranial Magnetic Stimulation at the Interface with Other Techniques: A Powerful Tool for Studying the Human Cortex. The Neuroscientist, doi: / BIBLIOGRAPHY

75 BIBLIOGRAPHY Ziemann, Iliac, Pauli, Meintzschel, and Ruge, U. Ziemann, T. V. Iliac, C. Pauli, F. Meintzschel, and D. Ruge. LearningModifies Subsequent Induction of Long-Term Potentiation-Like and Long-TermDepression-Like Plasticity in HumanMotor Cortex. The Journal of Neuroscience, 24, , BIBLIOGRAPHY 67

76 BIBLIOGRAPHY 68 BIBLIOGRAPHY

77 PART IV APPENDICES 69

78 70

79 Appendix A Hebbian principle D. O. Hebb postulated in 1949 a hypothesis about, that the timing of neural events are crucial parameters for the induction of changes in synaptic excitability. He stated the following: When an axon of cell A is near enough to excite a cell B and repeatedly or persistently takes part in firing it, some growth process or metabolic change takes place in one or both cells such that A s efficiency, as one of the cells firing B, is increased. [Hebb, 1949] It originally postulated that the strength of a synapse between a pre-synaptic and a post-synaptic neuron would increase when both neurons were activated in near-synchronicity. This model has later been refined by in vitro and in vivo animal experiments where it showed that synaptic efficacy induced by associative activity is dependent on the order of activity from the pre- and postsynaptic neuron. If the postsynaptic neuron generated an action potential after the afferent stimulus had generated an excitatory postsynaptic potential (EPSP) a long-term potentiation (LTP) was induced. When the postsynaptic cell is depolarized synchronously with the afferent input from another cell or two independent afferent signals reaches the same cell in sync, the LTP is called associative or Hebbian. In contrast if the postsynaptic neuron spiked before EPSP was induced by afferent stimulus, it would result in a long-term depression. This principle is called the temporally asymmetric Hebbian rule. [Classen et al., 2004], [Paulsen and Sejnowski, 2000]. The Hebbian principle has gained a lot of attention in the rehabilitation field. LTP is considered to be an important cellular mechanism in learning and memory and by inducing changes which will strengthen the synaptic excitability, it would be possible to alter the plasticity of e.g. the motor cortex. A study by U. Ziemann which followed the PAS protocol proved the hypothesis that motor learning prevents subsequent LTP-like plasticity but enhances subsequent LTD-like plasticity. [Ziemann et al., 2004] 71

80 72 A. Hebbian principle

81 Appendix B Transcranial magnetic stimulation Transcranial magnetic stimulation (TMS) can be used to investigate noninvasively and painlessly peripheral nerves, nervous propagation along the corticospinal tract and spinal roots of the human. The technique is based on the principle of electromagnetic induction and is widely used to stimulate both peripheral nerves and brain tissue (figure B.1) in studies encompassing e.g. motor control, movement disorders, stroke, and plasticity. It has proved to be a versatile technique and is now also being used in combination with electroencephalography (EEG), functional magnetic resonance imaging (fmri) and single unit recording. [Wassermann and Lisanby, 2001] Figure B.1: In transcranial magnetic stimulation a coil is placed on the scalp. The experimental setup is here illustrated for therapy to treat depression and schizophren. The disadvantages of TMS is that it is limited to brain areas at or near the cortical surface, because the effect of the stimulation reduces with distance from the coil [Wassermann and Lisanby, 2001]. The potential of applying TMS to neurorehabilitation is high because the method is routinely combined with other techniques. [Rossini and Rossi, 2007], [Ziemann, 2011] 73

82 74 B. Transcranial magnetic stimulation

83 Appendix C Electroencephalography Electroencephalography (EEG) is a medical imaging technique that reads scalp electrical activity generated by brain structures. It is a completely non-invasive procedure that can be applied repeatedly to patients, normal adults, and children with virtually no risk or limitation. The typically clinical use of EEG can be to diagnose epilepsy, brain death, encephalopathy or monitor anesthesia. For research EEG is used in neuroscience, psychophysiological research and cognitive science and psychology. [Misulis and Head, 2003] C.1 Generation of EEG rhythms The electrical activity that can be recorded on the scalp is a summation of excitatory and inhibitory postsynaptic potentials and action potentials. These potentials comes from neurons situated in the most superficial areas of the cerebral cortex, parallel to the scalp. The electrical signals coming from deeper regions, like basal nuclei have too small amplitude compared to the that generated by the most superficial layers and thus they cannot be distinguished, but the activated thalamocortical afferents interfere with the motor output. [Misulis and Head, 2003] C.1.1 Cortical potentials The neuronal task of transmitting information is mainly realized by synapses and action potentials revealed as electrical activity. [Misulis and Head, 2003] Na + AP AP Na + Action potential (AP) Continuous charge balance 1a Depolarization Rest Refractory Na + Na + AP Depolarization 1b Figure C.1: The ionic circuit in depolarization. [Despopoulos and Silbernagl, 2003] The receptor activation may have either an excitatory or inhibitory effect on the membrane site, thus excitatory or inhibitory postsynaptic potentials, depending on what ions they increase the 75

84 C.1 Generation of EEG rhythms permeability to.these potentials are the most representative electrical activity that can be recorded from the scalp. The action potential consists of a transient depolarization of the dendritical membrane which is conducted along the soma s and axon s membrane by changes in its ionic permeability. Certain neurotransmitters act on the channels within the neuronal membrane in specialized sites and increase their permeability to the positive ions that are placed outside the cell in equilibrium state. This phenomenon is called depolarization (figure C.1). It creates in fact a dipole, a negative extracellular environment and a positive one inside the cell, that is actually the opposite of the equilibrium state and moves progressively farther. When all of the channels within one neuron s membrane are open, a peak potential is reached, followed by repolarisation, which consists of the ionic channels shutting, which are time-dependent, and the returning to the equilibrium membrane potential. A refractory period follows, blocking possible bidirectional conduction. The myelinated axons drive the influx faster that unmyelinated ones, due to the myelin sheath which implies propagation from node to node, preventing the ionic leak. [Misulis and Head, 2003] C.1.2 Scalp potentials The electrical activity recorded at the scalp surface is a summation of different electrical signals that cross the meninges, the skull and the scalp before being detected by scalp electrodes. Inside the brain most of the motor neurons have a vertical orientation with a superficial dendritical ramification, a large cell body perpendicular to scalp and deep axons. This creates a columnar organization of the cerebral cortex. EEG is a recording of vector potentials generated by cortical activity, with amplitude correlated with the total area of activated cortex and also with the synchronization of a number of activated neurons. When the most superficial cortical parts have a positive field potential and the deeper ones are negative, the vector is vertically oriented and pointing to the scalp electrodes with its positive ending. The scalp and the inactive tissues have an electrical activity as well that is recorded by EEG. [Misulis and Head, 2003] C.1.3 Classification of EEG rhythms For obtaining basic brain patterns of individuals, subjects are instructed to close their eyes and relax. Brain patterns form wave shapes that are commonly sinusoidal (figure C.2). [Misulis and Head, 2003] α β ϑ δ Normal 100µV 1 s Abnormal Frequency 8 13 Hz Hz 4 7Hz 0.5 3Hz Paroxysmal spikes Paroxysmal waves 3 Hz spikes and waves Figure C.2: The different types of brain waves; Alpha (8-13 Hz), Beta (14-30 Hz), Theta (4-7 Hz) and Delta (0.5-3 Hz). [Despopoulos and Silbernagl, 2003] Usually, they are measured from peak to peak and normally range from 0.5 to 100 µv in amplitude, 76 C. Electroencephalography

85 C.2 Recording EEG which is about 100 times lower than ECG signals. By means of Fourier transform power spectrum from the raw EEG signal is derived. In power spectrum contribution of sine waves with different frequencies are visible. Although the spectrum is continuous, ranging from 0 Hz up to one half of sampling frequency, the brain state of the individual may make certain frequencies more dominant. [Misulis and Head, 2003] C.2 Recording EEG Surface electrodes are the most common method for recording EEG. Each of the electrodes are fixed to the skin and consist gel, which give a better connection between the electrode and the skin. There are many other ways of recording EEG, such as using needle, sphenoidal, subdural strip or depth electrodes. C.2.1 Electrode position The electrodes are strategically placed across the scalp and a reference electrode is usually placed on one or both earlobes. The electrodes are placed according to The International Federation system (figure C.3). The letters F, T, C, P and O stands for Frontal, Temporal, Central, Parietal and Occipital. Even numbers refers to electrode positions on the right hemisphere and odd numbers refer to positions on the left. The skull is divided into the transversal and median planes with 10 % and 20 % intervals between each electrode. [Misulis and Head, 2003] Figure C.3: The electrode placement system for recording EEG signals. A) Superior view, B) left side view. C.2.2 Artifacts Among basic evaluation of the EEG traces belongs scanning for signal distortions called artifacts. Usually it is a sequence with higher amplitude and different shape in comparison to signal sequences that doesn t suffer by any large contamination. The artifact in the recorded EEG may be either patient-related or environmental. Biological artifacts are unwanted physiological signals that may significantly disturb the EEG. Environmental artifacts, such as AC power line noise, can be decreased by decreasing electrode impedance and by shorter electrode wires. [Misulis and Head, 2003] C. Electroencephalography 77