Automated Sleep Stage Analysis Using Hybrid Rule-Based and Case-Based Reasoning

Size: px
Start display at page:

Download "Automated Sleep Stage Analysis Using Hybrid Rule-Based and Case-Based Reasoning"

Transcription

1 Automated Sleep Stage Analysis Using Hybrid Rule-Based and Case-Based Reasoning ( )

2 Automated Sleep Stage Analysis Using Hybrid Rule-Based and Case-Based Reasoning

3 Ph. D. Dissertation Automated Sleep Stage Analysis Using Hybrid Rule-Based and Case-Based Reasoning ( ) Haejeong Park August 2000 Interdisciplinary Program of Medical and Biological Engineering Major Seoul National University

4 Automated Sleep Stage Analysis Using Hybrid Rule-Based and Case-Based Reasoning Presented to the Graduate School of Seoul National University in Partial Fulfillment of the Requirements for The Degree of Doctor of Philosophy by Haejeong Park Interdisciplinary Program of Medical and Biological Engineering Major Seoul National University This dissertation is approved for The Degree of Doctor of Philosophy June 2000 Chairman Vice Chairman Member Member Member

5 Automated Sleep Stage Analysis Using Hybrid Rule-Based and Case-Based Reasoning Doctoral Dissertation submitted in June of 2000 to the Graduate School of Seoul National University in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Interdisciplinary Program of Medical and Biological Engineering Major by Haejeong Park Interdisciplinary Program of Medical and Biological Engineering Major Seoul National University

6 For from Him and through Him and to Him are all things. To Him be the glory forever! Amen. Romans 11:36

7 i Abstract The objective of this study is to enhance the reliability of the automated sleep staging using context-based classification, and to make a system capable of learning and explaining to the classification. To achieve these goals, we propose an automated method for sleep staging using hybrid rule- and case-based reasoning. The system first performs rule-based sleep stage scoring according to the Rechtschaffen and Kale s sleep-scoring rule (1968) and then supplements the scoring with case-based reasoning. This method is comprised of a patient calibration unit, a signal processing unit, a rulebased scoring unit, and a case-based scoring unit. The patient calibration unit determines the basic threshold sets in order to adapt to various subjects. The signal processing unit rejects artifacts, extracts features and detects events from polysomnographic signals. The rule-based scoring unit applies basic standard rules to the facts derived from signal processing unit. If the rule-based scoring unit fails to reason above the confidence threshold, the case-based scoring unit executes additional scoring. We applied this methodology to three recordings of normal sleep and three recordings of obstructive sleep apnea (OSA). The average agreement rate in normal recordings was 87.5 % and case-based scoring enhanced the agreement rate by 5.6 %. This architecture showed several advantages over the other analytical approaches in sleep scoring: high performance on sleep disordered recordings, the explanation facility, and the learning ability. The results suggest that this combination of rule-based reasoning and case-based reasoning is promising for automated sleep scoring and it is also considered to be a good model of the cognitive scoring process. Keywords: sleep staging, case-based reasoning, rule-based reasoning, Student-number: automatic analysis, hybrid reasoning

8 ii Table of Contents Abstract....i Table of Contents...ii List of Figures..vi List of Tables. viii Nomenclature....x Chapter 1. INTRODUCTION Automation of Sleep Staging: Background Sleep diagnosis and polysomnography Sleep stage scoring standard: R & K criteria Necessity for automation of sleep diagnosis Researches on Automated Sleep Staging Research on automated sleep staging Discussion Statement of the Problem Related Researches Objective and Overview of the Study..14 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS Research-oriented Analyzing Software Design Strategy and Overview Main functions Extended functions Visualization.. 25

9 iii 2.2. Digital Polysomnograph Hardware Design Strategy and Overview Hardware Specifications Visualization and Discussion...33 Chapter 3. HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING Introduction Rule-based Reasoning Case-based reasoning Hybrid Rule-based and Case-based Reasoning...45 Chapter 4. DATA PREPARATION Introduction Subject Specific Calibration Alpha wave frequency range Tone of chin EMG Artifact Rejection Removing ECG interference to EEG by AR interpolation Removing harmonic noises Removing drift caused by sweat Feature Extraction from Background signal activities Wave segment of EEG LPC of EEG State of EOG Tone of chin EMG Waveform Detection Introduction Sleep Spindle 60

10 iv K-complex Slow delta wave Arousals REMs Reliability value...66 Chapter 5. RULE-BASED STAGING Introduction Event verification Single epoch reasoning Multi-epoch adjusting Reliability value Final decision Implementation Discussion...80 Chapter 6. CASE-BASED STAGING Introduction Case representation Case Retrieval Indexing problem Similarity Symmetry Rules for managing case retrieval Case Revise and Retain policy Establishing case-base Discussion...92

11 v Chapter 7. RESULT Performance Evaluation Evaluation I: performance at full epochs Evaluation II: performance at coincident epochs Benchmark test for stages Comparison with other approaches: Hybrid Rule-Based Reasoning and Neural Network Learning facility and Explanation facility Learning facility by case accumulation Explanation of decision Chapter 8. DISCUSSION.114 Chapter 9. CONCLUSION..117 REFERENCES..119 Appendix A Appendix B (Korean Abstract)..139 ACKNOWLEDGEMENT..141 BIOGRAPHICAL SKETCH..144

12 vi List of Figures Figure 1-1. Block diagram of digital polysomnographic system. 3 Figure 1-2. A Display of Stage REM 7 Figure 1-3. A Display of Stage W 7 Figure 1-4. A Display of Stage 1 7 Figure 1-5. A Display of Stage 2 8 Figure 1-6. A Display of Stage 3 8 Figure 1-7. A Display of Stage 4 8 Figure 1-8. A Display of MT 9 Figure 1-9. General framework of sleep classification 10 Figure 2-1. Overall architecture of Intelligent PolySomnographic System (IPSS) 18 Figure 2-2. Software functional architecture with modular design 20 Figure 2-3. A Display of Main Software 26 Figure 2-4. A Display of Automated Analysis 27 Figure 2-5. Time Synchroniztion Example. 30 Figure 2-6. Amplifier and Acquisition Hardware 31 Figure 2-7. Display of IPSS Hardware 33 Figure 3-1. Structure of intelligent system 35 Figure 3-2. Architecture of the automated sleep scoring engine 38 Figure 3-3. A general structure of a rule-based expert system 40 Figure 3-4. A general architecture of a case-based reasoning system 43 Figure 4-1. An example of determining the alpha frequency band 50

13 vii Figure 4-2. An exemplary display of determining the threshold of chin EMG tone 51 Figure 4-3. Elimination of ECG artifacts in EEG using AR interpolation method 53 Figure 4-4. Adaptive notch filtering using LMS algorithm 54 Figure 4-5. Wave Segment Index of EEG 56 Figure 4-6. Classification of EOG state 58 Figure 4-7. An example of the relationship between Hypnogram and EOG state at normal sleep 59 Figure 4-8. Detecting sleep spindle using STFT 61 Figure 4-9. Parameters for detecting K-complex. 62 Figure An exemplary display of the automated detection of K-complex 62 Figure Slow delta wave detection using Fujimori algorithm 63 Figure The detection procedure of micro arousal using band filtered signals 64 Figure REMs detection procedure using Fuzzy inference 65 Figure An exemplary display of REMs detection 66 Figure 5-1. The Flow Diagram of RBSU 69 Figure 5-2. Architecture of RBSU based on CLIPS 80 Figure 6-1. Diagram of data flow to the CBSU and the scoring process in the CBSU 83 Figure 7-1. An illustrative example of normal sleep hypnogram. 96 Figure 7-2. Neural network architecture 108 Figure 7-3. Reliability and case-base growth 111 Figure 7-4. Distribution of stages at the case-base. 111

14 viii List of Tables Table 1-1. Outline of Sleep Scoring Criteria According to Standard Criteria [1] 6 Table 3-1. Comparison of Conventional programming Systems and RBR Systems 41 Table 3-2. The comparison of Rule Based Reasoning and Case based Reasoning 46 Table 6-1.The problem fields of the case used in CBSU 85 Table 6-2. List of similarity functions and weights of fields 86 Table 7-1. Scoring agreement of a normal recording between manual scoring and RBSU 97 Table 7-2. Scoring agreement of a normal recording between manual scoring and hybrid R+CBSU 98 Table 7-3. Scoring agreement of all recordings (3523 epochs from four subjects) between manual scoring and RBSU 99 Table 7-4. Scoring agreement of all recordings (3353 epochs from four subjects) between manual scoring and RBSU(C) 100 Table 7-5. Scoring agreement of all recordings between manual scoring and CBSU 101 Table 7-6. Scoring agreement of all recordings between manual scoring and the hybrid R+CBSU 102 Table 7-7. Scoring agreement rate between manual scoring and automated algorithm 102

15 ix Table 7-8. Scoring agreement of typical recordings between manual scoring and the hybrid R+CBSU. 105 Table 7-9. Scoring agreement of typical recordings between manual scoring and the hybrid R+CBSU with the aid of computer scoring 105 Table Benchmark test of the Hybrid R+CBSU 107 Table The agreement table between the manual scoring and Hybrid neuralnetwork and Rule-based expert system 109 Table Agreement rate of reasoning techniques 109 Table Performance of retention 112 Table An example of decision explained by RBSU 113

16 x Nomenclature CBR : Case-Based Reasoning CBSU : Case-Based Scoring Unit CLIPS : C Language Integrated Production System ECG : Electrocardiogram EEG : Electroencephalogram EMG : Electromyogram EOG : Electrooculogram IPSS : Intelligent PolySomnographic System LPC : Liner Predictive Coding MT : Movement Time NPT : Nocturnal Penile Tumescence NREM : Non Rapid Eye Movement OSA : Obstructive Sleep Apnea PCU : Patient Calibration Unit R & K : Rechtschaffen and Kale RBR : Rule-Based Reasoning RBSU : Rule-Based Scoring Unit REM : Rapid Eye Movement SaO2 : Oxygen Saturation SEM : Slow Eye Movement SPU : Signal Processing Unit

17 Chapter 1 INTRODUCTION In vain you rise early and stay up late, toiling for food to eat for LORD grants sleep to those he loves. Psalms 127: Automation of Sleep Staging: Background Sleep diagnosis and polysomnography Sleep has come to the surface as a major health issue in recent decades and has attracted considerable attention in the traditional medical area recently. Changes in life style from threatened schedules, excessive work and new technological situation has caused complex effects on sleep quality. Difficulties in sleep raised problems ranging from diminished functionality to medical disorder. The International Classification of Sleep Disorders (1990) lists 84 sleep disorders [1]. According to epidemiological studies on sleep disorders, a large population (In USA, 30 % for insomnia, 1% for OSA) suffers from sleep related disorders [1]. This situation is expected to become worse. Therefore, current phase of sleep is not only and individual problem but also a social problem. Sleep is a typical rhythm that is regulated by the circadian process. It has indwelling patterns of body state that are divided into rapid eye movement (REM) sleep

18 Chapter 1. INTRODUCTION 2 and nonrapid eye movement (NREM) sleep. NREM sleep is further subdivided into Stages 1-4 [1][2]. Stage 1 and Stage 2 are light sleep; Stage 3 and Stage 4 are deep sleep, or generally termed slow wave sleep. Specifically, REM and NREM sleep alternate every 90 minutes. In the first part of the sleep, there is more NREM sleep, especially deep sleep, and REM sleep predominates in the second half of the sleep. This dynamic structure of sleep can be examined by polysomnography. Long-term changes of brain state are projected onto the outward physiological signals such as electroencephalogram (EEG), electrooculogram (EOG), electromyogram (EMG), electrocardiogram (ECG), blood pressure, respiratory efforts, oxygen saturation (SaO2) and so on. Polysomnography records all these signals in order to diagnose sleep structure and sleep disorders. Especially, recording sleep EEG, EOG and EMG is fundamental for sleep stage scoring (in brief, sleep staging or sleep scoring). According to the Rechtschaffen and Kales suggestion [2][3], central leads (C3-A2 or C4-A1) of EEG are the minimal requirement for sleep staging. Occipital leads (O2-A1 or O1-A2) are additionally recorded. EOG leads are placed at the outer corners of the eyes. To detect vertical as well as horizontal eye movements, one electrode is placed slightly above and one slightly below the eyes. Three EMG leads are placed in the mental and submental areas and the voltage between two of these three is monitored. Sleep stage scoring gives information on the quality of sleep based on the sleep structure [3]. In addition to the examination of sleep structure, an important area in sleep diagnosis is the examination of the sleep disorders by measuring sleep-related events such as obstructive apnea, snoring and periodic leg movements that interrupt normal sleep. Oral and nasal airflow signals, abdominal and thoracic movement for respiration are used for diagnosing respiratory disorders such as sleep apnea and hypopnea.

19 Chapter 1. INTRODUCTION 3 BED 1 Infrared Camera EEG, EOG Digital Video Recorder Main PC Oro-nasal Airflow, Snoring, EMG ECG, Thoracic respiration Polysomno Amplifier A/D & DSP Abdominal respiration, NPT Oximetry EMG, Body Position CD-ROM Remote PC (Windows NT) LAN (TCP/IP) Figure 1-1. Block diagram of digital polysomnographic system. Sixteen channels of physiological signals are digitally converted and managed by IPSS (Intelligent PolySomnographic System) Recording snoring sound and oxygen saturation also aid diagnosis of respiration. ECG is used for circulatory disorders and EMG is used for behavior disorders. Nocturnal penile tumescence (NPT) test is an additional area that polysomnography supports. Video monitoring during sleep is also a basic requirement in polysomnography. Even to the diagnosis of sleep disorders caused by sleep events, sleep stage scoring is essential. Without sleep staging, sleep diseases cannot be diagnosed completely. Figure 1-1 shows a block diagram of the digitalized polysomnographic system. This system includes sensor part, amplifier part, digitalization part and managing part Sleep stage scoring standard: R & K criteria Sleep is not a uniform biological state. It consists of rapid eye movement (REM) sleep and Stage 1, Stage 2, Stage 3 and Stage 4 in nonrapid eye movement (NREM) sleep according to the current sleep staging standard proposed by Rechtschaffen and

20 Chapter 1. INTRODUCTION 4 Kales (in brief, R & K criteria) [2]. Generally, sleep stage classification also includes wakefulness (Stage W) and movement time (MT). Time is divided into epochs (commonly 30 seconds for each epoch). The sleep stage assigned to each epoch is the stage occupying the majority of the time within that epoch. Sleep stage scoring depends on the characteristics of electrophysiological signals: EEG, EOG, and chin EMG. Stage W (wakefulness) is characterized by EEG alpha activities that appear in more than 50 % of the epoch. Stage 1 is characterized by the low voltage mixed frequency EEG without rapid eye movements. Stage 1 is scored when less than 50% of an epoch contains alpha waves and the criteria for deeper stages of sleep are not met. Slow rolling eye movements often are present and the level of muscle tone is equal or diminished compared to that during wakefulness. Stage 2 is defined by the presence of sleep spindles (oscillations of Hz with a duration of seconds) and/or K- complexes (a high-amplitude biphasic wave of at least 0.5 second duration, an initial sharp positive voltage followed by a negative deflection slow wave). Stage 3 and Stage 4 are characterized by slow delta waves of EEG. Slow delta wave is defined as EEG activity slower than 2 Hz with peak-to-peak amplitude greater than 75 microvolt. If slow delta waves occupy 20-50% of an epoch, the epoch is scored as Stage 3, and if they occupy more than 50%, it is scored as Stage 4. Stage REM is defined by a relatively low voltage, mixed frequency EEG in conjunction with episodic rapid eye movements (REMs) with low amplitude EMG tone. The standard sleep staging criteria in adults, according to the three electrographic parameters, are outlined in Table 1-1 [1]. Besides these characteristic definitions of each sleep stage, Rechtschaffen and Kales criteria propose several smoothing rules in order to score obscure epochs that cannot be dealt with by the above definitions [2]. Smoothing rules are based on the

21 Chapter 1. INTRODUCTION 5 assumption that a sleep state has the tendency to persist throughout epochs. These rules refer to the electrophysiological context between epochs prior to and posterior to the target epoch. For example, if there is no spindle, K-complex, and no other events in the target epoch, then smoothing rules propose to look for events at prior and posterior epochs and if spindles are found, then the epoch is scored as Stage 2. This kind of contextual information is essential in practical scoring. Exemplary displays of one-epoch polysomnographic signals at seven sleep stages (Stage REM, Stage W, Stage 1, Stage 2, Stage 3, Stage 4, and MT) are shown from Figure 1-2 to Figure 1-8. Each epoch of 30 seconds including EEG (C3/A2 and O2/A1), EOG (left and right) and chin EMG of a young adult male volunteer is recorded on a Grass Instruments Company Model 78 polygraph at the Sleep Research Center of Seoul National University Hospital. The central and occipital EEGs, left and right EOGs and chin EMG are recorded with a low-frequency cutoff of 0.3 Hz and a high-frequency cutoff of 100 Hz.

22 Chapter 1. INTRODUCTION 6 Table 1-1. Outline of Sleep Scoring Criteria According to Standard Criteria [1] Stage/State EEG EOG EMG Relaxed Wakefulness NREM Stage 1 Stage2 Eyes Closed: rhythmic alpha (8-13 cps); prominent in occipital; attenuates with attention Eyes open: relatively low voltage, mixed frequency Relatively low voltage, mixed frequency May be theta (3-7 cps) activity with greater amplitude Vertex sharp waves Synchronous high-voltage theta bursts in children Background: relatively low voltage, mixed frequency Sleep spindles: waxing, waning, cps (>0.5sec) K complex: negative sharp wave followed immediately by slower positive component (> 0.5sec); spindles may ride on Ks; Ks maximal in vertex; spontaneous or in response to sound Stage3 >=20% <=50 % high amplitude (> 75 uv), slow frequency (< 2cps); maximal in frontal Voluntary control; REMs or none; blinks; SEMs when drowsy SEMs Occasionally SEMs near sleep onset None, picks up EEG Stage4 > 50% high amplitude, slow frequency None, picks up EEG Tonic activity, relatively high; voluntary movement Tonic activity, may be slight decrease from waking Tonic activity, low level Tonic activity, low level Tonic activity, low level REM Relatively low voltage, mixed frequency Phasic REMs Tonic suppression; phasic twitches Movement time Obscured Obscured Very high activity Adopted from Principles and Practice of Sleep Medicine, 2nd Ed. W.B. SAUNDERS company, which is modified from Rechtschaffen A, Kales A (eds): A Manual of Standardized Terminology: Techniques and Scoring System for Sleep Stages of Human Subjects. Los Angeles, UCLA Brain Information Service/Brain Research Institute, 1968.

23 Chapter 1. INTRODUCTION 7 EEG:C3/A2 EEG:O2/A1 100 uv 0 1 sec EOG:Left EOG:Right Rapid Eye Movements EMG:Chin Figure 1-2. A Display of Stage REM: REM sleep is scored when the EEG pattern is relatively low voltage, mixed frequency, the EMG is tonically suppressed and the EOG contains rapid eye movements. EEG:C3/A2 EEG:O2/A1 100 uv 1 sec 0 EOG:Left EOG:Right EMG:Chin Figure 1-3. A Display of Stage W: Rhythmic alpha waves are prominent in central and occipital. This epoch is scored as Wake. EEG:C3/A2 EEG:O2/A1 EOG:Left EOG:Right EMG:Chin Vertex Sharp Wave Slow Eye Movement 100 uv 0 1 sec Figure 1-4. A Display of Stage 1: The attenuation of waking EEG alpha and the presence of slow eye movement are indicators of Stage1. Vertex sharp waves are a common feature of the onset of Stage 1.

24 Chapter 1. INTRODUCTION 8 EEG:C3/A2 K-COMPLEX SPINDLE 100 uv EEG:O2/A1 0 1 sec EOG:Left EOG:Right EMG:Chin Figure 1-5. A Display of Stage 2: Sleep spindles (oscillating wave with the frequency between Hz) and K-complexes (high voltage, sharp rising and sharp falling wave) with relatively low voltage mixed frequency background EEG are found. This epoch is scored as Stage 2. EEG:C3/A2 EEG:O2/A1 100 uv 0 1 sec EOG:Left EOG:Right EMG:Chin Figure 1-6. A Display of Stage 3: Stage 3 is scored when the EEG pattern consists of high-voltage (>=75uV), slow (<=2 Hz) activity in 20 % or more, but less than 50 % of an epoch. EEG:C3/A2 EEG:O2/A1 100 uv 0 1 sec EOG:Left EOG:Right EMG:Chin Figure 1-7. A Display of Stage 4: Stage4 is characterized by a predominance ( > 50 %) of high-voltage, slow waves in the EEG.

25 Chapter 1. INTRODUCTION 9 EEG:C3/A2 EEG:O2/A1 100 uv 0 1 sec EOG:Left EOG:Right EMG:Chin Figure 1-8. A Display of MT: Movement time is scored when movements arise from sleep, immediately precede sleep, and obscure the EEG activity for at least one-half the scoring epoch Necessity for automation of sleep diagnosis Computerization or automation of sleep research has been carried out on the two basic desires to satisfy two basic goals [4]. The first goal is to discover new electrophysiological patterns not revealed by visual inspection of EEG and other signals. The second goal is to save costs and time by imitating human sleep scoring. In the traditional method of sleep stage scoring, a sleep specialist reads a paper polysomnographic recording of the whole night s sleep and assigns sleep stages to each epoch according to R & K criteria. However, the quantity of electrophysiological signals is so enormous that it takes about 3-4 hours to score one subject manually and requires a large amount of labor. This keeps the method from widespread usage and hinders the advancement of sleep research. A reliability problem of manual sleep staging exists, with a relatively poor interscorer agreement rate, especially in the disturbed sleep scorings. These factors make the automation of sleep research essential not only as a laborsaving method but also as a consistent quantitative tool.

26 Chapter 1. INTRODUCTION Researches on Automated Sleep Staging Previous attempts to automated analysis of sleep were mainly directed towards imitating the R & K rules in order to save scoring time and further objectify the procedure [4]. Numerous attempts of automated sleep staging share common features to achieve common goal. Sleep stage scoring is a kind of classification problem and follows the general classification framework. General classification can be characterized by the information reduction and information mapping process. Kil and Shin [5] summarize this approach to classification as 1) preprocessing to remove as many confusion factors as possible and data projection onto appropriate spaces in which signal attributes can be better captured and be less sensitive to extraneous variables, such as interference and environmental noise, 2) extraction of many pertinent features and optimization and 3) construction of appropriate decision statistics or classifier topologies that take advantage of the underlying good feature distribution. This framework is shown in Figure 1-9. Possible algorithm feedback or interaction Raw data ( EEG, EOG, EMG ) Preprocessing and Enhancement Feature Extraction Classifier Stage Figure 1-9. General framework of sleep classification

27 Chapter 1. INTRODUCTION Research on automated sleep staging Most research on automated sleep stage scoring follows the framework of Figure 1-9. The differences are in the selection of feature sets and the classification method. Feature sets extracted from the electrophysiological signals such as EEG, EOG, and EMG, can be categorized as those of the time domain and those of the frequency domain. A basic feature extraction algorithm of the time domain is period analysis (either zero-crossing or peak analysis) frequently used with band-pass filtering [4][6]. Recently, several non-linearity indicating features such as the bicoherence index [7] and the correlation dimension of chaos theory have been tried [8]. The most representative feature extraction methods in the frequency domain are the FFT-based spectral analysis. Delta (0-4Hz), theta (4-8Hz), alpha (8-12 Hz), spindle (12-16Hz) and beta (16-30 Hz) band frequency powers are commonly used as features. Also, model based features such as autoregressive (AR) coefficients have been frequently used. These features are used with a single set or with mixed sets. In EOG signals, the most commonly used feature is the correlation coefficient. In the EMG signal, the variance of the signal is often used as a feature. Another categorization of the previous researches is based on the classification method. The statistical approach, the neural network approach and the syntactic approach are three major approaches. The statistical approach assumes physiological features on the statistical base for classification. Features are generated by a statistical nature and therefore the underlying system can be modeled or assumed on the probability theory [9]. Neural network approaches have recently received considerable attention [10-14]. This classification is performed by the inter-relationship between features and can be applied to new

28 Chapter 1. INTRODUCTION 12 situations by adapting weights on the connections. The syntactic approach deals with grammars or rules [15-20]. They require parsing and basically intuitive structures. The rule-based expert approach is the representative of this approach [15]. Recently, hybrid classification approaches have been actively researched [21][22]. Some researchers applied fuzzy inference [23] or hybrid neural network and expert system [22]. It should be noted that sleep staging is also based on the existence of characteristic waveforms such as sleep spindle, K-complex, and rapid eye movement with background activities of electrophysiological signals. To detect these waveforms has also been a widely researched area [24-30] Discussion No approach has shown significantly better performance than the other approaches. According to the published results, the agreement rate between automated scoring and manual scoring is % in recordings of normal young adults [31-35]. In sleepdisordered cases, where sleep signals are more irregular with various artifacts and innate complexity of the disturbed sleep, the agreement rate is reported somewhat lower at a level of % [4][36-37]. We believed that the unsatisfactory agreement rates of these algorithms, especially in the sleep-disordered cases, primarily resulted from relatively limited usage of the contextual information Statement of the Problem Even though numerous efforts have been made for automation, very few methods

29 Chapter 1. INTRODUCTION 13 are routinely used. In clinical practice, most sleep laboratories do not use automated analysis in place of manual sleep staging. There is an obvious reason: low man/machine agreement in sleep stage scoring. The primary reasons for the low reliability of the current systems, especially in the sleep-disordered cases, can be summarized as follows: (1) inter-group variability of sleep, which means different sleep structures depending on the particular sleep disorder; (2) inter-subject variability of sleep; (3) the existence of many kinds of artifacts in longterm monitoring; (4) rarely used contextual information; and (5) the intrinsic problems in the basic concept of the current R & K sleep staging system. In spite of the many arguments that have recently arisen on the last fifth reason [4][38], we decided to mimic the current R & K system in this study and therefore we ruled out considering this reason in our study. Another problem in clinical usage is that human specialists would not know the automated scoring process, and they would therefore receive no explanation of the decision. In other words, previous algorithms are not heuristic enough directly. This is another reason why a number of automated systems fail to receive confidence from specialists. Except for the neural network approach, the knowledge acquired from experience cannot be added to each new scoring process. Even in the neural network approach, knowledge is acquired by the developer, not by the practical users. In summary, low reliability from less usage of contextual information, no explanation facility and no learning facility are problems of this study.

30 Chapter 1. INTRODUCTION Related Researches To overcome low reliability, contextual information is considered important under the R & K system, where sleep staging is largely based on the signal context. Several rule-based expert systems or hybrid expert systems have been developed in order to include the contextual information [15-17,19-21]. Several neural network systems with context-containing features have been researched recently [14][39]. But basically, they have limitations in using context and no support for the explanation facility and learning facility. Rule based reasoning has been a leading method of the expert system that can contain human knowledge including context [40]. This system supports an explanation of the decision. Even though this method is widely used as part or all of the commercial systems, rule-based reasoning seems to have less progress in the performance. Recently, case based reasoning has been actively researched as a new artificial intelligence concept for classification, decision and planning [41-43]. This method has the merit of learning by experience and dealing with domain knowledge totally. Besides, integrating different approaches, so called, hybrid reasoning approaches is a general trend in artificial intelligence. Especially, hybrid case based reasoning and rule based reasoning have received much attention in many application areas [44-48]. However, there has been no introduction of a case-based reasoning concept or a casebased hybrid reasoning technology into sleep stage analysis Objective and Overview of the Study The objective of this study is to enhance the reliability of automated sleep staging

31 Chapter 1. INTRODUCTION 15 using context-based classification, and to make a system capable of learning and explaining to the classification. To achieve these goals, we analyzed the human approach to sleep staging. First, human experts follow the standard rules of Rechtschaffen and Kales. Also, they often use heuristics and memories from past cases if they cannot apply rules directly. These two complementary aspects motivated us to integrate rule-based reasoning (RBR) and case-based reasoning (CBR). In this paper, we suggest the architecture of the hybrid sleep staging system, especially focusing on the efficiency of CBR with RBR in the sleep diagnosis. Four important parts compose this dissertation. The first part describes a researchoriented tool for sleep analysis in order to support further research. The second part describes data preparation methods including patient calibration, artifact rejection, feature extraction and waveform detection. The third part describes rule-based reasoning as the first step of the hybrid reasoning process for sleep staging. The fourth part describes the case-based reasoning method applied to sleep staging. In the fifth part, the experiment and discussion will be described. Each of these parts is described in a separate chapter. In this chapter, the necessity of automated sleep staging has been described. This chapter has also reviewed automated sleep staging approaches and discussed problems in the current automated system, and has suggested the direction of the thesis. In Chapter 2, a research-oriented system for sleep analysis named IPSS is introduced. The first part describes the strategy and implementation of the software system and the second part describes those of hardware system. Chapter 3 describes overall the architecture of the system especially focusing on the classifier with artificial intelligence. In this chapter, rule-based reasoning and case-

32 Chapter 1. INTRODUCTION 16 based reasoning are summarized. The hybrid reasoning approach using rule-based reasoning and case-based reasoning is also introduced. Chapter 4 describes the data preparation methodology for the decision step. Patient calibration, artifact rejection, feature extraction and waveform detection methods are illustrated. Chapter 5 describes the process of applying rule-based reasoning to sleep stage scoring. Event verification, single epoch reasoning and multi-epoch reasoning are described. This chapter introduces the concept of reliability value and describes the method of final decision. Chapter 6 describes case-based sleep staging method. Case representation, case retrieval and case maintenance are described. Further optimization area of case-based reasoning is also discussed. Chapter 7 describes the result of applying the hybrid reasoning system to sleep stage scoring. Performance is evaluated in terms of the objectives of the system. Chapter 8 summarizes the main idea of this dissertation, discusses about the problems of the system and suggests direction in which this work can be extended. In Chapter 9, contributions of this thesis are summarized with final conclusion.

33 Chapter 2 IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS In this chapter, a brief overview of a digitalized polysomnographic system and interactive analysis software for sleep research is presented. Various modalities and the large amount of sleep signals make it difficult to access sleep physiology directly. Several commercial systems provide analysis tools in order to deal with sleep data. However, most of them are limited to basic algorithms and are not efficient for further research, because they mainly focus on clinical usage. For both research and clinical purposes, we developed an intelligent polysomnographic system (IPSS) which is comprised of both software and hardware components. Software parts of IPSS are categorized into basic functions: user interface, automatic analysis and external interface. The software of IPSS was designed with the object-oriented programming method. Hardware comprises sensor part, amplifier part, analog to digital part, digital signal processing part and host interface part. Hardware system (named DSP system of IPSS) was designed with a modular architecture that is flexible for various usages. Figure 2-1 summarizes the overall structure of IPSS. In the first section of this chapter, the research-oriented software is presented, and in the second section, the digital polysomnographic hardware is described with an explanation of the strategy of its design.

34 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS 18 Intelligent Intelligent PolySomnographic PolySomnographic System System (IPSS) (IPSS) Software Clinical Verification Hardware Moving Image Recording Basic Functions User Interface Automatic Analysis External Interface USB, RS-232 for interfacing DSP communication A/D Amplifier control Sensor Amplifier OOP with C++ on Windows Software Oriented Modular Design Figure 2-1. Overall architecture of Intelligent PolySomnographic System (IPSS)

35 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS Research-oriented Analyzing Software A general purpose multi channel biosignal managing and analyzing software is presented in this section Design Strategy and Overview Design strategy is make IPSS software 1) modular; 2) useful for clinical usage; 3) appropriate in research area and 4) user-friendly and user-interactive. 1) Modular design IPSS software contains biosignal acquisition, biosignal storage management, biosignal display and biosignal edit and analysis module. IPSS also has video recording, reporting and managing database additionally. Biosignal acquisition method depends on the type of biosignal or the protocol of the signal. Signals may be acquired either from the direct hardware on the system or from the remote system through RS232 protocol - RS232 is a common protocol for general commercial system such as oximetry and blood pressure - or TCP/IP protocol. In order to interface with various types of signal protocol, modular design is necessary. The method of storage, display and analysis of the biosignal may differ according to the purpose of signal acquisition and these areas also require modular design. To deal with various types of signals and to satisfy modular concept, object oriented programming (OOP) is used for real implementation. The functional architecture of modular design is presented in Figure 2-2.

36 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS 20 Signal Display Signal Acquisition Signal Analysis File Manage Network Interface Coordinating Manager of IPSS Event Edit Video Record Analysis Report MATLAB Interface Patient Database Figure 2-2. Software functional architecture with modular design 2) Usefulness in clinical usage The minimum requirement of medical diagnosis system is usefulness in clinical usage. Several polysomnographic softwares are well-designed in an engineering sense, but poor for clinical usage. The target for clinically useful design is to retain merits of paper reading and to overcome limitations of paper reading. This is because sleep scoring has been standardized on the paper reading and many somnologists are used to

37 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS 21 papers. Paper reading has the merit of rather high temporal and amplitude resolution. With paper reading, it is easy to review the whole night s sleep trend and to access random epochs quickly. But, it cannot display signals with variable temporal resolution. Extensions such as scaling, position change and filtering are not allowed. The clinical feedback has guided this purpose. 3) Appropriateness in research area Another important aim of IPSS is to make an aided system for research. Automation or semi-automation is required to replace or help the specialist. In order to develop automated algorithms, the process of trial and error is unavoidable. Therefore, programming with C is not adequate because it requires much time and labor in coding and debugging. In addition, a fixed executable program does not satisfy users who may want to implement their own algorithms. In order to achieve this goal, we decided to interface our system with the widely used research tool MATLAB (Mathworks, USA). 4) User-friendly and user-interactive design A final consideration in design is supporting user-friendly and user-interactive interfaces. In order to help manual scoring on the digital system, speed and simplicity is the most important requirement in clinical usage. To satisfy this requirement, analysis of the work of sleep research center is performed Main functions

38 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS 22 1) Biosignal Acquisition Combination with the previously set-up analog polysomnograph is the first target of IPSS. Technique to stabilize data acquisition for the long time and real-time analysis is developed. Amplifier calibration, biosignal calibration, and event logging functions are implemented. 2) Biosignal Display IPSS displays biosignals with high screen resolution and with high speed. Sleep recording is different from other EEG recordings in that sleep recording follows a fixed chart speed, 1 cm/1 sec, in order to display one 30 second epoch signals on a page. Without any scrolling, the display system should display a 30 second signal on one screen shot. IPSS is appropriate for the high-resolution display (above 1600x1200) that is required to display EEG with clinically meaningful signal resolution. For example, the spatial resolution of 1500 pixels for a 30 second epoch is 50 pixels/sec, which is a slightly lower than the Nyquist frequency of EEG (60 Hz, assuming that EEG is distributed below 30 Hz frequency). This system also supports multiple display modes that are 60, 120 and 240 seconds/epoch. Long duration display modes (120 or 240 seconds/epoch) are preferable in analyzing respiratory activity. The other difference of IPSS from general EEG monitoring is that IPSS is designed to effectively display different signals with different sampling frequencies on a single screen. In practical sleep scoring, the page-scrolling speed is essential. In order to score an obscure epoch, scorers repeatedly check previous and posterior epochs. Therefore, we enhanced the display speed by using a file and memory management technique and optimizing the plotting process.

39 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS 23 Easy control of the display properties such as gain, color, baseline, and display type is another user-friendly feature of IPSS. 3) Manual Sleep Staging and Waveform Marking For semi-automatic or manual scoring purpose, manual sleep staging tool and waveform (or event) marking tool are implemented. Because the manual marking associated with paper reading takes so much time, IPSS automatically calculates the information of the marked waveforms. Automatically marked waveforms can be revised. Waveform searching and direct access to waveforms are also available. A summarization tool calculates and displays overall sleep information from marked waveforms. To evaluate the agreement rate between human scoring and automated scoring, and between human scorers, IPSS supports multiple scoring profile system. 4) Interface with Automated Analysis Automated diagnosis of sleep including sleep staging is a major focus of IPSS. A large quantity of sleep signals makes it difficult for analysis. Therefore, analyzing sleep requires an interactive tool that makes it possible to access data and to apply the user s own function. Developing an automated system with a C code is time consuming and yields a result that is not fit for end users. MATLAB is a widely used powerful tool that can be easily programmed for research purposes. We implemented a complete interface with MATLAB. After verification of the algorithm with MATLAB, we coded algorithms into C++. With the MATLAB, the user can send and get data with IPSS software and can

40 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS 24 manage IPSS software easily by the DDE (dynamic data exchange) method. IPSS software can access MATLAB algorithms without any interface. 5) Analysis Report Using the result of scoring, a basic summary of the diagnosis is generated including the sleep hypnogram, sleep indexes, and the event lists. Documentation of the result of diagnosis provides clinically meaningful information such as sleep latency, the total REM epoch, hypnogram, etc. on both Microsoft Excel and Microsoft Word Extended functions 1) Digital Video Recording The importance of video monitoring during polysomnographic recording is increasing, but the analog-taped storage method currently in use has difficulties in random access and synchronization with the polysomnographic data. To overcome these limitations, we need to develop a digitalized integration and synchronization method. Digitization of the moving image during sleep requires a minimum level of detecting relatively fast movements. It also requires a minimal volume of data for storage and communication. We analyzed several standards of digital moving image compression to find the optimal storing format for the video image during sleep. Based on this analysis, we developed an integration method using the format H. 261, which has better performance than MPEG-1 at the low bit rate of 128 kbps. We also used the QCIF format in H.261 to increase the frame rate, and added the time index to the H.261 format to support random access. We synchronized these video images with the

41 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS 25 polysomnographic data [49]. 2) Patient Database, File Managing Unit, Network Interfacing Unit and Other User Interface Easy access to the database of the patient and diagnosis is implemented in order to manage the total information effectively. The file managing tool manages all files generated during long-term recording. The network interfacing unit is implemented to support remote recording. User-friendly interfaces such as a remote controller and foot switch are also implemented for easy sleep staging Visualization Figure 2-3 shows the main screen shot of the IPSS software. REVIEW mode of a 30 second 16 channels of polysomnogram is displayed. On the upper left position, the epoch number and the scored stage are displayed. The bottom status bar indicates the sleep stage profiles, or hypnogram. The screenshot of the automated analysis is shown in Figure 2-4.

42 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS 26 O2Sat Respiration ECG EMG EOG EEG Subject 20ms Page & Stage Scoring Tools Page Tools Marking Tools Analysis Tools Setup Toole Snoring Body Position Hypnogram Figure 2-3. A Display of Main Software Amplifier Tools

43 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS 27 Figure 2-4. A Display of Automated Analysis 2.2. Digital Polysomnograph Hardware A general purpose multi channel biosignal amplifier and acquisition hardware is presented in this section Design Strategy and Overview Polysomnogram signals are divided into two modalities according to the characteristic frequency range; 1) high frequency AC-coupled signals such as EEG, ECG and EMG and 2) low frequency DC-coupled signals such as nasal and oral airflow, respiratory effort, oxygen saturation and so on. Each signal has a significantly different bandwidth and amplitude. Both modalities of polysomno signals are recorded by

44 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS 28 polysomnograph. Traditional polysomnograph has multi-channel analog amplifiers of AC-coupled type and DC-coupled type. These amplifiers include a variable gain amplifier and variable cut-off frequency low pass filters, high pass filters and notch filter. Human expert sets up the amplifier settings according to the recording purpose. Pen-writer device is also an essential part of polysomnography. However, this traditional polysomnograph uses general-purpose amplifiers and pen-writer, and this make it complex, large and expensive for widespread use in the sleep research. Recently, commercial polysomnographs tend to be smaller and digitalized by focusing specific purpose and sacrificing general usage. Those systems are designed for a fixed set-up usage and to measure specific signals. Using these systems to measure other signals of interest is not easy. In this study, we developed general-purpose, multi-channel electro-physiological amplifiers and a data acquisition system that are compact and can be easily modified for various applications by changing the software configuration. The basic strategy of this system is to maximize software-dependency and to minimize dependency on the analog hardware component. The specification areas that can be replaced by digital processing are 1) filtering and amplification gain selection, 3) time synchronization and 4) signal isolation. 1) Filtering and amplification gain selection Contrary to amplifiers for exclusive use, such as high resolution EEG and multi channel ECG, we focused on a system that deals with different frequency ranges and amplification gains. In traditional analog amplifiers, a low pass filter, a high pass filter, and 60 Hz/50

45 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS 29 Hz power-line notch filter are essential components [50]. The problem of analog amplifiers is inflexibility in changing frequency band. Certain amplifiers provide several cut-off frequency filters but this makes the system complex and expensive. If a sharp cut-off frequency response is required, then the amplifier gets more complex. Also, changing the filter cut-off frequency requires manual switching for each channel. Variation of the power line harmonic noise is another problem of analog amplifiers. The frequency of the Power line is not exactly 60 Hz /50 Hz and may vary with time, but there is no means of adapting to this variation in frequency the analog notch filter. Nonlinear phase distortion in analog filtering should not be neglected. In order to overcome the intrinsic limitations of analog filtering in multi-channel, variable purpose applications, we used digital filtering techniques using a high-speed digital signal processor. Arbitrary filters can be designed in the host PC and the coefficients of the filters are downloaded to the DSP system for real time filtering. The amplification gains of polysomno signals vary from the order of 100 to the order of In a general analog system, each channel requires a gain amplifier that can be configured by selecting switch. In this system, we used digitally programmable gain amplifiers following an analog multiplexer. This method reduces the hardware components needed. 2) Time synchronization between channels In specific applications such as brain mapping, time synchronization between channels is required. In order to achieve time synchronization, several systems use individual AD converters or sample and hold ICs for each channel. This also adds to the complexity of the system.

46 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS 30 In this system, we choose a software alternative. Using the over-sampling and decimation method, we can reduce the time delay between channels to a negligible level. For example, consider a four channel recording system with a sampling frequency of 250 Hz (1/Ts) at each channel. As in Figure 2-5 (a), the general multiplexing scheme set AD conversion rate of 250 x 4 =1000 Hz and the time delay between channels (Tf1) is 1/1000 sec. The time delay between the first and the last channel (Td1) is 3 x Tf1 = 3/1000 ms. In case of oversampling by two as shown in Figure 2-5 (b), the sampling frequency is 2000 Hz and Tf2=1/2000. Time delay Td2 is 3/2000 second and this is a half of Td1. Td1 Ts Td2 Ts C H 1 C H 1 C H 1 C H 1 C H 1 C H 1 C H 2 C H 2 C H 2 C H 2 C H 2 C H 2 C H 3 C H 3 C H 3 C H 3 C H 3 C H 3 C H 4 C H 4 C H 4 C H 4 C H 4 Tf1 (a) Tf2 (b) Figure 2-5. Time Synchroniztion Example. Time delay Td2 is shorter by a half than Td1 if it is oversampled twice. The oversampling and decimation scheme also has merit in the aliasing problem

47 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS 31 and only a simple passive filter or no filter is required for an antialiasing filter in the design. 3) Signal isolation Signal isolation between the analog and digital components is very important in the medical instrument. In a multi-channel system, the general design of the isolation is carried out by placing an analog isolation amplifier or an isolation transducer between the analog multiplexer and the A/D converter [50]. In order to overcome the limitations from nonlinear resolution of analog isolation, we decided to use the digital isolation placing an optocoupler at the serial interface of the A/D converter. Meanwhile, the isolation of power is accomplished by using a DC-to-DC converter. Figure 2-6 illustrates the amplifier and acquisition hardware block diagram. Protection Circuit x10 0 x10 x100 MUX control 1 Serial Paralle l USB PGA control 1 Electrode Box MUX MUX PGA PGA CH1 CH2 A/D DSP 102 isolation Digital Signal Procesor (filtering, data processing) PC interface (Serial, Parallel, USB port) GAU(16 ch) PGA control 2 MUX control 2 Figure 2-6. Amplifier and Acquisition Hardware

48 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS Hardware Module Specifications 1) Signal Amplification Module A high precision instrument amplifier INA 2141 (Burr-Brown, USA) is used for preamplifier. It is composed of two differential amplifier sets in one package and gain of 10 or 100 can be set. Preamplified 16 channel signals are multiplexed by two 8-channel multiplexer (MAX308) and amplified by a programmable gain amplifier PGA ) AD Conversion Module We achieved digitization using a high-speed, high-resolution A/D converter DSP102 (Burr-Brown, USA), which comprises two identical A/D modules in a single chip. This converter can sample up to 200 KHz with 16-bit resolution. The digitized data is fed into DSP processor through serial communication. 3) Digital Signal Processing Module TMS320C32 (Texas Instrument, USA) is the main processor in the IPSS hardware. It is a dedicated digital signal processing chip that allows for high speed filtering and analysis. It is also used for overall control, storing and transporting digitalized signals to the host processor. 4) Communication module with Host Computer The interfaces between the DSP and the host computer are RS-232 serial port and USB protocol (universal serial bus). Digital video image for sleep are also integrated into IPSS.

49 Chapter 2. IPSS: A RESEARCH-ORIENTED SYSTEM FOR SLEEP DIAGNOSIS Visualization and Discussion Figure 2-7 shows the IPSS hardware system. Distinction of this system is digital filtering with the DSP processor that enables effective noise reduction, linear phase preservation and variable cutoff frequency. Because of the software dependent design, conversion to other applications such as QEEG and epilepsy diagnosis is possible without changing any hardware components. IPSS Hardware Figure2-7. Display of IPSS Hardware

50 Chapter 3 HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING 3.1. Introduction Sleep stage scoring is a kind of classification problem as is mentioned in Chapter 1. In the development of the automated sleep staging system, we followed the general framework of classification: preprocessing, signal enhancement, feature extraction and classification. The core of sleep stage classification is considered to be the selection of the exact classifier. By contrast to other classification problems, sleep stage scoring is so complex including statistics, abstraction and somewhat arbitrary rules that are not verified physiologically but are accepted by clinical trials. This chapter describes the hybrid architecture of sleep classification focusing on the hybrid intelligent classifier. The fundamental principles of designing the classifier in this system are 1) it should deal with contextual information; 2) it should explain the classification result; 3) it should support easy learning. These principles are adopted from human cognitive process.

51 Chapter 3. HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING 35 Despite a lot of researches, general numerical or analytical classifiers are thought to be inadequate for these principles. We decided to focus on the intelligent-based classifier because artificial intelligence has some properties that satisfy the above three principles. According to Bezdek [52] in [51], intelligence can be subdivided into computational intelligence, artificial intelligence, and biological intelligence according to the level of complexity as shown in Figure 3-1. Human Knowledge + sensory inputs biological intelligence organic Knowledge tidbits + sensor data artificial intelligence symbolic computation computational + numeric sensor intelligence Figure 3-1. Structure of intelligent system [ Bezdek, 1994] As Medsker summarized this structure in his book [53], computational processes deal with numeric data from sensors. Artificial intelligence is based on symbol-

52 Chapter 3. HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING 36 processing techniques and augments computational processes with rules and other nonnumerical knowledge tidbits. Neural network, self-organization map, genetic algorithms are examples of computational intelligence. Rule-based reasoning and case-based reasoning are in the style of artificial intelligence. Fuzzy logic is the most representative example of integrating numeric and semantic information. According to Bezdek model, the ultimate goal of intelligence is biological intelligence. The integration of different intelligent techniques is widely applied to classification problems. The reason for this hybrid approach is that each individual technology represents various aspects of human intelligence and, by integrating them, better understanding of human cognition may be possible and the augmentation of performance is expected [53]. Among various hybrid intelligent techniques, hybrid rule-based reasoning and case-based reasoning is regarded as a good model that satisfies the above three principles in the sleep classification problem. It is similar to human cognitive process in that in case of typical information, rules stored in memory are applied to score the sleep epoch and in case of ambiguous information, previous experiences are accessed from memory to find a solution. In order to implement this hybrid classifier, we designed a four-unit architecture including a Patient Calibration Unit (PCU), a Signal Processing Unit (SPU), a Rule- Based Scoring Unit (RBSU) and a Case-Based Scoring Unit (CBSU). With the information from the interview of the patient, biosignal calibration and recording-time preprocessing, the PCU sets the thresholds of basic tones in order to adapt to various subjects. These thresholds are used at the SPU, RBSU and CBSU. From the 30-second polysomnographic signals, the SPU eliminates artifacts, extracts features from background signal activities and detects special events. The RBSU applies

53 Chapter 3. HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING 37 basic standard rules to the facts that are composed of features and events derived from the SPU. If the RBSU fails to reason out above the confidence threshold, then CBSU executes additional scoring by retrieving similar cases from the case database. Figure 3-2 illustrates this architecture. The PCU, SPU and CBSU were implemented with MATLAB (Mathworks, U.S.A.). We used CLIPS developed by NASA as an implementation tool for RBSU [40]. These architectural units are described in the following chapters in succession. Chapter 4 describes the PCU and SPU. Chapter 5 describes the RBSU and Chapter 6 describes the CBSU. In this chapter, the introduction of rule-based reasoning and case-based reasoning will be summarized and the integration of these artificial intelligence techniques for sleep stage classification will be discussed.

54 Chapter 3. HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING 38 next epoch Yes Load Polysomno Data Signal Processing (SPU) Rule-Based Scoring (RBSU) Acceptable? No Case-Based Scoring (CBSU) Patient Calibration (PCU) Figure 3-2. Architecture of the automated sleep scoring engine: SPU, RBSU and CBSU are the major functional units. The PCU sets basic thresholds of tones from background activities. From the 30-second polysomnographic signals, the SPU rejects artifacts, extracts features from background signal activities and detects special events. The RBSU applies basic standard rules to the facts (features and events) derived from the SPU. If the RBSU fails to reason out acceptably, i.e., the overall reliability value (which will be discussed in the following section) of reasoning is lower than the predefined threshold, then the CBSU executes additional scoring.

55 Chapter 3. HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING Rule-based Reasoning Rule-based reasoning (RBR) is the most common type of knowledge-based expert systems. The terms rule-based system, expert system, knowledge-based expert system are often used synonymously. Newell and Simon demonstrated that much of human problem solving or cognition could be expressed by IF THEN-type production rules [40] such as following; IF it rains, THEN carry umbrella. IF there is a flame, THEN there is a fire. The Newell and Simon model of human problem solving in terms of long-term memory (knowledge base), short-term memory (working memory), and a cognitive processor (inference engine) is regarded as the basis of modern rule-based expert systems. A practical RBR system consists of a knowledge base, a working memory, and an inference engine [40]. The knowledge base contains the domain knowledge represented in the form of IF THEN type rules. Working memory is a global database of facts to which the rules are applied. Inference engine makes inferences by deciding which rules are satisfied by the facts, prioritizing rules and executing the rules according to their priority. Agenda is a prioritized list of rules created by the inference engine, whose patterns are satisfied by the facts in the working memory. Explanation facility explains the decision of the system and the user interface enables the user to communicate with the expert system. Knowledge acquisition facility is optional and provides an automatic way for the user to enter knowledge into the system. Figure 3-2 shows the general

56 Chapter 3. HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING 40 structure of a rule-based expert system. A RBR system solves problems by applying previously established rules to the given problem. The advantage of rule-based reasoning is ease of modification, and ease of addition of rules into the knowledge base due to the separation of the knowledge base and the inference engine. However, this system has limitations in representing human knowledge. Transferring human knowledge into an expert system is so challenging that it is called the knowledge acquisition bottleneck [40]. In addition, RBR does not benefit from the experience gained of its use nor does it learn from its past failures. In spite of these limitations, expert systems have been successful in dealing with real-world problems that conventional programming methodologies have been unable to solve, especially dealing with uncertain or incomplete information. A comparison of conventional programming systems and rule-based expert system is shown in Table 3-1[54]. KNOWLEDGE KNOWLEDGE BASE (RULES) BASE (RULES) INFERENCE ENGINE AGENDA WORKING MEMORY WORKING (FACTS) MEMORY (FACTS) USER INTERFACE User Expert Knowledge Engineer KNOWLEDGE ACQUISITION FACILITY EXPLANATION FACILITY Figure 3-3. A general structure of a rule-based expert system

57 Chapter 3. HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING 41 Table 3-1. Comparison of Conventional programming Systems and RBR System [Efraim Turban and Jay E. Aronson, 1998] Conventional Systems Information and its processing are usually combined in one sequential program. Program does not make mistakes (programmers do). Do not (usually) explain why input data are needed or how conclusions are drawn. Required all input data. May not function properly with missing data unless planned for. Changes in the program are tedious. The system operates only when it is completed. Execution is done on a step-by-step (algorithmic) basis Effective manipulation of large database. Representation and use of data. Efficiency is a major goal. Easily deal with quantitative data. Use numerical data representations. Capture, magnify, and distribute access to numeric data or information. RBR Systems Knowledge base is clearly separated from the processing (inference) mechanism. Program may make mistakes. Explanation is a part of most ES. Do not require all initial facts. Typically can arrive at reasonable conclusions with missing facts. Changes in the rules are easy to make. The system can operate with only a few rules (as the first prototype). Execution is done by using heuristics and logic. Effective manipulation of large knowledge bases. Representation and use of knowledge. Effectiveness is the major goal. Easily deal with qualitative data. Use symbolic knowledge representation. Capture, magnify and distribute access to judgment and knowledge.

58 Chapter 3. HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING Case-based reasoning In practical problem solving areas, especially in medically related fields, heuristics play an essential role in solving some types of problems. Heuristics are rules of thumb or empirical knowledge derived from experience. Heuristics are not guaranteed to work, but they may aid in the solution. Case-based reasoning (CBR) deals with cases that comprise heuristic knowledge, especially knowledge gained by experience. CBR is a relatively new field of artificial intelligence [41][55-57]. Instead of applying rules to the problem, CBR solves problems by making use of solutions to the previous problems. This system is based on two hypotheses on the nature of the world as follows: (1) Regularity: similar problems have similar solutions; (2) Recurrence: the occurred problems tend to recur [43]. Consequently, future problems are likely to be similar to current problems. Case is generally represented as a problem and solution pair. Previous problemsolution cases are accumulated in the case-base. The reasoning procedure of CBR is generally categorized into the following four cycles: case retrieve, reuse, revise, and retain. These cycles are illustrated in Figure 3-4 [42]. When a new problem is inserted, the reasoning system retrieves a similar case from the case-base using retrieval algorithm such as nearest-neighbor retrieval. The solution part of the retrieved case is reused for the problem. If the retrieved case does not match with the given problem sufficiently, the solution is revised to adapt to that problem by repair rules. If the solution is still not adequate, human intervention may be required to modify the solution, make a new case, and retain to the case-base for future use.

59 Chapter 3. HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING 43 Figure 3-4. A general architecture of a case-based reasoning system adopted from Aamodt & Plaza, 1994 The study of CBR has been driven by two motivations [43]. The first is the desire to model human reasoning and learning in cognitive science. The second is the pragmatic desire to advance the technology of artificial intelligence. Leake [43] summarized the advantage of the case-based reasoning over other artificial intelligence technology as below: (1) Knowledge acquisition: A CBR system can solve the problem without abstraction of rules from various situations and without complete knowledge of the application domain. The rule acquisition process can be laborious and unreliable: it may be difficult to elicit rules, and there is no assurance that those rules will actually be

60 Chapter 3. HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING 44 sufficient to characterize expert performance. Case-based reasoning reasons from previous cases and does not require decomposition of experience. (2) Knowledge Maintenance: Initial understanding of the problem is often imperfect, requiring system knowledge to be refined. Changes in knowledge structure may render existing knowledge obsolete. On the other hand, CBR allows users to aid in the acquisition and maintenance of knowledge by adding missing cases to the case database without expert intervention. Case-based reasoning can perform incremental learning. With a limited set of seed cases, a case-base can be augmented during the problem-solving procedure with a case retention policy. (3) User Acceptance: neural network systems cannot provide explanations of their decisions, and rule-based systems must explain their decisions by reference to their rules, which the user may not understand or accept. On the other hand, the results of casebased reasoning are based on previous cases that can be presented to users. However, we should keep in mind the limitations of CBR. Disadvantages of casebased reasoning summarized by Kolodner [55] are as follow: (1) A case-based reasoner might be tempted to use old cases blindly, relying on previous experience without validating it in the new situation. (2) A case-based reasoner might allow cases to bias him/her or it too much in solving new problem. (3) Often people, especially novices, do not consider the most appropriate sets of cases from the case-base when they are reasoning.

61 Chapter 3. HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING 45 In summary, CBR is context-based reasoning approach using previous experiences and has a lot of merits in practical areas despite some limitations. CBR research has not long history and has a large research area Hybrid Rule-based and Case-based Reasoning The different characteristics of RBR and CBR are listed in Table 3-2. The application areas of RBR are closed, narrow and well understood enough to be covered by rules, while CBR is appropriate for areas in which knowledge of the domain is not easily represented by rules, but is easily accumulated through the experience of more solved problems [41][56-57]. Because of the above different but complementary characteristics, integration of CBR with RBR has been researched in different tasks such as classification [45] and diagnosis [46][47]. It should be noted that neither source subsumes the other the general knowledge in the rules is not necessarily well represented by any given set of cases, while the eccentric knowledge in the cases is not necessarily captured by known rules [48].

62 Chapter 3. HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING 46 Table 3-2. The comparison of Rule Based Reasoning and Case based Reasoning adopted from Applying Case-Based Reasoning [41] Problem Area RBR Narrow, well understood, strong domain theory, stable over time Knowledge facts and IF-THEN rules Cases Representation System provides answers Precedents Explanation by trace of fired rules Precedents CBR Wide, poorly understood, weak domain theory, dynamic over time System can learn no, usually requires manual Yes, by case acquisition addition of new rules Etc well-defined knowledge Experience-rich and knowledge-poor Sleep staging involves the type of hybrid reasoning in which human experts use both rule-based knowledge and experience-based knowledge in order to score sleep stages. Rules play a major role in sleep staging because sleep stage scoring has the standard R & K criteria. To the extent human sleep staging is well understood in implementation detail, RBR is an efficient and powerful for scoring. However, implementation of the R & K criteria using RBR, has some limitations. (1) Inexactness: R & K criteria are a type of inexact reasoning in that many of the rules in the R & K criteria cannot be described to mathematical detail. They are largely based on linguistic propositions. Therefore, some definitions are too rough to implement directly. In certain descriptions, detailed definitions are even missed for practical

63 Chapter 3. HYBRID REASNONING ARCHITECTURE FOR SLEEP STAGING 47 implementation. R & K criteria are just a set of guidelines and rules of thumb in some points of view. (2) Coverage: Knowledge engineer cannot program all rules that meet the R & K criteria in implementation. In practice, the sleep phenomenon is so complex that complete abstraction for generating rules is not possible. In other words, it is impractical to understand the domain knowledge fully. These limitations can, to some extent, be overcome by using CBR concept because CBR deals with chunks of domain knowledge and does not require a detailed knowledge of domain, i.e. sleep staging. Therefore, our system adopted the hybrid RBR and CBR approach to analyze sleep stages. Among several integration model of RBR and CBR, the usual integration is to use RBR to deal with knowledge in a standard situation and to use CBR to deal with the problems that are not covered by rules and that require past problem solving experiences [45-46][57]. In these cases, RBR is usually applied first. When RBR fails to provide a reliable solution, CBR looks for past cases that best match the current problem. We decided to follow this RBR-first-CBR-last integration that uses rules to generate a strict sleep stage scoring, and cases to deal with details and exceptions to the rules caused by sleep complexity.

64 Chapter 4 DATA PREPARATION 4.1. Introduction In this chapter, as a data preparation process, patient calibration unit (PCU) and signal processing unit (SPU) are described. The electrophysiological signals of sleep vary according to sleep disorders and also between individual. Because of this, subject calibration is required to adapt to subject specific characteristics. This is the role of the PCU. The primary purpose of the SPU is to extract features from the electrophysiological signals and to support materials for the facts and cases that are used as sources of the next reasoning processes. We applied the following signal processing algorithms: 1) artifact rejection; 2) extraction of background information; and 3) event detection. Polysomnographic recording takes a long time and is exposed to a noisy environment such as detachment of electrodes by movement, interference across the channels, ECG artifacts in the EEG, 20 Hz harmonic noise, and drift due to sweat. The SPU first monitors these background noises to determine whether or not to apply the artifact rejection method and, if required, applies rejection algorithms. The rejection algorithms we developed were adaptive notch filtering for 20 Hz harmonics removal and autoregressive elimination of the ECG artifact in the EEG.

65 Chapter 4. DATA PREPARATION 49 To extract background information, we used both time-domain and frequencydomain algorithms. Time-domain algorithms comprise waveform finding and statistical methods using zero crossing rate, variance, and auto-regressive model parameters. In the frequency-domain, the power spectrum was mainly used for the calculation of the power of each frequency band, center frequency, peak frequency, and other quantities. The SPU also has special expert modules for detecting particular events such as spindles, K-complexes, rapid eye movements (REMs) and arousals. A reliability value of the event, P(E e), is assigned to the every detected event and is used as a reliability of the event at the reasoning step. The extracted information is translated into the syntax of the rule-based scoring unit Subject Specific Calibration As mentioned above, there exists inter-subject and inter-group variability in the characteristics of polysomnogram. One may have faster alpha activities compared with the others, and others have lower alpha amplitude or lower spindle density. To adapt to the peculiarity of each subject, the human scorer uses three types of heuristic information obtained from: 1) physiological signal calibration; 2) surveying the whole night s record; and 3) interviewing the patient. We modeled these types of information into algorithms using several parameters, such as a histogram of the EEG frequency bands and a histogram of the chin-emg variables. In this step, basic tones of the subject are estimated and thresholds are tuned according to the estimate. Brief summary of these algorithms are as follows;

66 Chapter 4. DATA PREPARATION Alpha wave frequency range Determining the alpha wave frequency range follows the method suggested in the paper [58]. Alpha wave frequency ranges are calculated from the power spectrum of every 10-second EEG. Peaks and troughs of every power spectrum are calculated to compose each histogram of peaks and troughs. From the histogram, the alpha frequency range is determined by calculating peak frequency around traditional alpha frequency bands (8-12 Hz). An exemplary histogram of alpha peaks and troughs are shown in Figure Hz Number freq [Hz] Figure 4-1. An example of determining the alpha frequency band: The bold solid line is the number of peaks detected at each frequency and the narrow dotted line indicates that of troughs.

67 Chapter 4. DATA PREPARATION Tone of chin EMG The tone of the chin EMG remains low during Stage REM and is an important index in differentiating Stage REM from Stage W. The PCU estimates the threshold that indicates low activity of the EMG. But, this tone is relative and the low threshold is determined by inspecting signal calibrations and the trend of the overall recording. By referencing the tone level during biocalibration (denoted by Tc), the lowest tone level of the entire recording (denoted by Tl), and the maximum tones in the histogram (Tm), the threshold of EMG tone (Thr) is determined. To calculate T1, Tm and Tc, every halfsecond EMG variance is calculated during whole epoch. Approximate algorithm is as follows; [ α( Tm Tl) + Tm] + (1 κ Tgen Thr = κ ) (4-1), where κ and α are weights and Tgen is a standard threshold determined by trial and error. Figure 4-2 shows an exemplary histogram of EMG tones of normal subject. Figure 4-2. An exemplary display of determining the threshold of chin EMG tone. The level of chin EMG tone (Thr ) is determined by calculating Tl, Tc and Tm.

68 Chapter 4. DATA PREPARATION Artifact Rejection Polysomnographic recordings have various artifacts such as ECG interference in the EEG, harmonic noises, and drifts caused by sweat. These artifacts are automatically measured and rejected if the levels of the artifacts are above rejection thresholds Removing ECG interference to EEG by AR interpolation The basic idea of the ECG artifact elimination using the AR interpolation method is to regard the QRS spike regions of the EEG as missing samples. These missing samples are interpolated using signal model and information contained in the neighboring samples. In general, perfect interpolation is not possible and distortion is inevitable. However, in order to minimize the error in interpolating, we adopted statistical information of the neighboring signal using an Autoregressive(AR) model [59]. (1) Evaluating the degree of the effect of ECG To determine whether to apply the artifact elimination algorithm, we first evaluated the degree of ECG-interference. In order to determine the ECG-corrupted segment, we found the R-wave in the ECG and calculated the correlation coefficient of the EEG and the ECG around the R-wave. If the correlation coefficient was above 0.5, then we applied the removal procedure. (2) The method of eliminating the artifact This algorithm is composed of a two-stage process. In the first stage, AR model

69 Chapter 4. DATA PREPARATION 53 coefficients are estimated from the known neighboring signal. In the second stage, these estimated model coefficients are used to interpolate the missing samples. The AR model of the signal x can be described as P x( n) = ak x( n k) + k= 1 e( n) (4-2), where a k is the model coefficients and e(n) is a zero mean excitation noise. Assuming that M samples (for one segment) from the N samples of the original signal are missing, we denote it as = { x( k),..., x( k + M 1)}. From the remaining known x Uk N-M samples, the missing samples x uk can be estimated by the forward and backward prediction. The result of the elimination process is illustrated in Figure 4-3. ECG in EEG ECG Filtered EEG Figure 4-3. Elimination of ECG artifacts in EEG using AR interpolation method. Top is the ECG contaminated EEG, middle is the reference ECG and the bottom is the ECG eliminated EEG after applying the AR interpolation method.

70 Chapter 4. DATA PREPARATION Removing harmonic noises Adaptive notch filtering is performed when 20 Hz or 60 Hz harmonic noises are continuously detected on the power spectrum of several epochs. The filtering method we used is an adaptive LMS algorithm with the following structure. Primary input d+n Reference Sinusoid 90' Delay x (n) x (n) Adaptive Filter y (n) y (n) + y(n) + Figure 4-4. Adaptive notch filtering using LMS algorithm. Reference sinusoid is 60 Hz wave for the 60 Hz notch filter. The weight updating equation is W ( n + 1) = W ( n) + 2βe( n) X ( n) (4-3), where W(n) is the weight vector, e(n) is the estimation error, and X(n) is the input vector. β indicates the adaptation coefficient. This algorithm has the merit of easy application to the arbitrary frequency Removing drift caused by sweat

71 Chapter 4. DATA PREPARATION 55 Low frequency band filter is applied to check drifts and then high pass filtering is applied if drifts exist Feature Extraction from Background signal activities Among various features, the most representative features to characterize the background activities are 1) wave segment of the EEG; 2) LPC (Linear Predictive Coding) coefficients of the EEG; 3) state of the EOG; and 4) tone of the chin EMG Wave segment of EEG The overall band powers during one 30-second epoch have been used as the main features in several papers [8][10]. However, these features give no precise information on the temporal distribution of the waves of different frequency bands during an entire epoch. This temporal distribution of the waves is important in sleep stage scoring, because many standard rules are defined by the waveform duration and distribution. For example, if the total duration of the alpha wave is prominent for more than 50 %, then the epoch is Stage W according to Rechtschaffen and Kales rule. In order to include the temporal wave distribution, one epoch is divided into 1-second intervals. Hamming windowed FFT (Fast Fourier Transform) is calculated to assign indexes that correspond to the dominant frequency band of the segment. Assigned indexes are denoted as Delta ( Hz), Theta (3-7 Hz), Alpha ( Hz), Beta1 (12-20 Hz), Beta2 (20-50 Hz), and Spindle ( Hz). Figure 4-5 shows an exemplary display of wave segment index.

72 Chapter 4. DATA PREPARATION 56 EEG Hz Hz Hz Hz Hz Hz 50 0 T T T T T D D T D T D T T A T T T A A A T N A T D T A D T A Time [sec] Figure 4-5. Wave Segment Index of EEG : The dominant frequency of each 1 second interval is selected as the index of the segment.

73 Chapter 4. DATA PREPARATION LPC coefficients of EEG The total power spectrum of one epoch s EEG is estimated using the MEM (Maximum Entropy Method) by calculating LPC coefficients. A linear prediction model is defined as P x( m) = a x( m k ) + e( m) (4-4) k = 1 k, where e is a random signal with the variance of 2 σ e The power spectrum of an autoregressive process is given by P MEM XX σ ( f ) = P 1 a k= 1 2 e k e j2πfk. (4-5) State of EOG Several signal characteristics are found at the EOG according to the sleep stages. The state of the EOG is classified into one of five classes: SEM (slow eye movement), DRIFT, DELTA, QUIET, and NORMAL. The 125 Hz sampled EOG is decimated by 10 and the power spectrum is estimated by the MEM method. To classify EOG state, three types of parameters are calculated: (1) total cross spectral power between LEOG and REOG at the frequency range of Hz (denoted as lp1and rp1 for LEOG and REOG); (2) total cross spectral power between the LEOG and REOG at the frequency range of Hz (denoted as lp2 and rp2 for each EOG), and (3) the correlation coefficient between the left and right EOG (denoted as crcf). Figure 4-6

74 Chapter 4. DATA PREPARATION 58 shows the procedure of the EOG state classification and Figure 4-7 shows the relationship between the EOG state and the sleep hypnogram. Calculate power spectrum of both EOG & correlation coefficient. Is the spectral power ( Hz) of left and right EOG higher than the slow eye movement threshold SEM_THR? YES Is both EOG negatively correlated? YES status = 'SEM' NO NO status = 'DRIFT' Is the spectral power ( Hz) of left and right EOG higher than the slow eye movement threshold SEM_THR? Is the correlation of both EOG positive? YES status = 'DELTA' NO Is the spectral power ( Hz) of left and right EOG lower than the quiet threshold QUIET_THR? YES status = 'QUIET' NO status = 'NORMAL' Figure 4-6. Classification of EOG state

75 Chapter 4. DATA PREPARATION 59 MT REM WAKE Q KSH Hypnogram & EOG state QUIET SEM DRIFT NORMAL DELTA Figure 4-7. An example of the relationship between Hypnogram and EOG state at normal sleep Tone of chin EMG The tone of the chin EMG remains low during Stage REM and is an important index in differentiating Stage REM from Stage W. The tone of each epoch is calculated from the average value of every half-second EMG variance between inter-quartile ranges. The inter-quartile range is adopted to reject burst-type high activities. The dynamic range of the EMG tone is so high that we used the logarithm of the EMG variance as the tone of the epoch as shown by the following equations. 1 Vs = N N 1 Te = log N RV = exp [ xs( t) xs ], t = 1 S S IQR where x Vs, where N ( Vs) { ( Tc Tl) / α log(10) } s 1 = N S N t = 1 x ( t) s = numof { S IQR( Vs) } (4-6)

76 Chapter 4. DATA PREPARATION 60, where Vs is the variance of a segment in the epoch and Te is the tone of the epoch. RV is the reliability value of the chin EMG that will be explained in the next section Waveform Detection Introduction Current sleep stage scoring is largely dependent on typical waveform events such as sleep spindle, K-complex, and rapid eye movement. Arousal is also an important event that shows transition between stages. Two approaches have been applied on the event detection problem in the EEG. One is the time domain approach and the other is the frequency domain approach. Sleep spindle and alpha activity are narrow frequency signals and the frequency domain approach may be better. Arousal events can be easily detected by frequency change. K-complex, slow delta wave, and saw-tooth wave may be effectively detected by the time domain approach Sleep Spindle In this system, we used the short time Fourier transform (STFT) method to detect spindles. A wave of which the power of the Hz band is relatively high is regarded as a spindle. Figure 4-8 shows an example of spindle detection using STFT.

77 Chapter 4. DATA PREPARATION Hz Frequency [Hz] EEG(C3:A2) Time [sec] Figure 4-8. Detecting sleep spindle using STFT K-complex A wave with rapid rising and rapid falling above the 70-microvolt level at the 3-7 Hz band is considered a K-complex. The detection procedure is basically a time domain approach. Calculating parameters (RISING_DIST, FALLING_DIST, RISING_DUR and FALLING_DUR illustrated in Figure 4-9), K-complex is determined if a waveform meets the condition (20 uv < RISING DIST < 120 uv) and (60 uv < FALLING DIST < 180 uv) and (RISING DUR + FALLING DUR > 250 msec). Figure 4-10 shows an example of automated detection of K-complex.

The AASM Manual for the Scoring of Sleep and Associated Events

The AASM Manual for the Scoring of Sleep and Associated Events The AASM Manual for the Scoring of Sleep and Associated Events Summary of Updates in Version 2.1 July 1, 2014 The American Academy of Sleep Medicine (AASM) is committed to ensuring that The AASM Manual

More information

Basics of Polysomnography. Chitra Lal, MD, FCCP, FAASM Assistant professor of Medicine, Pulmonary, Critical Care and Sleep, MUSC, Charleston, SC

Basics of Polysomnography. Chitra Lal, MD, FCCP, FAASM Assistant professor of Medicine, Pulmonary, Critical Care and Sleep, MUSC, Charleston, SC Basics of Polysomnography Chitra Lal, MD, FCCP, FAASM Assistant professor of Medicine, Pulmonary, Critical Care and Sleep, MUSC, Charleston, SC Basics of Polysomnography Continuous and simultaneous recording

More information

LEARNING MANUAL OF PSG CHART

LEARNING MANUAL OF PSG CHART LEARNING MANUAL OF PSG CHART POLYSOMNOGRAM, SLEEP STAGE SCORING, INTERPRETATION Sleep Computing Committee, Japanese Society of Sleep Research LEARNING MANUAL OF PSG CHART POLYSOMNOGRAM, SLEEP STAGE SCORING,

More information

Simplest method: Questionnaires. Retrospective: past week, month, year, lifetime Daily: Sleep diary What kinds of questions would you ask?

Simplest method: Questionnaires. Retrospective: past week, month, year, lifetime Daily: Sleep diary What kinds of questions would you ask? Spencer Dawson Simplest method: Questionnaires Retrospective: past week, month, year, lifetime Daily: Sleep diary What kinds of questions would you ask? Did you nap during the day? Bed time and rise time

More information

linkedin.com/in/lizziehillsleeptechservices 1

linkedin.com/in/lizziehillsleeptechservices  1 BSS2015 Hands-On Tech Breakfast SCORING SLEEP USING AASM GUIDELINES: A BRIEF INTRODUCTION Lizzie Hill BSc RPSGT EST Specialist Respiratory Clinical Physiologist, Royal Hospital for Sick Children, Edinburgh

More information

SLEEP STAGING AND AROUSAL. Dr. Tripat Deep Singh (MBBS, MD, RPSGT, RST) International Sleep Specialist (World Sleep Federation program)

SLEEP STAGING AND AROUSAL. Dr. Tripat Deep Singh (MBBS, MD, RPSGT, RST) International Sleep Specialist (World Sleep Federation program) SLEEP STAGING AND AROUSAL Dr. Tripat Deep Singh (MBBS, MD, RPSGT, RST) International Sleep Specialist (World Sleep Federation program) Scoring of Sleep Stages in Adults A. Stages of Sleep Stage W Stage

More information

CHAPTER 6 INTERFERENCE CANCELLATION IN EEG SIGNAL

CHAPTER 6 INTERFERENCE CANCELLATION IN EEG SIGNAL 116 CHAPTER 6 INTERFERENCE CANCELLATION IN EEG SIGNAL 6.1 INTRODUCTION Electrical impulses generated by nerve firings in the brain pass through the head and represent the electroencephalogram (EEG). Electrical

More information

Polysomnography Artifacts and Updates on AASM Scoring Rules. Robin Lloyd, MD, FAASM, FAAP 2017 Utah Sleep Society Conference

Polysomnography Artifacts and Updates on AASM Scoring Rules. Robin Lloyd, MD, FAASM, FAAP 2017 Utah Sleep Society Conference Polysomnography Artifacts and Updates on AASM Scoring Rules Robin Lloyd, MD, FAASM, FAAP 2017 Utah Sleep Society Conference x Conflict of Interest Disclosures for Speakers 1. I do not have any relationships

More information

Arousal detection in sleep

Arousal detection in sleep Arousal detection in sleep FW BES, H KUYKENS AND A KUMAR MEDCARE AUTOMATION, OTTHO HELDRINGSTRAAT 27 1066XT AMSTERDAM, THE NETHERLANDS Introduction Arousals are part of normal sleep. They become pathological

More information

CLASSIFICATION OF SLEEP STAGES IN INFANTS: A NEURO FUZZY APPROACH

CLASSIFICATION OF SLEEP STAGES IN INFANTS: A NEURO FUZZY APPROACH CLASSIFICATION OF SLEEP STAGES IN INFANTS: A NEURO FUZZY APPROACH J. E. Heiss, C. M. Held, P. A. Estévez, C. A. Perez, C. A. Holzmann, J. P. Pérez Department of Electrical Engineering, Universidad de Chile,

More information

Non-contact Screening System with Two Microwave Radars in the Diagnosis of Sleep Apnea-Hypopnea Syndrome

Non-contact Screening System with Two Microwave Radars in the Diagnosis of Sleep Apnea-Hypopnea Syndrome Medinfo2013 Decision Support Systems and Technologies - II Non-contact Screening System with Two Microwave Radars in the Diagnosis of Sleep Apnea-Hypopnea Syndrome 21 August 2013 M. Kagawa 1, K. Ueki 1,

More information

EEG Arousals: Scoring Rules and Examples. A Preliminary Report from the Sleep Disorders Atlas Task Force of the American Sleep Disorders Association

EEG Arousals: Scoring Rules and Examples. A Preliminary Report from the Sleep Disorders Atlas Task Force of the American Sleep Disorders Association EEG Arousals: Scoring Rules and Examples A Preliminary Report from the Sleep Disorders Atlas Task Force of the American Sleep Disorders Association Sleep in patients with a number of sleep disorders and

More information

NATIONAL COMPETENCY SKILL STANDARDS FOR PERFORMING POLYSOMNOGRAPHY/SLEEP TECHNOLOGY

NATIONAL COMPETENCY SKILL STANDARDS FOR PERFORMING POLYSOMNOGRAPHY/SLEEP TECHNOLOGY NATIONAL COMPETENCY SKILL STANDARDS FOR PERFORMING POLYSOMNOGRAPHY/SLEEP TECHNOLOGY Polysomnography/Sleep Technology providers practice in accordance with the facility policy and procedure manual which

More information

Combining EEG with Heart Rate Training for Brain / Body Optimization. Combining EEG with Heart Rate Training. For Brain / Body Optimization

Combining EEG with Heart Rate Training for Brain / Body Optimization. Combining EEG with Heart Rate Training. For Brain / Body Optimization Combining EEG with Heart Rate Training For Brain / Body Optimization Thomas F. Collura, Ph.D. March 13, 2009 DRAFT There is a growing interest in combining different biofeedback modalities, in particular

More information

Western Hospital System. PSG in History. SENSORS in the field of SLEEP. PSG in History continued. Remember

Western Hospital System. PSG in History. SENSORS in the field of SLEEP. PSG in History continued. Remember SENSORS in the field of SLEEP Mrs. Gaye Cherry: Scientist in Charge Department of Sleep and Respiratory Medicine Sleep Disorders Unit Western Hospital PSG in History 1875: Discovery of brain-wave activity

More information

A Modified Method for Scoring Slow Wave Sleep of Older Subjects

A Modified Method for Scoring Slow Wave Sleep of Older Subjects Sleep, 5(2):195-199 1982 Raven Press, New York A Modified Method for Scoring Slow Wave Sleep of Older Subjects Wilse B. Webb and Lewis M. Dreblow Department of Psychology, University of Florida, Gainesville,

More information

The AASM Manual for the Scoring of Sleep and Associated Events

The AASM Manual for the Scoring of Sleep and Associated Events The AASM Manual for the Scoring of Sleep and Associated Events The 2007 AASM Scoring Manual vs. the AASM Scoring Manual v2.0 October 2012 The American Academy of Sleep Medicine (AASM) is committed to ensuring

More information

Sleep diagnostics systems

Sleep diagnostics systems Sleep diagnostics systems DeVilbiss Healthcare introduces the SleepDoc Porti diagnostics systems powered by the technology and expertise of Dr Fenyves und Gut. From a 5 channel respiratory screener up

More information

Artifact Recognition and Troubleshooting

Artifact Recognition and Troubleshooting Artifact Recognition and Troubleshooting 2017 Focus Fall Super Session The Best of the Best For Respiratory Therapists and Sleep Technologists The Doubletree Hilton Hotel Pittsburgh, PA Thursday Sept.

More information

Processed by HBI: Russia/Switzerland/USA

Processed by HBI: Russia/Switzerland/USA 1 CONTENTS I Personal and clinical data II Conclusion. III Recommendations for therapy IV Report. 1. Procedures of EEG recording and analysis 2. Search for paroxysms 3. Eyes Open background EEG rhythms

More information

Outlining a simple and robust method for the automatic detection of EEG arousals

Outlining a simple and robust method for the automatic detection of EEG arousals Outlining a simple and robust method for the automatic detection of EEG arousals Isaac Ferna ndez-varela1, Diego A lvarez-este vez2, Elena Herna ndez-pereira1 and Vicente Moret-Bonillo1 1- Universidade

More information

Proceedings 23rd Annual Conference IEEE/EMBS Oct.25-28, 2001, Istanbul, TURKEY

Proceedings 23rd Annual Conference IEEE/EMBS Oct.25-28, 2001, Istanbul, TURKEY AUTOMATED SLEEP STAGE SCORING BY DECISION TREE LEARNING Proceedings 23rd Annual Conference IEEE/EMBS Oct.25-28, 2001, Istanbul, TURKEY Masaaki Hanaoka, Masaki Kobayashi, Haruaki Yamazaki Faculty of Engineering,Yamanashi

More information

VCE Psychology Unit 4. Year 2017 Mark Pages 45 Published Feb 10, 2018 COMPREHENSIVE PSYCHOLOGY UNIT 4 NOTES, By Alice (99.

VCE Psychology Unit 4. Year 2017 Mark Pages 45 Published Feb 10, 2018 COMPREHENSIVE PSYCHOLOGY UNIT 4 NOTES, By Alice (99. VCE Psychology Unit 4 Year 2017 Mark 50.00 Pages 45 Published Feb 10, 2018 COMPREHENSIVE PSYCHOLOGY UNIT 4 NOTES, 2017 By Alice (99.45 ATAR) Powered by TCPDF (www.tcpdf.org) Your notes author, Alice. Alice

More information

AUTOMATED SLEEP SCORING SYSTEM USING LABVIEW. A Thesis PARIKSHIT BAPUSAHEB DESHPANDE

AUTOMATED SLEEP SCORING SYSTEM USING LABVIEW. A Thesis PARIKSHIT BAPUSAHEB DESHPANDE AUTOMATED SLEEP SCORING SYSTEM USING LABVIEW A Thesis by PARIKSHIT BAPUSAHEB DESHPANDE Submitted to the Office of Graduate Studies of Texas A&M University in partial fulfillment of the requirements for

More information

Outline of Talk. Introduction to EEG and Event Related Potentials. Key points. My path to EEG

Outline of Talk. Introduction to EEG and Event Related Potentials. Key points. My path to EEG Outline of Talk Introduction to EEG and Event Related Potentials Shafali Spurling Jeste Assistant Professor in Psychiatry and Neurology UCLA Center for Autism Research and Treatment Basic definitions and

More information

Novel single trial movement classification based on temporal dynamics of EEG

Novel single trial movement classification based on temporal dynamics of EEG Novel single trial movement classification based on temporal dynamics of EEG Conference or Workshop Item Accepted Version Wairagkar, M., Daly, I., Hayashi, Y. and Nasuto, S. (2014) Novel single trial movement

More information

Sleep Spindle Detection Based on Complex Demodulation Jia-bin LI, Bei WANG* and Yu ZHANG

Sleep Spindle Detection Based on Complex Demodulation Jia-bin LI, Bei WANG* and Yu ZHANG International Conference on Biomedical and Biological Engineering (BBE 216) Sleep Spindle Detection Based on Complex Demodulation Jia-bin LI, Bei WANG* and Yu ZHANG Department of Automation, School of

More information

MOVEMENT RULES. Dr. Tripat Deep Singh (MBBS, MD, RPSGT, RST) International Sleep Specialist (World Sleep Federation program)

MOVEMENT RULES. Dr. Tripat Deep Singh (MBBS, MD, RPSGT, RST) International Sleep Specialist (World Sleep Federation program) MOVEMENT RULES Dr. Tripat Deep Singh (MBBS, MD, RPSGT, RST) International Sleep Specialist (World Sleep Federation program) 1. Scoring Periodic Limb Movement in Sleep (PLMS) A. The following rules define

More information

Human Brain Institute Russia-Switzerland-USA

Human Brain Institute Russia-Switzerland-USA 1 Human Brain Institute Russia-Switzerland-USA CONTENTS I Personal and clinical data II Conclusion. III Recommendations for therapy IV Report. 1. Procedures of EEG recording and analysis 2. Search for

More information

What Every Clinician Should Know About Polysomnography

What Every Clinician Should Know About Polysomnography What Every Clinician Should Know About Polysomnography Susheel P Patil MD PhD Introduction Electrical Concepts Current, Voltage, Resistance Capacitance, Inductance, and Impedance Differential Amplifiers

More information

SleepRT applications for routine and research

SleepRT applications for routine and research SleepRT applications for routine and research OSG BVBA Bussestraat 17 2840 RUMST BELGIUM Tel: +32 (0) 15 32 13 73 Fax: +32 (0) 15 32 13 93 E-mail: info@osg.be Website: www.osg.be VAT number: BE 0425.381.820

More information

A 74-year-old man with severe ischemic cardiomyopathy and atrial fibrillation

A 74-year-old man with severe ischemic cardiomyopathy and atrial fibrillation 1 A 74-year-old man with severe ischemic cardiomyopathy and atrial fibrillation The following 3 minute polysomnogram (PSG) tracing was recorded in a 74-year-old man with severe ischemic cardiomyopathy

More information

Physiology Unit 2 CONSCIOUSNESS, THE BRAIN AND BEHAVIOR

Physiology Unit 2 CONSCIOUSNESS, THE BRAIN AND BEHAVIOR Physiology Unit 2 CONSCIOUSNESS, THE BRAIN AND BEHAVIOR In Physiology Today What the Brain Does The nervous system determines states of consciousness and produces complex behaviors Any given neuron may

More information

Hyatt Moore IV 1, 2 Steve Woodward 3, PhD Emmanuel Mignot 1, MD, PhD. Electrical Engineering Stanford University, CA 2

Hyatt Moore IV 1, 2 Steve Woodward 3, PhD Emmanuel Mignot 1, MD, PhD. Electrical Engineering Stanford University, CA 2 HIGH RESOLUTION DETECTION OF POLYSOMNOGRAPHY BASED PHASIC EVENTS OF REM SLEEP IN POSTRAUMATIC STRESS DISORDER IMPROVING TOOLS FOR PSG ANALYSIS OF REM SLEEP IN PTSD Hyatt Moore IV 1, 2 Steve Woodward 3,

More information

Neurostyle. Medical Innovation for Better Life

Neurostyle. Medical Innovation for Better Life Neurostyle Medical Innovation for Better Life Neurostyle Pte Ltd is a company dedicated to design, develop, manufacture and distribute neurological and neuromuscular medical devices. Strategically located

More information

Appendix 1. Practice Guidelines for Standards of Adult Sleep Medicine Services

Appendix 1. Practice Guidelines for Standards of Adult Sleep Medicine Services Appendix 1 Practice Guidelines for Standards of Adult Sleep Medicine Services 1 Premises and Procedures Out-patient/Clinic Rooms Sleep bedroom for PSG/PG Monitoring/Analysis/ Scoring room PSG equipment

More information

Webinar Q&A Report Noninvasive, Automated Measurement of Sleep, Wake and Breathing in Rodents

Webinar Q&A Report Noninvasive, Automated Measurement of Sleep, Wake and Breathing in Rodents Webinar Q&A Report Noninvasive, Automated Measurement of Sleep, Wake and Breathing in Rodents Q: How does the PiezoSleep system distinguish between inactivity (quiet wake) vs. sleep? B. O Hara: The piezo

More information

Sleep stages. Awake Stage 1 Stage 2 Stage 3 Stage 4 Rapid eye movement sleep (REM) Slow wave sleep (NREM)

Sleep stages. Awake Stage 1 Stage 2 Stage 3 Stage 4 Rapid eye movement sleep (REM) Slow wave sleep (NREM) Sleep stages Awake Stage 1 Stage 2 Stage 3 Stage 4 Rapid eye movement sleep (REM) Slow wave sleep (NREM) EEG waves EEG Electrode Placement Classifying EEG brain waves Frequency: the number of oscillations/waves

More information

Where experience connects with innovation

Where experience connects with innovation Where experience connects with innovation Gold Standard For sleep apnea detection and so much more Many thousands of satisfied users, all over the world NOX T3 IS A RESPIRATORY PORTABLE SLEEP RECORDER

More information

Excessive Daytime Sleepiness Associated with Insufficient Sleep

Excessive Daytime Sleepiness Associated with Insufficient Sleep Sleep, 6(4):319-325 1983 Raven Press, New York Excessive Daytime Sleepiness Associated with Insufficient Sleep T. Roehrs, F. Zorick, J. Sicklesteel, R. Wittig, and T. Roth Sleep Disorders and Research

More information

MESA PSG SLEEP READING CENTER. SRC MANUAL OF OPERATIONS and SCORING RULES

MESA PSG SLEEP READING CENTER. SRC MANUAL OF OPERATIONS and SCORING RULES MESA PSG SLEEP READING CENTER SRC MANUAL OF OPERATIONS and SCORING RULES MESA PSG Sleep 1 SRC Manual of Operations & Scoring Rules Table of Contents Page Number 1. Objectives 4 2. Structure 4 3. Certification

More information

Sleep Stage Estimation By Evolutionary Computation Using Heartbeat Data and Body-Movement

Sleep Stage Estimation By Evolutionary Computation Using Heartbeat Data and Body-Movement Sleep Stage Estimation By Evolutionary Computation Using Heartbeat Data and Body-Movement 1 Hiroyasu MAtsushima, 2 Kazuyuki Hirose, 3 Kiyohiko Hattori, 4 Hiroyuki Sato, 5 Keiki Takadama 1-5, First Author

More information

Polysomnography and Sleep Disorders

Polysomnography and Sleep Disorders 23 Polysomnography and Sleep Disorders Jean K. Matheson, Randip Singh, and Andreja Packard Summary The classification of sleep disorders is based both on clinical and neurophysiological criteria and is

More information

Physiology Unit 2 CONSCIOUSNESS, THE BRAIN AND BEHAVIOR

Physiology Unit 2 CONSCIOUSNESS, THE BRAIN AND BEHAVIOR Physiology Unit 2 CONSCIOUSNESS, THE BRAIN AND BEHAVIOR What the Brain Does The nervous system determines states of consciousness and produces complex behaviors Any given neuron may have as many as 200,000

More information

SleepSign System. FAQs User List Software Overview

SleepSign System. FAQs User List Software Overview FAQs User List Software Overview SleepSign System SSA100W SleepSign System - Frequently Asked Questions 1. Why should I use SleepSign software? 2. Can SleepSign import digital data generated from my BIOPAC

More information

H-Reflex Suppression and Autonomic Activation During Lucid REM Sleep: A Case Study

H-Reflex Suppression and Autonomic Activation During Lucid REM Sleep: A Case Study Sleep 12(4):374-378, Raven Press, Ltd., New York 1989 Association of Professional Sleep Societies Short Communication H-Reflex Suppression and Autonomic Activation During Lucid REM Sleep: A Case Study

More information

Sleep Stages and Scoring Technique

Sleep Stages and Scoring Technique CHAPTER 3 Sleep Stages and Scoring Technique Raman K. Malhotra Alon Y. Avidan Introduction to Sleep Stage Scoring The original Rechtschaffen and Kales sleep scoring manual of 1968, commonly known as the

More information

PSD Analysis of Neural Spectrum During Transition from Awake Stage to Sleep Stage

PSD Analysis of Neural Spectrum During Transition from Awake Stage to Sleep Stage PSD Analysis of Neural Spectrum During Transition from Stage to Stage Chintan Joshi #1 ; Dipesh Kamdar #2 #1 Student,; #2 Research Guide, #1,#2 Electronics and Communication Department, Vyavasayi Vidya

More information

Atlas, rules, and recording techniques for the scoring of cyclic alternating pattern (CAP) in human sleep

Atlas, rules, and recording techniques for the scoring of cyclic alternating pattern (CAP) in human sleep Sleep Medicine 3 (2002) 187 199 Consensus Report Atlas, rules, and recording techniques for the scoring of cyclic alternating pattern (CAP) in human sleep Mario Giovanni Terzano a, *, Liborio Parrino a,

More information

Split Night Protocols for Adult Patients - Updated July 2012

Split Night Protocols for Adult Patients - Updated July 2012 Split Night Protocols for Adult Patients - Updated July 2012 SUMMARY: Sleep technologists are team members who work under the direction of a physician practicing sleep disorders medicine. Sleep technologists

More information

11/18/13 ECG SIGNAL ACQUISITION HARDWARE DESIGN. Origin of Bioelectric Signals

11/18/13 ECG SIGNAL ACQUISITION HARDWARE DESIGN. Origin of Bioelectric Signals ECG SIGNAL ACQUISITION HARDWARE DESIGN Origin of Bioelectric Signals 1 Cell membrane, channel proteins Electrical and chemical gradients at the semi-permeable cell membrane As a result, we get a membrane

More information

ELEC ENG 4BD4 Lecture 1. Biomedical Instrumentation Instructor: Dr. Hubert de Bruin

ELEC ENG 4BD4 Lecture 1. Biomedical Instrumentation Instructor: Dr. Hubert de Bruin ELEC ENG 4BD4 Lecture 1 Biomedical Instrumentation Instructor: Dr. Hubert de Bruin 1 Cochlear Implant 2 Advances in Vision (Retinal Stimulation) 3 Argus II Implant 4 Mini Gastric Imaging 5 Taser 6 Shock

More information

Beyond the Basics in EEG Interpretation: Throughout the Life Stages

Beyond the Basics in EEG Interpretation: Throughout the Life Stages Beyond the Basics in EEG Interpretation: Throughout the Life Stages Steve S. Chung, MD, FAAN Chairman, Neuroscience Institute Director, Epilepsy Program Banner University Medical Center University of Arizona

More information

ISSN: ISO 9001:2008 Certified International Journal of Engineering and Innovative Technology (IJEIT) Volume 2, Issue 10, April 2013

ISSN: ISO 9001:2008 Certified International Journal of Engineering and Innovative Technology (IJEIT) Volume 2, Issue 10, April 2013 ECG Processing &Arrhythmia Detection: An Attempt M.R. Mhetre 1, Advait Vaishampayan 2, Madhav Raskar 3 Instrumentation Engineering Department 1, 2, 3, Vishwakarma Institute of Technology, Pune, India Abstract

More information

HST-582J/6.555J/16.456J-Biomedical Signal and Image Processing-Spring Laboratory Project 1 The Electrocardiogram

HST-582J/6.555J/16.456J-Biomedical Signal and Image Processing-Spring Laboratory Project 1 The Electrocardiogram HST-582J/6.555J/16.456J-Biomedical Signal and Image Processing-Spring 2007 DUE: 3/8/07 Laboratory Project 1 The Electrocardiogram 1 Introduction The electrocardiogram (ECG) is a recording of body surface

More information

Practical 3 Nervous System Physiology 2 nd year English Module. Dept. of Physiology, Carol Davila University of Medicine and Pharmacy

Practical 3 Nervous System Physiology 2 nd year English Module. Dept. of Physiology, Carol Davila University of Medicine and Pharmacy Electroencephalography l h (EEG) Practical 3 Nervous System Physiology 2 nd year English Module Dept. of Physiology, Carol Davila University of Medicine and Pharmacy What is EEG EEG noninvasively records

More information

Speed - Accuracy - Exploration. Pathfinder SL

Speed - Accuracy - Exploration. Pathfinder SL Speed - Accuracy - Exploration Pathfinder SL 98000 Speed. Accuracy. Exploration. Pathfinder SL represents the evolution of over 40 years of technology, design, algorithm development and experience in the

More information

Neonatal EEG Maturation

Neonatal EEG Maturation Neonatal EEG Maturation Cindy Jenkinson, R. EEG T., CLTM October 7, 2017 Fissure Development 3 http://www.hhmi.org/biointeractive/develop ment-human-embryonic-brain 4 WHAT IS IMPORTANT TO KNOW BEFORE I

More information

Polysomnography Course Session: Sept 2017

Polysomnography Course Session: Sept 2017 Polysomnography Course Session: Sept 2017 General Information Polysomnography course will be held at SLEEP AND ALERTNESS CLINIC Med-West Medical centre 750 Dundas St. W., Suite 2-259 (Conference Room)

More information

Assessment of Sleep Disorders DR HUGH SELSICK

Assessment of Sleep Disorders DR HUGH SELSICK Assessment of Sleep Disorders DR HUGH SELSICK Goals Understand the importance of history taking Be able to take a basic sleep history Be aware the technology used to assess sleep disorders. Understand

More information

EEG, ECG, EMG. Mitesh Shrestha

EEG, ECG, EMG. Mitesh Shrestha EEG, ECG, EMG Mitesh Shrestha What is Signal? A signal is defined as a fluctuating quantity or impulse whose variations represent information. The amplitude or frequency of voltage, current, electric field

More information

EPILEPTIC SEIZURE DETECTION USING WAVELET TRANSFORM

EPILEPTIC SEIZURE DETECTION USING WAVELET TRANSFORM EPILEPTIC SEIZURE DETECTION USING WAVELET TRANSFORM Sneha R. Rathod 1, Chaitra B. 2, Dr. H.P.Rajani 3, Dr. Rajashri khanai 4 1 MTech VLSI Design and Embedded systems,dept of ECE, KLE Dr.MSSCET, Belagavi,

More information

Development of OSA Event Detection Using Threshold Based Automatic Classification

Development of OSA Event Detection Using Threshold Based Automatic Classification Development of OSA Event Detection Using Threshold Based Automatic Classification Laiali Almazaydeh, Khaled Elleithy, Varun Pande and Miad Faezipour Department of Computer Science and Engineering University

More information

AGING AND SLEEP STAGE EFFECTS ON ENTROPY OF ELECTROENCEPHALOGRAM SIGNALS

AGING AND SLEEP STAGE EFFECTS ON ENTROPY OF ELECTROENCEPHALOGRAM SIGNALS University of Kentucky UKnowledge University of Kentucky Master's Theses Graduate School 2008 AGING AND SLEEP STAGE EFFECTS ON ENTROPY OF ELECTROENCEPHALOGRAM SIGNALS Swetha Vennelaganti University of

More information

Electroencephalograph EEG-9200J/K

Electroencephalograph EEG-9200J/K Electroencephalograph EEG-9200J/K Efficient and user-friendly EEG-9200 is a Windows XP desktop PC based electroencephalograph that is efficient and easy to use with 8 ch DSA trendgraph, 3D mapping, high

More information

Biomedical Signal Processing

Biomedical Signal Processing DSP : Biomedical Signal Processing What is it? Biomedical Signal Processing: Application of signal processing methods, such as filtering, Fourier transform, spectral estimation and wavelet transform, to

More information

AASM guidelines, when available. Does this mean if our medical director chooses for us to use an alternative rule that our accreditation is at risk?

AASM guidelines, when available. Does this mean if our medical director chooses for us to use an alternative rule that our accreditation is at risk? GENERAL G.1. I see that the STANDARDS FOR ACCREDITATION state that we are to use the recommended AASM guidelines, when available. Does this mean if our medical director chooses for us to use an alternative

More information

Thought Technology Ltd.

Thought Technology Ltd. Thought Technology Ltd. 8205 Montreal/ Toronto Blvd. Suite 223, Montreal West, QC H4X 1N1 Canada Tel: (800) 361-3651 ۰ (514) 489-8251 Fax: (514) 489-8255 E-mail: mail@thoughttechnology.com Webpage: http://www.thoughttechnology.com

More information

Assessment of Reliability of Hamilton-Tompkins Algorithm to ECG Parameter Detection

Assessment of Reliability of Hamilton-Tompkins Algorithm to ECG Parameter Detection Proceedings of the 2012 International Conference on Industrial Engineering and Operations Management Istanbul, Turkey, July 3 6, 2012 Assessment of Reliability of Hamilton-Tompkins Algorithm to ECG Parameter

More information

Brain Computer Interface. Mina Mikhail

Brain Computer Interface. Mina Mikhail Brain Computer Interface Mina Mikhail minamohebn@gmail.com Introduction Ways for controlling computers Keyboard Mouse Voice Gestures Ways for communicating with people Talking Writing Gestures Problem

More information

Recognition of Sleep Dependent Memory Consolidation with Multi-modal Sensor Data

Recognition of Sleep Dependent Memory Consolidation with Multi-modal Sensor Data Recognition of Sleep Dependent Memory Consolidation with Multi-modal Sensor Data The MIT Faculty has made this article openly available. Please share how this access benefits you. Your story matters. Citation

More information

Nov versus Fam. Fam 1 versus. Fam 2. Supplementary figure 1

Nov versus Fam. Fam 1 versus. Fam 2. Supplementary figure 1 a Environment map similarity score (mean r ).5..3.2.1 Fam 1 versus Fam 2 Nov versus Fam b Environment cofiring similarity score (mean r ).7.6.5..3.2.1 Nov versus Fam Fam 1 versus Fam 2 First half versus

More information

Development of a portable device for home monitoring of. snoring. Abstract

Development of a portable device for home monitoring of. snoring. Abstract Author: Yeh-Liang Hsu, Ming-Chou Chen, Chih-Ming Cheng, Chang-Huei Wu (2005-11-03); recommended: Yeh-Liang Hsu (2005-11-07). Note: This paper is presented at International Conference on Systems, Man and

More information

A Sleeping Monitor for Snoring Detection

A Sleeping Monitor for Snoring Detection EECS 395/495 - mhealth McCormick School of Engineering A Sleeping Monitor for Snoring Detection By Hongwei Cheng, Qian Wang, Tae Hun Kim Abstract Several studies have shown that snoring is the first symptom

More information

states of brain activity sleep, brain waves DR. S. GOLABI PH.D. IN MEDICAL PHYSIOLOGY

states of brain activity sleep, brain waves DR. S. GOLABI PH.D. IN MEDICAL PHYSIOLOGY states of brain activity sleep, brain waves DR. S. GOLABI PH.D. IN MEDICAL PHYSIOLOGY introduction all of us are aware of the many different states of brain activity, including sleep, wakefulness, extreme

More information

PD233: Design of Biomedical Devices and Systems

PD233: Design of Biomedical Devices and Systems PD233: Design of Biomedical Devices and Systems (Lecture-7 Biopotentials- 2) Dr. Manish Arora CPDM, IISc Course Website: http://cpdm.iisc.ac.in/utsaah/courses/ Electromyogram (EMG) Skeletal muscles are

More information

This presentation is the intellectual property of the author. Contact them for permission to reprint and/or distribute.

This presentation is the intellectual property of the author. Contact them for permission to reprint and/or distribute. Modified Combinatorial Nomenclature Montage, Review, and Analysis of High Density EEG Terrence D. Lagerlund, M.D., Ph.D. CP1208045-16 Disclosure Relevant financial relationships None Off-label/investigational

More information

ISSN: (Online) Volume 3, Issue 7, July 2015 International Journal of Advance Research in Computer Science and Management Studies

ISSN: (Online) Volume 3, Issue 7, July 2015 International Journal of Advance Research in Computer Science and Management Studies ISSN: 2321-7782 (Online) Volume 3, Issue 7, July 2015 International Journal of Advance Research in Computer Science and Management Studies Research Article / Survey Paper / Case Study Available online

More information

From a screener to a full PSG System in the palm of your hand Upgradeable at any time!

From a screener to a full PSG System in the palm of your hand Upgradeable at any time! SOMNOscreen plus Up to 33 channels Ambulatory & online Continuous Determination of Obstruction Level Wireless real-time data transfer From a screener to a full PSG System in the palm of your hand Upgradeable

More information

MES 9000 MUSCULOSKELETAL EVALUATION SYSTEM

MES 9000 MUSCULOSKELETAL EVALUATION SYSTEM MES 9000 MUSCULOSKELETAL EVALUATION SYSTEM The new MES 9000. Now you can completely and objectively evaluate and document Dynamic Range of Motion, Static and Dynamic EMG and Muscle Testing All with one

More information

Nature Neuroscience: doi: /nn Supplementary Figure 1

Nature Neuroscience: doi: /nn Supplementary Figure 1 Supplementary Figure 1 Hippocampal recordings. a. (top) Post-operative MRI (left, depicting a depth electrode implanted along the longitudinal hippocampal axis) and co-registered preoperative MRI (right)

More information

Relax, you re in safe hands EMBLETTA MPR

Relax, you re in safe hands EMBLETTA MPR Relax, you re in safe hands EMBLETTA MPR EMBLETTA MPR Good health begins with a good night s sleep. So the faster and more accurately the causes of sleep disturbances are diagnosed, the quicker your patients

More information

A Brain Computer Interface System For Auto Piloting Wheelchair

A Brain Computer Interface System For Auto Piloting Wheelchair A Brain Computer Interface System For Auto Piloting Wheelchair Reshmi G, N. Kumaravel & M. Sasikala Centre for Medical Electronics, Dept. of Electronics and Communication Engineering, College of Engineering,

More information

Development of 2-Channel Eeg Device And Analysis Of Brain Wave For Depressed Persons

Development of 2-Channel Eeg Device And Analysis Of Brain Wave For Depressed Persons Development of 2-Channel Eeg Device And Analysis Of Brain Wave For Depressed Persons P.Amsaleka*, Dr.S.Mythili ** * PG Scholar, Applied Electronics, Department of Electronics and Communication, PSNA College

More information

Milena Pavlova, M.D., FAASM Department of Neurology, Brigham and Women's Hospital Assistant Professor of Neurology, Harvard Medical School Medical

Milena Pavlova, M.D., FAASM Department of Neurology, Brigham and Women's Hospital Assistant Professor of Neurology, Harvard Medical School Medical Milena Pavlova, M.D., FAASM Department of Neurology, Brigham and Women's Hospital Assistant Professor of Neurology, Harvard Medical School Medical Director, Faulkner EEG and Sleep Testing Center Course

More information

An Alpha-Wave-Based Binaural Beat Sound Control System using Fuzzy Logic and Autoregressive Forecasting Model

An Alpha-Wave-Based Binaural Beat Sound Control System using Fuzzy Logic and Autoregressive Forecasting Model SICE Annual Conference 2008 August 20-22, 2008, The University Electro-Communications, Japan An Alpha-Wave-Based Binaural Beat Sound Control System using Fuzzy Logic and Autoregressive Forecasting Model

More information

Sleep Diagnostics. With your investment in the SomnoStar

Sleep Diagnostics. With your investment in the SomnoStar Sleep Diagnostics With your investment in the SomnoStar system, you are receiving not only the most sought after sleep system on the market today, but also a partnership with a company dedicated to serving

More information

Emotion Detection Using Physiological Signals. M.A.Sc. Thesis Proposal Haiyan Xu Supervisor: Prof. K.N. Plataniotis

Emotion Detection Using Physiological Signals. M.A.Sc. Thesis Proposal Haiyan Xu Supervisor: Prof. K.N. Plataniotis Emotion Detection Using Physiological Signals M.A.Sc. Thesis Proposal Haiyan Xu Supervisor: Prof. K.N. Plataniotis May 10 th, 2011 Outline Emotion Detection Overview EEG for Emotion Detection Previous

More information

Summary of Features and Performance

Summary of Features and Performance MICHELE SLEEP SCORING SYSTEM Summary of Features and Performance Suite PE438, Princess Elizabeth Building 1 Morley Ave / Winnipeg, Manitoba / R3L 2P4 Canada phone 1 877 949 3202 / fax 204 943 6295 Table

More information

Hybrid EEG-HEG based Neurofeedback Device

Hybrid EEG-HEG based Neurofeedback Device APSIPA ASC 2011 Xi an Hybrid EEG-HEG based Neurofeedback Device Supassorn Rodrak *, Supatcha Namtong, and Yodchanan Wongsawat ** Department of Biomedical Engineering, Faculty of Engineering, Mahidol University,

More information

PCA Enhanced Kalman Filter for ECG Denoising

PCA Enhanced Kalman Filter for ECG Denoising IOSR Journal of Electronics & Communication Engineering (IOSR-JECE) ISSN(e) : 2278-1684 ISSN(p) : 2320-334X, PP 06-13 www.iosrjournals.org PCA Enhanced Kalman Filter for ECG Denoising Febina Ikbal 1, Prof.M.Mathurakani

More information

Relax, you re in safe hands

Relax, you re in safe hands ST/ST+ proxies will be available soon. Natus medical Incorporated 1501 Industrial Rd San Carlos, CA 94070 USA 1-800-303-0306 1-650-802-0400 D-1303-073 Natus Europe GmbH Robert-Koch-Str. 1 82152 Planegg,

More information

A bioimpedance-based cardiovascular measurement system

A bioimpedance-based cardiovascular measurement system A bioimpedance-based cardiovascular measurement system Roman Kusche 1[0000-0003-2925-7638], Sebastian Hauschild 1, and Martin Ryschka 1 1 Laboratory of Medical Electronics, Luebeck University of Applied

More information

A TECH S TOOLKIT FOR THE PEDIATRIC SLEEP LAB

A TECH S TOOLKIT FOR THE PEDIATRIC SLEEP LAB A TECH S TOOLKIT FOR THE PEDIATRIC SLEEP LAB Craig Canapari, MD craig.canapari@gmail.com drcraigcanapari.com: Updated syllabus will be here along with link to visual presentation. Twitter: DrCanapari INTRODUCTION

More information

Testing the Accuracy of ECG Captured by Cronovo through Comparison of ECG Recording to a Standard 12-Lead ECG Recording Device

Testing the Accuracy of ECG Captured by Cronovo through Comparison of ECG Recording to a Standard 12-Lead ECG Recording Device Testing the Accuracy of ECG Captured by through Comparison of ECG Recording to a Standard 12-Lead ECG Recording Device Data Analysis a) R-wave Comparison: The mean and standard deviation of R-wave amplitudes

More information

Sleep Staging with Deep Learning: A convolutional model

Sleep Staging with Deep Learning: A convolutional model Sleep Staging with Deep Learning: A convolutional model Isaac Ferna ndez-varela1, Dimitrios Athanasakis2, Samuel Parsons3 Elena Herna ndez-pereira1, and Vicente Moret-Bonillo1 1- Universidade da Corun

More information

ARMA Modelling for Sleep Disorders Diagnose

ARMA Modelling for Sleep Disorders Diagnose ARMA Modelling for Sleep Disorders Diagnose João Caldas da Costa 1, Manuel Duarte Ortigueira 2, Arnaldo Batista 2, and Teresa Paiva 3 1 DSI, Escola Superior de Tecnologia de Setúbal, Instituto Politécnico

More information

Procedures in the Sleep Laboratory

Procedures in the Sleep Laboratory AAST Technologist Fundamentals Date: May 7, 2017 Focus Conference Location: Orlando, Florida Workshop Procedures in the Sleep Laboratory Laree Fordyce, RST, RPSGT, CCRP Conflict of Interest Disclosures

More information

EEG Electrode Placement

EEG Electrode Placement EEG Electrode Placement Classifying EEG brain waves Frequency: the number of oscillations/waves per second, measured in Hertz (Hz) reflects the firing rate of neurons alpha, beta, theta, delta Amplitude:

More information

Ultrashort Sleep-Wake Cycle: Timing of REM Sleep. Evidence for Sleep-Dependent and Sleep-Independent Components of the REM Cycle

Ultrashort Sleep-Wake Cycle: Timing of REM Sleep. Evidence for Sleep-Dependent and Sleep-Independent Components of the REM Cycle Sleep 10(1):62-68, Raven Press, New York 1987, Association of Professional Sleep Societies Ultrashort Sleep-Wake Cycle: Timing of REM Sleep. Evidence for Sleep-Dependent and Sleep-Independent Components

More information