HST 583 fmri DATA ANALYSIS AND ACQUISITION Neural Signal Processing for Functional Neuroimaging Neuroscience Statistics Research Laboratory Massachusetts General Hospital Harvard Medical School/MIT Division of Health, Sciences and Technology HST.583: Functional Magnetic Resonance Imaging: Data Acquisition and Analysis Harvard-MIT Division of Health Sciences and Technology Dr. Emery Brown
Outline Spatial Temporal Scales of Neurophysiologic Measurements Neural Signal Processing for fmri Signal Processing for EEG in the fmri Scanner Combined EEG/fMRI Conclusion
THE STATISTICAL PARADIGM (Box, Tukey) Question Preliminary Data (Exploration Data Analysis) Models Experiment Model Fit (Confirmatory Analysis) Goodness-of-fit not satisfactory Assessment Satisfactory Make an Inference Make a Decision
Spatio-Temporal Scales EEG + fmri
Neurons Kandel, Schwartz & Jessell
Action Potentials (Spike Trains) Neuron Stimuli
2. SIGNAL PROCESSING for fmri DATA ANALYSIS Question: Can we construct an accurate statistical model to describe the spatial temporal patterns of activation in fmri images from visual and motor cortices during combined motor and visual tasks? (Purdon et al., 2001; Solo et al., 2001)
What Makes Up An fmri Signal? Hemodynamic Response/MR Physics i) stimulus paradigm a) event-related b) block ii) blood flow iii) blood volume iv) hemoglobin and deoxy hemoglobin content Noise Stochastic i) physiologic ii) scanner noise Systematic i) motion artifact ii) drift iii) [distortion] iv) [registration], [susceptibility]
Physiologic Response Model: Block Design
Physiologic Model: Event-Related Design
Physiologic Response: Flow,Volume and Interaction Models 1 Flow Term 1 Volume Term 0.5 0.5 0 0 20 40 60 80 100 120 0 0 20 40 60 80 100 120 1 Interaction Term f a =1 f b =-0.5 0.5 0 0 20 40 60 80 100 120 f c =0.2 0.6 0.4 0.2 0 Modeled BOLD Signal -0.2 0 20 40 60 80 100 120
Scanner and Physiologic Noise Models
fmri Time Series Model Baseline Activation Drift AR(1)+White x () t = m + b t+ s () t + v () t P P P P P Activation Model t P = time, = spatial location s P ( t-d ) = (base +Blood O stimulus) p O 2 (base +Blood volume stimulus) vol 2 IR IR
β 2σ β Correlated Noise Model Pixelwise Activation Confidence Intervals for the Slice β β + 2σ β
Signal Processing for EEG in the fmri Scanner How can we remove the artefacts from EEG signals recorded simultaneously with fmri measurements? (Bonmassar et al. 2002)
Ballistocardiogram Noise 150 Outside Magnet 100 EEG signal (uv) 50 0-50 -100-150 0 1 2 3 4 5 6 7 8 9 10 Time (sec) 150 Inside Magnet 100 EEG Signal (uv) 50 0-50 -100-150 0 1 2 3 4 5 6 7 8 9 10 Time (sec)
Faraday s Induced Noise v B φ ε = N t A Fundamental Physical Problem w/ EEG/fMRI: Motion of the EEG electrodes and leads generates noise currents! Machine Motion helium pump, vibration of table, ventilation system Physiological Motion heart beat (ballistocardiogram), breathing, subject motion
Noise vs. Signal... The Noise: Ballistocardiogram: >150 µv @ 1.5T in many cases Motion: > 200 µv @ 1.5T The Signal: ERPs: < 10 µv, reject epochs if > 50 µv Alpha waves: < 100 µv
Adaptive Filtering Use a motion sensor to measure the ballistocardiogram and head motion Place near temporal artery to pick up ballistocardiogram Use motion signal to remove induced noise
Adaptive Filter Algorithm Observed signal True underlying EEG y ( t) = s( t) + n( t) Induced noise Linear time-varying FIR model for induced noise n( t) = N 1 k = 0 w t ( k) m( t k) Motion sensor signal FIR kernel
Data 5 subjects Alpha waves 10 seconds eyes open, 20 seconds eyes closed over 3 minutes Visual Evoked Potentials (VEPs) Motion Head-nod once per 7-10 seconds for 5 minutes Added simulated epileptic spikes
Results: Alpha Waves
Results: Alpha Waves Outside Magnet
Results: Alpha Waves Eyes Closed Eyes Open 35 30 35 30 Frequency (Hz) 25 20 15 10 Frequency (Hz) 25 20 15 10 5 0 0 20 40 60 80 Time (sec) Before Adaptive Filtering 5 0 0 20 40 60 80 Time (sec) After Adaptive Filtering
COMBINED EEG/fMRI What are the advantages to combining EEG and fmri?( Liu, Belliveau and Dale 1998)
Combined EEG/fMRI Combines high temporal resolution of EEG with high spatial resolution of fmri Applications Event related potentials EEG-Triggered fmri of Epilepsy Sleep Anesthesia
The Sequence used in Simultaneous EEG/fMRI Stimulus Presentation 15 sec of 4-8 Hz Checkerboard Reversal 15 sec of fixation 15 sec of 4-8 Hz Checkerboard Reversal 15 sec of fixation fmri trigger fmri Window 30 sec EEG/VEP Window 30 sec EEG trigger 100 msec RT TO Time
Combining EEG and fmri (A) fmri regions of activation for 2 subjects. The fmri activity was consistently localized to the posterior portion of the calcarine sulcus. (B) Anatomically constrained EEG (aeeg). The cortical activity was localized along the entire length of the calcarine sulcus. (C) Combined EEG/fMRI (feeg). The localizations are similar to the fmri results and considerably more focal than the unconstrained EEG localizations
Spatiotemporal Dynamics of Brain Activity following visual stimulation
Cortical activations changes over time Seven snapshots of the cortical activity movie, without and with fmri constraint. The peaks of activity occur at the same time for both the EEG (alone) localization and the fmri constrained localization. Spatial extent of the fmri constrained EEG localization is more focal than the results based on EEG measurements alone.
Conclusion Well Poised Question Careful Experimental Design/Measurement Techniques Signal Processing Analysis Is An Important Feature of Experimental Design, Data Acquisition and Analysis. Data Analysis Should Be Carried Out Within the Statistical Paradigm.