Head Movement Compensation for Pediatric MEG Data using Signal-Space Separation (SSS)Dan WehnerHarvard-MIT Division of Health Sciences and TechnologyMGH/MIT/HMS Athinoula A. Martinos Center for Biomedical Imaging2/23/07Overview• Quantitative assessment of head movement in children• Dipole localization error due to head movement• SSS-correction and Equivalent Current Dipoles (ECD)• SSS-correction and Minimum Norm Estimates (MNE)• Practical use of continuous HPI and SSS toolsCollection of evoked MEG data with children• Large heartbeat artifact in data due to closer proximity of heart to MEG sensors• Large number of blink-related trial rejections• Small head size compared to the sensor array• Head movement during long (~1hr) cognitive taskHead position inside MEG helmetAdultChildMEG recording• 20 children (8-12 years old)• MEG signals recorded continuously– Sampling rate: 601 Hz– Filter 0.03-200 Hz• Averaged offline with -100 to 800 ms epoch• Rejection criteria:– >150µV in EOG, > 500 fT in gradiometers• Baseline corrected (100 ms pre-stimulus), and low pass filtered at 40 Hz• Low amplitude sinusoidal currents were fed to 4 HPI coils positioned on subject’s headAuditory Oddball Task• Subjects listened to a stream of standards interspersed with occasional deviants1000 standards with 100 deviants of each condition (8% each)/pat/ /cat//bat/ /rat/standard deviant deviant deviantAssessment of Head Movement• Position and orientation of head computed every 200 ms• Translation (x, y, z) and displacement vectors of head motion were segmented into 10-second bins• Average and variance of head motion in each time bin was calculatedRelative head motion for one childSummary for all subjects• Average displacement of head: 12 mm (range 3-26 mm)• Average variance – x: 3.7 mm, y: 15 mm, z: 23 mm• More variation in head movement for z-direction relative to x- and y-directions (p < 0.01, p < 0.05)• More variation in head movement for y-direction relative to x-direction (p < 0.01)Dipole localization error due to head movement• MRI (TR = 2530 ms, TE = 3.25ms, flip angle = 7º, voxel size = 1.3 x 1.0 x 1.3 mm3)• Source space with ~7000 sources (5 mm spacing)• Forward model calculated using BEM and initial head position• Additional forward models at 1 second intervals using HPI information• All forward solutions treated as simulated data– Current dipoles fit to field patterns (mne_dipole_fit)– Sensor positions in all cases corresponded to initial HPI measurementSimulated Dipole ResultsLH RHLateral Lateral MedialMedial1062mmLH RHLateral Lateral MedialMedial1395mm25155mm604020mmStandard DeviationMeanMaximumSimulated Dipole Results cont.Another SubjectLHRHMaxMeanSt. DevMovement Compensation with SSS V (r) =αlmYlm(θ,ϕ)rl +1m=−llÂ+βlmrlYlm(θ,ϕ)m=−llÂl =0•Âl =0•Â B = −µ0∇V ,∇2V = 0 αlm,βlm= scalars, r = r φ=αlmalm+βlmblmm=−llÂl =1LoutÂm=−llÂl =1LinÂ1)2)3) φ= Sx = SinSout[ ]xinxout      Sin= a1,−1...aLin,Lout[ ],Sout= b1,−1...bLin,Lout[ ]xin=α1,−1...αLin,Lout[ ]T,xout=β1,−1...βLin,Lout[ ]T4)From Taulu et al., 2005Movement Compensation with SSS• Calculate harmonic amplitudes attached to head (from continuous hpi)• Model movement of subject as movement of sensor array• Calculate signals in a virtual array locked to the subject’s head ˆ x =ˆ x inˆ x out      = Sφˆ φ in= Sinˆ x in†Equivalent Current Dipole (ECD) modeling• N100m response to repeated stimulus “pat”• Spherical source model with origin at (x = 0, y = 0, z = 40 mm)• SSS applied two ways– Within each run– All data transformed to head position at beginning of experiment• Single dipole fit to N100m response in each hemisphere with and without SSS-correction• Dependent variable: increase in GOF of the fitted ECD after SSS-correctionMEG sensor data for one childMean change in locationLH: 5 mmRH: 5.6 mmMean increase in GOFLH: 1.52% (p<0.01)RH: 0.97 % (n.s.)Increase in GOF in LH: 15/20 subjectsRH: 14/20 subjectsMinimum-Norm Estimates and SSS• Forward solution was computed using– Initial head position from run 1 only– Average of head positions from each of 5 runs– SSS-correction within each run– SSS-correction to head position at beginning of experiment• Dependent variable: difference in mean MNE amplitude within a small cortical patch surrounding peak N100m responseResults for one childMNE Results for one childSeries of planned pairwise t-tests showed that peak N100m response inLH: SSS-entire expt. > SSS-each run = uncorrected-allruns > uncorrected-run1RH: SSS-entire expt. > SSS-each run > uncorrected-allruns > uncorrected-run1 Group statistics:Discussion• Averaging forward solutions from several runs gives a more robust signal than using the forward solution from run 1 alone (Uutela et al., 2001)• SSS-correction further sharpened the N100m response significantly– SSS-correction to the head position at the beginning of the experiment gave the largest N100m valuesConclusions• SSS is an effective way to compensate for head movements in children during MEG recordings• Sharpening of brain responses resulting in higher effective SNR may be crucial for some cognitive expts. if choice of stimuli and/or attention of child is limitedUsing Continuous HPI • Attach HPI coils as usual• Take initial HPI measurement• After starting MEG recording, run the script ./neuro/dacq/bin/cont_hpi_on– Continuous hpi will start– You can filter it out of the display by applying LPF to display• After finishing MEG recording, run the script ./neuro/dacq/bin/cont_hpi_off• The continuous HPI measurements will be stored in the raw measurement filesUsing the SSS program MaxFilter• Software located on megws1, megws2, megws3• command line $NEUROMAG_ROOT/bin/util/maxfilter where $NEURMAG_ROOT is /space/orsay/8/megdev/megsw-neuromag• Run MaxFilter on raw data files– Very important to mark bad channels in raw file– maxfilter -f <infile.fif> -o <outfile.fif> -v -movecomp -hp– maxfilter -help • Average data as usual• Make sure you look at data in xplotter before making MNE/dSPM movies!• Manual located at $NEUROMAG_ROOT/manuals/MaxFilter.pdfAcknowledgmentsThesis SupervisorsMatti HämäläinenSeppo AhlforsElfar AdalsteinssonMEG Tech SupportDeirdre FoxeDan WakemanMody Lab MembersSurina BashoCherif SahyounMaria ModyNeuromag GroupJukka