Home

MaxFilter 2.0 Manual

1. ENMO NM system channels 306 204 outfile infile_quat fif or infile_mc fif origin fit to isotrak or 0 0 40 mm frame head psinv direct iterative st off off regularize on on hpistep ms 200 200 hpisubt amp amp Ipfilt off off Table 4 2 2006 10 31 The default parameter values of MaxMove are summarized in Table 4 2 NM21993A CHAPTER 5 MaxAve parameters 5 1 Background Program MaxAve is an off line version of the on line averager see Neuro mag Data Acquisition User s manual Section 3 5 On line averaging MaxAve can be applied to raw data files either before or after Maxwell fil tering The program reads the acquisition parameters from the input FIFF file and repeats averaging in off line mode Program MaxAve can be started from the GUI or from a terminal window command line The GUI has also controls for setting and changing the rejection limits Other parameter values set during data acquisition can be changed by setting and editing a parameter file 5 2 Launching the program NM21993A You can access MaxAve by starting the GUl as explained in Section 2 2 on page 12 You can select the input and output filenames and set rejection limits from the menu Averager Load raw data Select the raw data file to be averaged Output file Select the file for saving the averages Rejection limits Change averaging rejection limits When you have selected the input filename the GUI main dis
2. Note If the origin is set outside of the sensor array or closer than 5 cm to the nearest sensor the program reports an error and terminates 3 2 Harmonic basis functions MaxFilter evaluates the harmonic basis functions for all MEG sensors The default orders of the harmonic expansions are set to L 8 and Loy 3 Which have turned out sufficient in most practical applications see the publications listed on page 7 You can change the expansion orders by selecting Dimensions from the Filter menu Figure 3 2 F Expansion orders Inside order El Outside order sal OK Cancel Help Figure 3 2 Setting the expansion orders On command line you can change the orders with the options in L and out Lout 22 2006 10 31 NM21993A MaxFilter parameters Ej NM21993A The total number of multipole components becomes 2 a M L 1 Loy4 1 2J and must not exceed the number of good MEG channels In practice the largest values of L and L are limited for avoiding numerical instabil ities out Note If L lt 5 L gt 11 Lout lt 1 or Lout 5 the program reports an error and terminates After computing the harmonic expansion bases the program estimates the theoretical signal to noise ratio SNR of each multipole component To avoid increasing of the noise virtual channels with smallest SNRs are neglected see Section A 4 on page 47 for details The SNRs of the harmon
3. 30 Reconstruction of sensor signalS ooooocccooooo 31 3 6 Default parameters 0002 cee ee 32 Chapter 4 MaxMove parameters 33 4 1 Data transformation 0 eee 33 4 2 Head position tracking 00 cece eee eee eee 35 4 3 Head position estimation ooooooooom m 37 4 4 Movement compensation 00ee eee eee eee 38 4 5 Low pass filtering 000 cece eee eee eee 39 4 6 Default parameters 0 020 c eee eee 40 Chapter 5 MaxAve parameters 41 5 1 Background vi e cena wile ae eee we Nese fe ee 41 5 2 Launching the program 00 eee 41 5 3 Selecting files in is A hea ve 43 5 4 Averaging parameters 00 cee eee eee eee 43 Appendix A Maxwell filtering in a nutshell 45 A 1 Signal space separation 00 cece eee eee eee 45 A 2 Harmonic amplitudes 200 cee eee eee eee 46 A 3 Pseudoinverse 0 dd ee ewe ead da 47 A 4 Optimization of virtual channel selection 47 Appendix B Elekta Neuromag MEG sensors 49 B 1 Sensor typeS ix 030 toa Ee AA 49 B 2 Scaling between magnetometers and gradiometers 50 B 3 Manipulation of sensor types 2 200 e eee eee 51 Appendix C Temporal subspace projection 53 Appendix D Head position estimation 55 D 1 HPI signals occ foc ce ee a a 55 D 2 Coordinate matching 000 eee ee eee eee 56 D 3 HPI Channels coria rs aaa ae 57 D 4 Head position file format o oooo
4. MaxFilter fits to isotrak data or to sensor locations 2 Inside expansion order 8 80 terms outside expansion order 3 15 terms Automated bad channel detection on using limit 7 0 Fine ealihratinn an neina fnenrnidatahacec acciaas cal dat t 10 058 Head movement Fitting error 0 90 mm No warnings 21 Execute STOP No errors Processing 34 data buffers time 20 0 s 4 Running MaxFilter please wait Select File gt Show log to view details 5 k a Figure 2 2 MaxFilter GUI main window The menubar A text area for showing current parameter values Drawing area which shows the head position estimation parameters Display area for showing the processing status A message text label at the bottom of the window AA A The menus and controls are described in the following sections When you have defined the input and output filenames and modified the parameters press Execute to start Maxwell filtering The progress scale bar indicates the number of processed buffers You can press STOP to cancel processing Then you can modify the parameters and try again press Execute The labels next to STOP button indicate the number of warnings and errors reported by MaxFilter During execution the background colour of these labels is green If there are warnings background of the label warnings changes to red and the label reports the number of warnings If MaxFilter is terminated due to
5. View manual Starts the Acrobat reader program to view this manual On version This item shows the current version and compilation date of Max Filter MaxAve and MaxFilter_GUI NM21993A 2006 10 31 15 Using MaxFilter A e 16 2 5 Loading data MaxFilter accepts evoked or raw data FIFF files as input New data are loaded by selecting Load data from the File menu Figure 2 3 Besides file selection the dialog has a control button for parameter settings As default upon selecting a new input file the program cleans all parameters values and resets the GUI dialogs If you press the button Reuse the previ ous settings the program applies the parameter values that were applied for the previous file Select file to load Filter nttaavi root data chpi phantom 05 1006 fif Directories Files 051006 dip04_chpi_still fif q 0 dip04_nochpi fif dip4_chpi_3pos_sss fif Z 051 f n 051006 dip4_chpi_moving fif dip4_chpi_rot fif dip4_chpi_rot2 at1317 fif dip4_chpi_rot2 fif ai P _ Reuse the previous settings Current file tta chpi phantom 051006 dip4_chpi_3pos fil OK Filter Cance Help Figure 2 3 Data loading dialog When the input data are successfully loaded the working directory is changed to the directory containing the file You can change the working directory by selecting Set directory from the File menu All processed files wi
6. limit should be between 0 9 and 1 MaxMove options trans trans_file fif default Transforms MEG data in input_file fif into the sensor array defined by the channel info and transformation in trans_file fif or into default head position movecomp inter Estimates head positions and compensates head movements in a continuous raw data file Option inter defines what to do when con tinuous HPI is off headpos Estimates head positions and stores head position parameter data but does not compensate head movements 2006 10 31 NM21993A Command line arguments g hp pos_file txt Stores the estimated head position parameters in a separate ascii file hpistep n Sets the head position estimation interval in ms hpisubt amp base off Sets the HPI signal subtraction method hpicons Tries to improve the consistency between isotrak points and initial HPI fitting results linefreq lf Sets the basic line interference frequency 50 or 60 H2 lpfilt corner Applies low pass IIR filtering to remove residual HPI coil signals Other options site sitename Tries to load the fine calibration and cross talk correction data from the default directory e g neuro databases sss neuro databases ctc files sss_cal_sitename dat and ct_sparse_sitename dat cal sss_cal_file dat off Applies the fine calibration data defined in file sss_cal_file dat or switches the fine calibration off ctc ct_matrix_file fif
7. the program can identify and suppress artifact waveforms which arise in the Sy space Transformation of MEG data between different head positions By transforming the component b into harmonic amplitudes i e virtual channels MEG signals in a different head posi tion can be estimated easily Compensation of disturbances caused by head movements By extracting head position indicator HPI signals applied continuosly during a measurement the data transformation capability is utilized to estimate the corresponding MEG sig nals in a static reference head position 8 2006 10 31 NM21993A Introduction E NM21993A 1 4 Software safety MaxFilter is easy to apply and the default settings provide good results in most cases Maxwell filtering as implemented in this program is an irre versible operation The expansion corresponding to the space outside of the sensor array b is discarded before saving the result In addition MaxST projects out artifact waveforms identified in the sensor space Note The original recording cannot be reconstructed from the result FIFF file Therefore it is very important to keep also the original data files on suitable backup media after applying MaxFilter This manual contains important hazard information which must be read understood and observed by all users General limitations of the program are included in the following Chapters For your convenience all warnings that appear in the manual a
8. Basic Maxwell filtering input_file_sss fif e MaxST input_file_ssst fif e Head position estimation input_file_quat fif e Movement compensation input_file_mc fif If you want to set the output filename manually you can enter the desired name in the file selection dialog text field Output file 2006 10 31 17 El Using MaxFilter 18 2 6 2 2 6 3 2 6 4 On command line you can define the name of the output file with the option o output_file fif Output is not generated 1f the output file already exists In that case you should either remove the old file set a new output filename or select Force to ignore warnings Data packing The result of the program is saved in a FIFF file You can define the data packing with the option format type where type can be short 16 bit short packing float 32 bit float packing or long 32 bit integer pack ing Note If you don t select data packing 1 Raw data files are saved in float format Therefore the result file becomes twice as large as the input file if the original data were packed as short 2 Evoked data are packed in the same format as the input file Changing of data packing may be needed for example in processing 32 bit data acquired with an Orion system Neuromag Data Analysis Soft ware release 3 3 and earlier cannot process data stored with 32 bit inte ger packing Therefore the data packing needs to be short or float when saving the result file
9. and T isa 3x 1 translation vector Matching of the point sets is done by using the analytic solution based on the quaternions presented in P J Besl and N D McKay A Method for Registration of 3 D Shapes IEEE Trans Patt Anal Machine Intell 14 239 255 1992 Altogether seven quaternion parameters qo q are needed to define the transformation between the coordinate systems Using unit quater nions reduces the number of independent parameters to six because do 9 9 93 1 Parameters qp q3 define the rotation matrix and parameters q4 qs q give the translation The relation between the quaternion parameters and coordinate transformations is 2 2 2 2 do 41 24 93 2 414 9093 2 9193 9092 2 2 2 2 R 2 9192 9093 qo 42 41 43 2 q4243 4041 2 2 2 2 2 4143 4042 2 9 93 9091 qdo 43 91 A a JT T La 75 46 2006 10 31 NM21993A Head position estimation E D 3 HPI channels When MaxMove options are in use the program estimates the head posi tion parameters and saves them as 9 new raw data channels with the names QUAT001 CHPI001 quaternion parameter q1 QUAT002 CHPI002 quaternion parameter q2 QUAT003 CHPI003 quaternion parameter q3 QUAT004 CHPI004 quaternion parameter q4 QUAT005 CHPI005 quaternion parameter q5 QUAT006 CHPI006 quaternion parameter q6 QUAT007 CHPI007 goodness of fit QUAT008 CHPI008 HPI estimation error QUAT009 CHPI009 estimated movement velocity
10. dip04 chpi_rot raw mc ff OK Filter Cancel Help Figure 5 2 Data selection You can set the name of the output file by selecting Output file from the Averager menu If you do not specify an output filename the program tries to save the results in the file named as input_file_ave fif where input_file fif is the name of the raw data file 5 4 Averaging parameters Averaging variables are defined in the file NEUROMAG_ROOT setup maxave maxave vars Normally the program extracts the parameters from the input FIFF file and saves them in a temporary file Averaging is performed with these parameter values and the temporary parameter file is automatically cleaned after saving the averaged data 2006 10 31 43 5 MaxAve parameters 44 If you press the button Save parameters to file you can enter a filename e g ave par where the program writes in ascii format all data acquisiton parameters extracted from the input file You can then open and edit this file in a text editor and save it before running MaxAve Please see Neuro mag Data Acquisition User s manual Section 3 5 On line averaging for the description of the averaging parameters Thereafter you can apply the changed parameter values using GUI The program automatically loads the file and reads the modi fied values before execution Command line Apply option par ave par If you need to change only the rejection limits you can sele
11. fif le fif ctc only o orig fif maxfilter f infi maxfilter f infi Thereafter you can overlay the output files and compare the differences This kind of comparison is inherently utilized in bad channel detection Section 3 3 and in MaxST Section 3 4 Option reconst does not utilize the SNR optimization Section 3 2 3 6 Default parameters The default settings for Maxwell filtering are summarized in Table 3 1 ENM NM system NM 122 channels 306 204 122 outfile infile_sss fif origin fit to isotrak or 0 0 40 mm frame head badlimit 7 7 7 La 8 8 8 Lout 3 3 2 CTC yes no no CAL yes no no psinv direct iterative iterative st off off off regularize on on off Table 3 1 meuro databases ctc ct_sparse Sif meuro databases ct_sparse fif 2006 10 31 NM21993A CHAPTER 4 MaxMove parameters 4 1 Data transformation NM21993A Direct comparison of different MEG measurements is very difficult even if the data were acquired with the same device This applies both to data from the same subject measured several times or data from several sub jects In addition movements of the patient s or subject s head during the recording cause distortions in the MEG signals As described in Sections 1 2 and A 2 MaxFilter transforms the MEG data into harmonic function amplitudes which can also be interpreted as virtual channels MaxMove utiliz
12. fit to slow background disturbances to separate their contribution from the HPI signals Also line frequency signals are modelled if the tag FIFF_LINE_FREQ has been included in the raw data file You can control how to subtract the HPI signals before Maxwell filtering The choices are Amplitude Subtract the sine amplitudes Baseline Subtract both the sine amplitudes and the linear baseline OLE Do not subtract HPI signals Note HPI signal subtraction with the choice Baseline removes DC and slow frequency components lt 1 Hz Thus it acts as a high pass filter The corner frequency is determined by the HPI estimation interval shorter step results in higher corner frequency On command line you can select the option headpos if you want just to estimate the head positions You can set the HPI amplitude estimation length with the option hpistep n where n is the interval in millisec onds You can also include the line frequency and 1ts harmonics with the option linefreq lf You can control the HPI signal subtraction with the option hpisubt amp base off where the argument refers to one of the three selections above You can also save the estimated head positions in a separate ascii file with the option hp filename 2006 10 31 NM21993A MaxMove parameters EJ NM21993A 4 3 Head position estimation Note MaxFilter assumes that the normal HPI fitting is done during the data acquisition The program reports an erro
13. head frame The quaternion parameters are saved as extra channels in the result FIFF file Details of the quaternion matching are presented in Appendix D The command line option hpicons tries to reduce the initial mis match The program defines then new coordinate transformations by omit ting one coil at time The transformation giving smallest mismatch is selected and the isotrak point of the mismatching coil is adjusted accord ingly This option is useful for example if the attachment of an HPI coil has moved after digitization 2006 10 31 37 El MaxMove parameters 38 4 4 Movement compensation If you also want to compensate head movements select Compensate movements and save with processed data The program searches first 1f the quaternion data channels are already included in the input file e g if the head positions were estimated earlier If they are not found the head positions are estimated as presented above MaxFilter transforms the data to a static reference head position and at the same time performs the nor mal software shielding When processing a raw data buffer the program performs the data trans formation only if the HPI fitting is successful If the HPI fit failed a warn ing is given and the results are skipped from the output The output data during failed HPI fits shows zero values in all MEG channels Often head movements may induce other disturbances such as slowly varying fields due to magneti
14. in Appendix B Table B 1 The Elekta Neuromag system combines two types of sensors and an appropriate scaling between them has to be applied before combining the data for Maxwell filtering The default scaling factor between magnetom eter and gradiometer channels is 100 It can be changed with the com mand line option magscale see Appendix B 2 Sometimes it may be necessary to manipulate the sensor types with the command line options T2 or T3 for backward compatibility with older versions of data analysis programs Appendix B 3 Sensor noises Often sensor noise levels are used in source modelling studies The noises can be defined as standard deviations from the baseline during a specified time window In the absence of brain signals empty room data the noise values are typically uncorrelated and obey normal distribution Maxwell filtering however modifies the sensor noise properties and the baseline noises may become correlated Therefore statistical parameters such as confidence intervals and volumes are incorrect if analysis software uses sensor noises estimated from the baselines Note Correlations of the sensor noises must be taken into account if the Maxwell filtered sensor noises are applied in source modelling Calibration adjustment and cross talk correction Maxwell filtering can be applied to improve the standard calibration of MEG systems The adjustment includes accurately defined sensor orienta tions and
15. j 1 where P is the number of coils az j is the amplitude at sensor k from the jth coil p are respectively the angular frequency and phase of the jth sine signal and n is the noise The phase can be represented as a lin ear combination of a sine and a cosine term When N data samples have been collected with sampling frequency f 1 T the data can be pre sented in matrix form as The signal amplitudes fed to the coils are represented by the N x 2P matrix A sin iT yo cos _ piT and x is the amplitude vector of the coils The amplitudes can be esti mated in the least squares sense as al x AA A b Besides sin and cos terms for the P coils this model can be augmented to include the contribution of interfering magnetic fields Line frequency and 2006 10 31 55 D Head position estimation 56 its harmonics can be easily included e g 0 50 100 300 if the basic line frequency is 50 Hz and low pass filtering frequency is 300 Hz Furthermore low frequency disturbances can be modelled as where r is the slope and d constant offset fitted by linear regression to input data b t of channel k D 2 Coordinate matching Let the point set y pj 1 P represent the digitized head frame coordinates of the M HPI coils and x 7 1 P represent the same coordinates in the device frame The relation between the point sets can be expressed as y Rx yt T where R isa 3 x3 rotation matrix
16. read o output_file fif Defines the FIFF file where the results are written origin x0 y0 z0 Sets the origin of the expansions to the point x0 y0 z0 in the selected coordinate frame x0 y0 z0 must be given in mm frame device head Sets the coordinate frame to device or head The origin coordinates are given in this frame in Lin Sets the order of the expansion for b SOME class Sets the order of the expansion for b yg bad bad_chl bad_ch2 bad_chn Marks static bad channels bad_chn refers to the logical channel number e g 0741 1842 2623 2006 10 31 59 E Command line arguments 60 autobad on off nraw Switches the automated bad channel detection on or off or sets the number of tags to be scanned from the beginning of raw data file badlimit value Sets the standard deviation threshold of the automated bad channel detection skip tl_start tl_end tn_start tn_end Skips segments of raw data The skip intervals are given as pairs of time points in seconds from the start of the file format short float long Sets the data packing format in the output file torce Bypasses the warnings and error messages def Lists the default parameter values maint Lists maintenance options MaxST options st buflen Applies the spatiotemporal MaxST optionally sets the raw data buffer length in seconds corr limit Changes the subspace correlation threshold for MaxST
17. to select the head position estimation without compensation or the movement compensation of all data blocks in the input file see Section 4 4 Head movements m Head movement Estimate and save with original data F Compensate movements and save with processed data Save head position parameters in a separate file HPI estimation Head position estimation interval 200ms HPI signal subtraction Amplitude OK Cancel Help Figure 4 2 Setting head movement estimation and compensation If you select Estimate and save with original data the program estimates and subtracts the HPI coil signals estimates the head positions and saves the quaternion channels However the program does not perform any Maxwell filtering operations but writes the original data after HPI signal subtraction If you wish to save the head position parameters in a separate ascii file choose Save head position parameters in a separate file the format is shown in Appendix D 4 When processing a raw data buffer the program first estimates the ampli tudes of each HPI coil Sinusoidal signals of the HPI coils are estimated and subtracted from the measured data see Appendix D for details You can also set the HPI signal extraction and head position estimation inter val the choices are 200 500 and 1000 ms 2006 10 31 35 El MaxMove parameters 36 Besides sinusoidal HPI signals the program also performs a linear
18. v and represent the harmonic basis function values N Lp M L and the vectors X x contain the har monic amplitudes out The amplitudes x can be estimated from measured signals b as in Xout x in Sib out where St is the pseudoinverse of S see the next Section The signals b can then be separated as b b bour Din SinX in in Doy 8 X o out out Suppression of external interference can be performed by leaving out the contribution b yt The harmonic amplitudes Qt of the inside expansion can be interpreted as virtual channels which are independent of the sensor array and head position Optimization of the virtual channel selection is described in Sec tion A 4 46 2006 10 31 NM21993A Maxwell filtering in a nutshell A NM21993A A 3 Pseudoinverse The pseudoinverse St can be calculated with standard numerical proce dures 1 ss The condition number of St describes the stability Typically the condi tion number for the Elekta Neuromag sensor array is below 500 It may however increase dramatically e g if the magnetometer channels are omitted Therefore we have developed an iterative method to improve the stability Instead off all expansion terms together the iterative pseudo inverse is composed of separate block matrices 1 Sk SiS Sy where subscript k indicates that the submatrix contains only the expansion terms of the kth order k 1
19. 006 dip04_nochpi fif dip4_chpi_3pos fif dip4_chpi_3pos_sss fif dip4_chpi_moving fif dip4_chpi_rot fif dip4_chpi_rot2 at1317 ff 5 dip4_chpi_rot2 fif E i x 2 4 Transform to default head position Current transformation file ata chpi phantom 051006 dip04_chpi_stillfi ok Filter Cancel Help Figure 4 1 Selecting the data transformation options The data transformation does not require continuous head position indica tor HPI signals during recordings Thus it can be applied to all Elekta Neuromag systems even if continuous head position tracking was not uti lized Warning When transforming data from one head position into another special care should be taken to ensure that both head positions are accu rately defined in respect to the sensor array 34 2006 10 31 NM21993A MaxMove parameters EJ NM21993A 4 2 Head position tracking During the recording the head position has to be tracked by feeding con tinuous sinusoidal signals to 4 5 head position indicator HPI coils see Neuromag Data Acquisition User s manual Section 6 2 Head position indicator Head position tracking can be done only if the continuous HPI was applied during the recording Old hardware e g Neuromag 122 may however not support the continuous HPI When you want to estimate the head positions select Head movement compensation from the Move menu Figure 4 2 The box Head movement can be used
20. 1l m The columns of V and V span the waveforms of b t and b t Subspace intersection between the waveforms can be found with the QR decomposition V in Oink in gt Vs OR mxm where oO and RER from the nxm matrices Q O as A mxm matrix C is composed T C QnQ The singular values o of matrix C define the principal angles 0 between the two subspaces cos 0 0 The intersection of the sub spaces contains the waveforms corresponding to 6 1 In practice the subspace correlation limit in MaxFilter is set to 0 98 direct pseudoin verse or 0 995 iterative pseudoinverse 2006 10 31 53 g Temporal subspace projection If there are p principal values exceeding the correlation limit the program composes a projection operator 1 LL where L nx p contains the intersecting waveforms first p columns of V These waveforms are finally projected out as 2 T bin I LL b 54 2006 10 31 NM21993A APPENDIX D Head position estimation D 1 HPI signals NM21993A During the recording the head position has to be tracked by feeding con tinuous sinusoidal signals to 4 5 head position indicator HPI coils Typi cal frequencies of the signals are 154 158 162 166 and 170 Hz for Elekta Neuromag and Neuromag System or 293 307 314 321 and 328 Hz for new systems with the Orion electronics The resulting magnetic field on sensor k is then P b t Y a jsin t 9 n
21. 3 8 13 0 3 mm 25 324 0 0 8 13 0 3 mm 10 81 8 13 0 0 0 3 mm 10 81 Table B 1 MEG coil descriptions B 2 Scaling between magnetometers and gradiometers 50 The Elekta Neuromag data contains both magnetometer and planar gra diometer channels which have different SI units T and T m respec tively Therefore some scaling between the channel data and basis functions has to be applied before combining them for Maxwell filtering MaxFilter utilizes the inherent RMS noise levels of the sensors nn 3 fT sqrt Hz for magnetometers Ng 3 fT cm sqrt Hz for gradiometers The basic modelling equation b Sx see Appenix A 2 is presented in the form diag w b diag w Sx where the weight for the kth channel is w n n Thus the gradiom eter channel weight becomes 1 while the magnetometer channel weight is 100 Optionally the weight w for magnetometers can be set with the com mand line option magscale mult the gradiometer weight is still one 2006 10 31 NM21993A Elekta Neuromag MEG sensors B NM21993A B 3 Manipulation of sensor types Most Elekta Neuromag arrays consist of type T3 sensors Table B 1 Older versions of the data acquisition software may have however marked them as corresponding type T2 sensors In practice the only difference between types T2 and T3 is that the magnetometer sizes are different 25 8 mm for T2 21 mm for T3 MaxFilter applies the correct magne to
22. 7 0 1 O0 784 1 2 O0 g 1 000 err 1 2mm 1 000 0 003 0 014 0 0 783 05 0 2 80 3 05 g 0999 err 19mm 0 002 0 998 0 070 786 06 0 0 79 1 0 3 0 0 g 0 983err 1 0mm 0 014 0 070 0 997 0 4 78 6 29 27 793 15 g 1 000err 50mm 0 7 3 9 0 9 t 5 79 fe 0 14 cm g 0 99 v 0 19 cm s r 0 11 rad s d 0 10 cm JNo need to interpolate r 1 9 mm for samples 3420 3539 12 Warnings log Warning Trying to overwrite file data chpi phantom 060329 dip04_chpi_rot_raw_mc fif i a EL Execution command Yneuro bin util maxfilter gui f data chpi phantom 060329 dip04_chpi_rot_raw fif o data ch E Jl J a Hide Print Clear Help Save Figure 2 5 MaxFilter logging window 2006 10 31 NM21993A CHAPTER 3 MaxFilter parameters 3 1 Expansion origin NM21993A The outcome of the expansions for b and b depends on the location of the coordinate system origin In general best results of Maxwell filter ing are expected when the origin is defined so that the S space in Figure 1 1 on page 6 covers the whole brain The program defines an opti mal origin by Fitting to isotrak data All digitized points excluding cardinal landmark points are searched and a sphere is fitted to these points The fit is improved by dropping the worst outlier points e g tip of the nose The origin is set to the fitted point in the head coordi nate frame Setting according to Source Mode
23. Channel names QUATXXX indicate that movement compensation has not been applied while the channel names are changed to CHPIXXX after the data have been transformed to the static reference head position D 4 Head position file format When the head positions are saved to an ascii file the file contains a header line and one row for each fitted time interval Time ql q2 q3 q4 q5 q6 g value error velocity 0 000 0 04553 0 00785 0 05864 0 00058 0 00372 0 00306 0 99959 0 00065 0 00000 0 200 0 04424 0 00772 0 06070 0 00089 0 00330 0 00317 0 99959 0 00064 0 00006 0 400 0 04433 0 00769 0 06066 0 00089 0 00331 0 00316 0 99959 0 00064 0 00005 0 599 0 04446 0 00773 0 06067 0 00090 0 00329 0 00317 0 99959 0 00065 0 00014 0 799 0 04442 0 00766 0 06070 0 00089 0 00329 0 00316 0 99958 0 00065 0 00008 0 999 0 04438 0 00765 0 06068 0 00090 0 00329 0 00316 0 99959 0 00065 0 00008 NM21993A 2006 10 31 57 El Head position estimation 58 2006 10 31 NM21993A APPENDIX E NM21993A Command line arguments This section lists all options of the command line program maxfilter neuro bin util maxfilter f input_file fif options Common options version Shows the version number of the program help Shows brief information of available options Switches on verbose logging This option also displays detailed pro cessing history of the input file f input_file fif Defines the FIFF file where the evoked or raw data are
24. Elekta Neuromag MaxFilter User s Guide Software version 2 0 October 2006 ELEKTA Copyright O 2006 Elekta Neuromag Oy Helsinki Finland Elekta assumes no liability for use of this document if any unauthorized changes to the content or format have been made Every care has been taken to ensure the accuracy of the information in this document However Elekta assumes no responsibility or liability for errors inaccuracies or omissions that may appear in this document Elekta reserves the right to change the product without further notice to improve reliability function or design This document is provided without warranty of any kind either implied or expressed including but not limited to the implied warranties of merchantability and fitness for a particular purpose Elekta Neuromag MaxFilter and MaxShield are trademarks of Elekta This product is protected by the following issued or pending patents US2006031038 Signal Space Separation US6876196 Head position determination WO02005067789 DC fields W0O2005078467 MaxShield FI20050445 MaxST Printing History Neuromag p n Software Date Sth edition NM21492A D 1 1 January 2005 Ist edition NM21993A 2 0 October 2006 NM21993A Contents Chapter 1 Introduction 5 el OVervie Ww lt a AA A Seas 5 1 2 Maxwell filtering sai art RA 6 1 3 Software functionality oooooooommmmmmmmo 8 1 4 Software safety ooooooooomonnnanan a
25. G measurement results The parameters needed in calculations are set to well defined initial values separately for Elekta Neuromag Neuromag System and Neuromag 122 data The default settings are listed in Sec tions 3 6 and 4 6 and you can run the program without changing these values However sometimes it may be useful to tune the details of the Maxwell filtering operation In brief you can change the dimensions of the internal and external multi pole bases and the origin of the expansions Optionally you can manually identify and set bad channels that are not taken into account in the recon struction In cases where the source of interference is located inside or very close to the sensor array it is recommended to use the spatio tempo ral Maxwell filtering MaxST In addition you can transfer data between different head positions and compensate disturbances due to head move ments using MaxMove Warning It is important that the user inspects both the input and the out put data visually to judge the quality of the Maxwell filtering result 2006 10 31 11 E Using MaxFilter 12 2 2 Launching the program 2 2 1 Graphical User Interface The graphical user interface GUI is a program that collects the input parameters of MaxFilter or MaxAve and then runs the command line pro gram as a subprocess You can monitor the execution on a log window Section 2 7 The GUI can be launched in HP UX 11 Click the Neuromag toolbo
26. Ignoring warnings It is possible to bypass the warnings and error messages using the button Force to ignore warnings or the command line option force The pro gram tries to continue execution even if warnings or nonfatal error mes sages are encountered The program is however terminated if the orders of the expansions are improper or if the origin is too close to the sensors Note Normally the program checks if the output FIFF file already exists and refuses to overwrite an existing file When ignoring warnings the pro gram tries to overwrite an existing file without asking the user Processing history When saving processed data MaxFilter updates the processing history block if it is found or creates a new processing history The block includes the Maxwell filtering parameter values and information about the cross talk and fine calibration correction see Section 3 5 3 2006 10 31 NM21993A Using MaxFilter An example of the processing history block 104 104 103 204 212 113 104 105 104 105 104 105 1055 105 blo dat sci cre d 264 263 265 266 267 268 269 it 103 204 113 800 3417 270 271 l 900 901 ck ID e entist ator program 502 SSS task SSS coord frame SSS origin SSS ins order SSS outs order SSS nr chnls SSS components 502 501 block ID date creator program CTC matrix proj item
27. a block while argument off means that automated detection is not applied Argument nraw scans first nraw data buffers Value nraw 1 is equivalent with autobad on You can also define the threshold with the option badlimit r Warning If the threshold of the automated bad channel detection is too small the program may classify good channels as bads and if it is too high some bad channels may remain undetected Evoked data Each evoked response set is treated separately i e the channel selection may vary from set to set If the program finds more than 12 bad channels a warning is printed and the execution terminates If the fine calibration is not in use it may happen that the autobad detection produces too many bad channels for any threshold values In such cases you should examine the input data to define the bad channels manually and repeat the program by disabling autobad and setting bad channels manually Alternatively you may apply MaxST Section 3 4 Raw data FIFF files containing raw data from a long recording may contain sections where the data quality has momentarily deteriorated e g due to bursts of artifacts on several channels In such a section the program may detect too many bad channels The execution is however not terminated but output MEG data on all channels are set to zero during such data blocks For raw data FIFF files the default nraw is 30 i e the program scans bad channels from the first 30 buffer
28. a fatal error background of the label errors turns to red 2006 10 31 13 El Using MaxFilter 14 The scale bar and the text label at the bottom indicate the status of the Maxwell filtering operation You can view more detailed information of the MaxFilter output and warnings by selecting Show log from the File menu When you start MaxMove processing see Chapter 4 the program starts head position estimation Estimated head postion parameters are drawn in the plotting area of the main window e white curve fitting error e red curve goodness of fit e green curve translational movement velocity e blue curve rotational velocity e yellow curve drifting from the inital head position If you click the left mouse button on the curves to select a timepoint the labels on the left side of the drawing area display the values at the selected time point Clicking of the right mouse button opens a dialog for control ling the vertical scales and for showing or hiding the curves 2 4 Menus 2 4 1 You can access any of the menu choices directly by first pressing the Alt lt underlined letter in the menu name gt and then lt underlined letter in the menu choice gt For example to select the Set directory item from File menu press A1t f followed by d The same procedure applies to menus found in other windows of the program as well File Load data Open a new file for processing Section 2 5 Output options Set th
29. aa i aa 9 Chapter 2 Using MaxFilter 11 2 1 BACK GFOUIG oa o a a 11 2 2 Launching the program ooocccooccc oo 12 Graphical User lterace conte a Rea 12 Commandin 45 24 bare aa 12 2 3 GUI Main WiINdOW 000 eee ee 13 24 MOUSE Bee eee os 14 PM a aes dba Dal ate anaes 14 MGR una a Raya tee Aana Ob Sed Sebel S San aid dye ean ee 14 MOVE tein Bie a O Desired a iat acta as peat eat 15 Averagem wots werent oto a ig ale te het ye R E alg 15 Feb ss oan a oe eek aches ee need ae 15 2 9 Loading data vevs peras aa 16 2 6 Output options te ee net erate Ce ieee peas ea dats 17 Filename Senne iS oa 17 Data PACKING aula th bcp oe cre Aid weet dd Babe si 18 Ignoring Warnings ceden e ea 18 Processing MISION ta o de aa ag 18 2 7 Logging the output wie uri ra a a 20 Chapter 3 MaxFilter parameters 21 3 1 Expansion origin ic vein oie ra wee o wa wet 21 3 2 Harmonic basis functions 200e eee eee eee 22 3 3 Bad channels settings ooooomommmmm m 24 PROD A AAA Ac dati tar ert tages Gove ANN Sek By dodo CU eRe 24 Evoked dala ls sic oad je wth and A tana 25 Ra awdal inc suntan skate a fae Sa SS ates eB ade He ee jie heen So 25 SENSO AMICS a An Bod es ee cee as 26 3 4 MaxST settingS sucia eres 27 3 5 MEG SENSOMS oasa oi o a 29 Sensor ypes ia O aaa da EE E a 29 Sensor MOSES A A A 29 Calibration adjustment and cross talk correction 29 Changing the fine calibration and cross talk correction
30. al spherical shell inside of the sensor array encloses the subjects brain and another one encloses all MEG sensors The radii of the shells are determined respectively as the smallest and largest distances from the origin to the sensor locations The disturbing magnetic interferences are suppressed by omitting the har monic function components corresponding to unduly high spatial frequen cies by neglecting the S space component b and by reducing the S y Space component n out out Since the method is based directly on Maxwell s equations the operation can be called Maxwell filtering hence the name MaxFilter The basic Maxwell filtering operation can be regarded as spatial filtering because separation of b and b is done on the basis of the spatial pat terns and is independent of time Spatial separation can suppress only external interferences emanating from space S such as electromag netic pollution due to power lines radio communication traffic elevators etc External interference can also arise in the patient For instance nor mal cardiac and muscular activation cause fields detectable by MEG sen 6 2006 10 31 NM21993A Introduction EN NM21993A sors and any pieces of magnetized material in on the body may cause very large disturbances Identification and suppression of the Sy space components require addi tional knowledge of the temporal dynamics Spatio temporal extension to Maxwell filtering c
31. al subspaces Correla tion values close to 1 indicate intersecting waveforms which should be projected out of the data Mathematical basis of the temporal subspace projection is described in Appendix C MaxST can be applied to all FIFF data files with sufficiently long data for adequate statistics The program then reports the number of components which are projected out If the length of data is shorter than 600 samples the program reports an error and terminates In order to apply MaxST select Apply MaxST from the Filter menu Figure 3 4 You can toggle MaxST on or off and define the processing buffer length nbuf and subspace intersection correlation limit Po MaxST MaxST FF Use time domain extension MaxST Lenght of raw data buffer s la Subspace correlation limit 0 380 T oK Cancel Help Figure 3 4 Setting MaxST On command line you can use the options st nbuf and corr limit to control MaxST 2006 10 31 27 El MaxFilter parameters 28 MaxST switches the automated bad channel detection off The program however automatically scans all raw data tags to define the maximum number of saturated channels Normally both spatial Maxwell filtering and MaxST utilize an ordinary pseudo inverse in composing the harmonic function amplitudes Some times the program switches to an iterative method to compose the pseudo inverse and harmonic amplitudes see Appendix A 3 for details e Th
32. alled MaxST widens significantly the software shield ing capability of MaxFilter because MaxST can suppress also internal interferences that arise in the Sy space or very close to it Such internal interferences can be caused for example by magnetized pieces in on the subject s head such as dental work braces or magnetized left overs in burr holes or by pacemakers or stimulators attached to the patient Maxwell filtering inherently transforms measured MEG signals into vir tual channels in terms of harmonic function amplitudes Because the vir tual channels are independent of the device they offer a straightforward method for estimating corresponding MEG signals in other sensor arrays This function called MaxMove provides an elegant way to transfer MEG signals between different head positions and to compensate for distur bances caused by head movements during recordings The mathematical basis of Maxwell filtering is described briefly in Appendix A The Signal Space Separation algorithms and their applica tions are discussed in detail in 1 S Taulu M Kajola and J Simola Suppression of interference and artifacts by the signal space separation method Brain Topography 16 4 269 275 2004 2 S Taulu and M Kajola Presentation of electromagnetic multichannel data the signal space separation method Journal of Applied Physics 97 12 124905 June 2005 3 S Taulu J Simola and M Kajola Applications of the signal spac
33. amplitude noises In general the number of virtual channels can vary significantly If Lin 5 there are 35 components Qt while the number of components increases to 143 for 11 Correspondingly the number of external components B varies from 3 L 1 to 35 Lous 5 Virtual chan nels corresponding to highest spatial frequencies may become undetect able and therefore increase the noise in Maxwell filtering Therefore they need to be excluded MaxFilter 2 0 utilizes the total information in determining the most opti mal selection of virtual channels maximal total information also indicates best signal to noise ratio SNR Total information is maximized itera tively using a hypothetical random current density distributed over a spherical volume At each iteration round the virtual channel with small est SNR abs Q Nnm is neglected new pseudo inverse S is obtained and amplitudes at and 1 are recalculated Iteration is stopped when maximal is found 2006 10 31 NM21993A aPPENDIX B Elekta Neuromag MEG sensors B 1 Sensor types NM21993A The detection coil geometry in MaxFilter is similar to the one described in NM20568A Source Modelling Software User s Guide Appendix B Coil geometry information The flux is an integral of the magnetic field component normal to the coil plane Thus the basis matrix elements S for b are approximated by Ny in Sam 5 y Wisin tea Aip RA w
34. and the second sum amplitude coefficients f represents signals from sources outside of the array As the former volume contains the biomagnetic signal sources and the latter volume contains the external disturbance sources any mea sured signal can be uniquely decomposed into two magnetic subspaces with separate coefficients O m gt Bpm corresponding to the subspace spanning the biomagnetic signals and to the subspace spanning the exter nal disturbance signals The basis functions v are expressed in nm nm terms of the vector spherical harmonic VSH functions oY imY V m9 0 n 1 Y 0 fot m0 e nm iMY pm 0 0 nY amer te t Ze nm 9 30 sing where Y are the ordinary spherical harmonic functions and i is the imaginary unit Maxwell filtering is a process where 1 the expansions are terminated to the limit where the spatial frequencies of the n m components become too high such components are buried in the sensor noise and 2 the expan sion of f corresponding to external magnetic interferences is omitted after estimating both o and PB 2006 10 31 45 A Maxwell filtering in a nutshell A 2 Harmonic amplitudes MEG signals can be expressed in matrix form as 5 x gt in b Sx Sin Sour out where 7 5 7 JT Sin anes VN N Xin 1 1 Qy N E r JT Sor Dias Da ul X out Br Bm m Here the coefficients
35. chs 501 503 SSS cal chnls SSS cal coeff 503 901 900 proc history proc record SSS CTC SSS info info correction correction finecalib finecalib proc record proc history If the input file was already processed the program exits with an error message Warning SSS was already applied No output is produced The processing parameters of such files are shown on the GUI low win dow or using the command line option v You can however select Force to ignore warnings if you want to reprocess the input file despite the error message Note Forcing MaxFilter reprocessing may distort the result if different expansion origin or order settings are used NM21993A 2006 10 31 E Using MaxFilter 20 2 7 Logging the output You can display the output of MaxFilter and MaxAve by selecting Show log from the File menu Figure 2 5 The log window has three areas the top text area lists the normal stdout output of the program the middle text area is for displaying all st derr warnings and the lowest text area displays the execution command which the GUI composes for running MaxFilter or MaxAve 5 MaxFilter log x 0 4 78 6 29 25 792 15 g 1000er 50mm 0 3 40 0 9 a t 5 59 e 0 17 cm g 1 00 v 0 19 cm s r 0 06 rad s d 0 09 cm No need to interpolate r 1 1 mm for samples 3300 3419 Hpi fit OK movements mm s 7 7 9 4 7 1 77
36. ct Rejection limits from the Averager menu instead of saving the parameter file Figure 5 3 Rejection limits m Planar rejection T m EEG Y max 21e 13 max 21e 06 min 1e 13 min 21e 06 a noise 1e 13 noise 1e 06 slope 1e 13 slope 1e 06 i ji spike 1e 13 spike 1e 06 Magnetometer rejection T Misc Y max 5e 1 21e 06 ote EMGmax 45 min a 0 00015 noise 1 EOG max f slope 1 a 7 ECGmax spike 1 Figure 5 3 Setting averaging rejection limits You can enter desired rejection limits in the text fields A negative value indicates that the rejection parameter is not used 2006 10 31 NM21993A aPPENDIX A Maxwell filtering in a nutshell A 1 Signal space separation NM21993A MEG devices comprised of more than 300 signal channels provide gener ous oversampling of both biomagnetic and external disturbance magnetic fields Because the sensor array is located in a source free volume between the volume of interest inside of the helmet and the volume con taining all sources of external interference outside of the helmet it turns out that the magnetic signal space can be split into two separate linearly independent subspaces Yrm 0 a ae Bo mE Eu Mar Ho LL Ban Oml n 0m n n 0m n The first sum amplitude coefficients Q represents signals of interest emanating from the head surrounded by the sensors
37. e separation method IEEE Transactions on Signal Processing 53 9 3359 3372 2005 4 S Taulu and J Simola Spatiotemporal signal space separation method for rejecting nearby interference in MEG measurements Physics in Medicine and Biology 51 1759 1768 2006 2006 10 31 7 EN Introduction 1 3 Software functionality MaxFilter 2 0 software provides three separate programs MaxFilter The main application for doing Maxwell filtering called from a command line This program includes also MaxST and Max Move functions MaxAve Off line version of the on line averager provided for conve nient re averaging of raw data before or after Maxwell filte ring MaxFilter_GUI Graphical user interface program which collects the input arguments and then starts MaxFilter or MaxAve as a subpro cess Information of the data processing is displayed on the main display and in a log window The main functions of MaxFilter are Software shielding By subtracting the component b from measured signals b the program performs software shielding on the measured MEG data Automated detection of bad channels By comparing the reconstructed sum b b with measu red signals b the program can automatically detect if there are MEG channels with bad data that need to be excluded from Maxwell filtering Spatio temporal suppression of S space artifacts By subtracting the reconstructed waveforms b t b t from measured signals b t
38. e automated bad channel detec tion parameters On command line you can control bad channel settings using options bad autobad badlimit and magbad Autobad The program can determine automatically if there are MEG channels with spurious sensor artefacts Bad channel detection is performed by recon structing the signals b b int Bouto and by evaluating the difference b b b which in an ideal case should contain only white noise of the SQUID sensors Bad channels typically exhibit large values in b which apparently origi nate in the sensor space Sy The amplitude range is then calculated for 2006 10 31 NM21993A MaxFilter parameters E NM21993A 3 3 2 3 3 3 each channel as d b max b min k 1 M Average and standard deviation values d ye ds are calculated separately over magne tometer and planar gradiometer channels A channel is determined bad if di gt d 1 dsp The default threshold value in MaxFilter 2 0 is r 7 You can enable or disable autobad with the toggle button Detect bad channels automatically and set the threshold value in the Detection limit field The number nraw in the Scan raw tags field means that in the case of raw data the first nraw data buffers are scanned see Section 3 3 3 On command line you can apply the automatic bad channel detection with the option autobad on loff nraw Argument on indicates that the bad channel detection is done separately for each dat
39. e iterative method is always applied for NM122 and Neuromag Sys tem 204 channel data e Ifthe flat channel scanning finds more than 12 saturated channels even for a short period e When you select the command line option iterate Note MaxST is much more time consuming than spatial Maxwell filter ing Therefore you should first inspect the input data to judge if MaxST is needed The default length of data buffering in MaxST is four seconds Offsets and slow frequency variations are often seen also in the sensor space Sy sig nals Therefore MaxST suppresses DC and very slow frequency compo nents Note MaxST acts as a high pass filter where the cut off frequency is related to the buffer length Thus 4 second buffering corresponds to the cut off frequency of 0 25 Hz Longer buffers can account for slower background variations You can increase the buffer length and decrease the cut off frequency by setting nbuf larger Long buffers also increase the memory usage In the case of 306 channel Elekta Neuromag data 4 second buffers typically take about 50 MB of memory while increasing buffer length to 30 seconds expands the memory usage to about 400 MB 2006 10 31 NM21993A MaxFilter parameters Ej NM21993A 3 5 3 5 1 3 5 2 3 5 3 MEG sensors Sensor types Maxwell filtering can process all Elekta Neuromag sensor types using numerical integration over the pickup coils Details of the sensor types are collected
40. e output file and other output options Section 2 6 Set directory Set the current working directory Show log Show a log window to list the stdout and stderr outputs of MaxFilter and MaxAve Section 2 7 Exit Quit the program 2 4 2 Filter Origin Set the harmonic function expansion origin and coordinate frame Section 3 1 2006 10 31 NM21993A Using MaxFilter E Dimensions Set the harmonic function expansion orders Section 3 2 Bad channels Control the bad channel settings Section 3 3 Apply MaxST Set MaxST parameters Section 3 4 Fine calibration Set the fine calibration adjustment file Section 3 5 Cross talk compensation Set the cross talk correction file Section 3 5 2 4 3 Move Data transform Set the parameters for transforming MEG data between differ ent sensor arrays Section 4 1 Head movement compensation Set the parameters to estimate head positions and to compen sate head movements Sections 4 2 4 4 Low pass filter Set the low pass filtering Section 4 5 2 4 4 Averager Load raw data Select the raw data file to be averaged Section 5 3 Output file Select the file for saving the averages Section 5 3 Rejection limits Change averaging rejection limits Section 5 4 2 4 5 Help Why the beep At certain error situations an error dialog is not shown but the termi nal bell is rung This item gives a brief explanation of the reason
41. es the virtual channels in estimating the MEG signals corresponding to a different head position or sensor array The data transformation of MaxMove can be applied in e Conversion of data acquired from one subject patient in several record ing sessions into one reference head position e Conversion of data acquired from several subjects patients into one ref erence standard head position e Correction of disturbances due to head movements in a continuous recording The transformed data allow a strikingly more clear comparison than the original data providing thus a much better starting point e g for grand average studies When you want to transfer data into a different head position select Data transform from the Move menu Figure 4 1 You can then select the file containing the coordinate transformation for the reference head position or set the default transformation On command line you can select coordinate transformations with the option trans name where name is the FIFF file defining the coordi nate transformation of the reference head position If name default the head coordinate axes are identical to the device coordinate axes and the head coordinate frame origin corresponds to the location 0 0 0 of the device coordinate frame 2006 10 31 33 El MaxMove parameters Select coordinate transformation Filter nutaaviroovdata chpi phantom 051 o06 fii Directories Files os 1006 Sr 3 n 051
42. here F y are a set of N integration points covering the pickup coil loops of the sensor Vim kp is the value of the inside VSH function at Pip Myp are the coil normal directions at these points and w kp are the weights associated to the integration points This formula essentially cor responds to numerical integration of the magnetic field over the pickup loops of sensor k The VSH terms for b are approximated in a similar way out Table B 1 lists the parameters of the coil geometry descriptions employed in the software The colums of the table contain the following data 1 The number identifying the coil type This number is used in the coil descriptions found in the FIFF files 2 Description of the coil 3 Number of integration points 4 The locations of the integration points in coil coordinates 5 Weights assigned to the field values at the integration points 2006 10 31 49 El Elekta Neuromag MEG sensors Type Description Nx Fk Wk 2 Neuromag 122 5 44 7 68 0 mm 1 4 16 2mm planar gradiometer 11 1 7 68 0 mm 3012 Type T1 5 89 6 71 0 3 mm 1 4 16 69mm planar gradiometer 10 8 6 71 0 3 mm 3013 Type T2 planar gradiometer 3014 Type T3 planar gradiometer 3022 Type T1 0 0 0 0 0 3 mm 16 81 ae NC Orncles 9 99 9 99 0 3 mm 25 324 3023 Type T2 magnetometer 0 00 9 99 0 3 mm 10 81 9 99 0 0 0 3 mm 10 81 3024 Type T3 0 0 0 0 0 3 mm 16 81 magnetometer 8 1
43. ic components depend on the head position If the head is in the middle of the helmet typically about 20 of all virtual channels are neglected For L 8 the optimal selection involves 65 of 80 harmonic amplitudes On command line you can also apply the option regularize off if you want to apply all virtual channels The default setting is regu larize on which applies the optimized component selection 2006 10 31 23 E MaxFilter parameters 3 3 Bad channels settings 24 3 3 1 Successful Maxwell filtering requires that channels with artifacts or very poor data quality need to be excluded from the reconstructions The chan nels marked in the bad channel tag of the input FIFF file or manually marked bad in starting MaxFilter are treated as static bad channels i e they are automatically excluded You can also use the utility program neuro bin util mark_bad_fiff to mark permanently the channels in the input file that need to be excluded If you need to change bad channel detection settings select Bad channels from the Filter menu Figure 3 3 Bad channel setup Manual selection List of logical channel numbers i Autobad FF Detect bad channels automatically Detection limit zo Scan raw tags 2d OK Cancel Help Figure 3 3 Setting bad channels You can enter the logical channel numbers separated by space for setting manual bad channels You can also set th
44. led by the Elekta Neuromag service are sufficient Warning If the fine calibration and cross talk correction data are not available the performance of Maxwell filtering may not be as good as with the fine calibrated system 3 5 4 Changing the fine calibration and cross talk correction You can change fine calibration and cross talk correction files by select ing Fine calibration or Cross talk compensation from the Filter menu Figure 3 5 Fine calibration Cross talk correction a Filter b Filter neuro databases sss dat neuro databases ctc fit Ultestones Elles Directories s Files llases sss sss_cal dat ct_sparse fif ases sss sss_cal_biomag dat ses cte ct_sparse_biomag fif ases sss tmp WSSS_cal_otaniemi da ses ctc other ct_sparse_orion fif sss_cal_vallila dat t sparse_otaniemi fif ct_sparse_vallila fif l l 4 Z E 17 E Ep an i E i i _ Don t use fine calibration 4 Don t use cross_talk correction Current fine calibration file Current cross talk correction file aeuro databases sss sss_cal_otaniemi dal neuro databases ctc ct_sparse_otaniemi fil ok Filter cancel Help OK Filter Cancel Help Figure 3 5 Selecting a the calibration adjustment file b the cross talk correction file 30 2006 10 31 NM21993A MaxFilter parameters Ej NM21993A 3 5 5 On command line you can define where the correctio
45. ll be located in a correct directory automatically On command line you can define the name of the input file with the option f input_file fif 2006 10 31 NM21993A Using MaxFilter 2 NM21993A 2 6 Output options After loading the input data you can modify the output options by select ing Output options from the File menu Figure 2 4 Besides file selection Section 2 6 1 the dialog has controls for defining the output format and for setting data skips see Section 2 6 2 You can also press the button Force to ignore warnings if you want to process the data despite non fatal warnings see Section 2 6 3 Set output options Filter automount taavi root data chpi phantom 051006 fit Directories Files Tnpi phantom 05 10067 Ie dip04_chpi_still fif A hpi phantom 051006 dip04_nochpi fif dip4 chpi_3pos fif dip4_chpi_3pos_sss fif dip4_chpi_moving fif dip4 chpi_rot fif dip4_chpi_rot2 at1317 ff 7 dip4_chpi_rot2 ff 7 z E m Output format r Data skips lA Input data format Start sec End sec vy Short 16 bits w Long 32 bits v Float 32 bits _ Force to ignore warnings Output file vifroot data chpi phantom 051006 dip4_chpi_3pos_md OK Filter Cancel Help Figure 2 4 Output options 2 6 1 Filename If the output filename is not set the program tries automatically to create a file named according to e
46. lling If the number of isotrak points is too small the program uses the default setting according to the Source Modelling pro gram Xfit see NM20568A Source Modelling Software User s Guide section 5 3 The MEG sphere model The ori gin is set to point 0 0 40 mm in the head coordinate frame Fitting to sensors If the input file does not contain a suitable coordinate transfor mation the program fits a sphere to all sensor locations The origin is set to the fitted point in the device coordinate frame In the case of Elekta Neuromag the optimal device origin is at 0 13 6 mm Note If the sphere fitted to isotrak points extends outside of the sensor array for example due to isotrak points that were digitized outside of the head surface the program reports and error and stops execution In such cases you need to specify the frame and origin manually If you want to set the expansion origin and coordinate frame manually select Origin from the Filter menu Figure 3 1 On command line you can select whether to use the device or head coordi nate frame option frame and where to place the origin of the expan sions option origin 2006 10 31 21 E MaxFilter parameters Expansion origin m Inside expansion origin origin x mm 0 0 origin y mm oo originz mm 40 0 m Coordinate system wv device 2 head OK Cancel Help Figure 3 1 Setting the expansion origin and coordinate frame
47. magnetometer calibration factors and imbalance correction for the planar gradiometers In addition cross talk correction can be applied to reduce mutual interference between overlapping magnetometer and gradiometer loops of a sensor unit Currently these options are available only for 306 channel Elekta Neuro mag systems Fine calibration and cross talk matrix files are prepared and installed by the Elekta Neuromag service personnel 2006 10 31 29 El MaxFilter parameters Fine calibration adjustment is not performed if the fine calibration file is not found or the processing history of input_file fif already includes the fine calibration Likewise the cross talk correction is not done if the cross talk matrix file is not found or the processing history of input_file fif already includes the correction After opening the FIFF file the program attempts to load the channel cross talk correction and fine calibration data files The default files are respectively NEUROMAG_ROOT databases ctc ct_sparse fif and NEUROMAG_ROOT databases sss sss_cal dat The root directory can be defined via the environmental variable NEUROMAG_ROOT By default it points to directory neuro or opt neu romag For other locations you should set NEUROMAG_ROOT to desired directory path before running MaxFilter If you analyze data only from one Elekta Neuromag system the default cross talk correction and fine calibration data files instal
48. max L Loun 1 Estimation of the moments x x 18 then performed iteratively On each iteration round all expansion orders are processed one by one by set ting the moments of the kth order to zero and by estimating the moments of other orders from the sub blocks of the pseudo inverse matrix The iter ation converges typically in less than 10 rounds The iterative method is always applied as default to Neuromag System and Neuromag 122 data A 4 Optimization of virtual channel selection The celebrated Shannon s theory of information transmission can be applied in MEG as well A single magnetometer can be regarded as a noisy channel conveying information from the sources in the brain Its output b t is the sum of the signal s t and noise n t When b t and n t are normally distributed and independent the information gained per one sample is J 1 2 log P 1 where P is the power signal to noise ratio In the case of multichannel arrays the channels need to be orthogonalized to make them independent of each other before the total information of the sensor array is calculated Because Maxwell filtering transforms the measured signals b t into orthogonal virtual channels we can directly utilize this property in evaluating the total information 2006 10 31 47 A Maxwell filtering in a nutshell 48 1 my la zL ledge n m nm where N m represents the sensor noise n t converted into the VSH
49. meter size on the basis of the calibration factors even if sensors of type T3 are marked as T2 sensors Data analysis source modelling software may however utilize too large dimension for magnetometers but such an error does not have a significant effect on the analysis results MaxFilter option T3 was included to correct the sensor types if needed and to remove further modelling inaccuracies due to wrong sensor size New data acquisition programs version 3 3 and thereafter set the correct sensor types automatically Yet older version of the source modelling pro grams xfit mce may not recognize sensor types T3 Therefore MaxFil ter provides the option T2 for convenience it marks all magnetometer and gradiometer channels as type T2 sensors Note that this option may cause a small numerical error in the source modelling programs as explained above 2006 10 31 51 B Elekta Neuromag MEG sensors 52 2006 10 31 NM21993A APPENDIx c Temporal subspace projection NM21993A The spatio temp oral extension to Mawell filtering MaxST utilizes first spatial reconstructions of b and b for each sensor and each sample and subtracts them from the measured data b t b t bj 0 D 4 0 The data b and b are packed in nxm matrices B and B m sen sors and n samples and decomposed with the singular value decomposi tion Bin UinSinVin Sin diago in in B U SV S diag o sos Ss where k
50. n freezes when the HPI signals are switched off and performes Maxwell filtering using the latest head position 2006 10 31 NM21993A MaxMove parameters EJ 4 5 Low pass filtering Even when the HPI signals are subtracted residual sinusoidal signals may remain e g during rapid head movements You can apply infinite impulse response IIR low pass filtering to remove the remaining HPI signals by selecting Low pass filter from the Move menu Figure 4 3 You can toggle the filter on or off and set the corner frequency rc IIR low pass filtering Set LP filtering F Use low pass filtering Corner fequency Hz oo _ OK i Cancel Help k Figure 4 3 Controlling low pass filtering On command line you can select the option 1pfilt freq where freq is the low pass corner frequency Typical HPI signal frequencies are 154 158 162 166 and 170 Hz Thus the default corner frequency value is 120 Hz MaxFilter also includes some finite impulse response FIR filters for downsampling the result file You can invoke the downsampling with the option ds factor where fact is the downsampling ratio and the low pass frequency is determined from the sampling frequency sf according to Table 4 1 factor frequency sf 2 sf 6 sf 9 sf 12 n BR ww N Table 4 1 NM21993A 2006 10 31 39 El MaxMove parameters 40 4 6 Default parameters
51. n data are loaded from with the options ctc and cal These options are needed if you need to analyze data that were recorded with different measurement devices For convenience MaxFilter includes option site sitename which tries to load the files NEUROMAG_ROOT databases ctc ct_sparse_sitename fif and NEUROMAG_ROOT databases sss sss_cal_sitename dat Note The program reports an error and terminates if the selected files are not found or if they do not contain appropriate data By default the fine calibration and cross talk correction are always attempted if suitable files are found Sometimes they however need to be switched off for example with simulated data or evoked FIFF files exported from the graph program when the adjustments were already applied in the original raw data file The current version of MaxFilter does not check automatically if the selected fine calibration or cross talk correction files are consistent with the input data file Therefore you should be very careful when e Changing the cross talk or fine calibration correction files from the default ones e Processing data that were recorded with other measurement devices Warning Special care should be taken to ensure that right fine calibra tion data are used for imported or old data for which the default calibra tion does not apply The program has also a command line maintenance option ctc only see the next Section Reconstruction of se
52. nsor signals Maxwell filtering transforms measured MEG data inherently to harmonic function amplitudes which can be interpreted as virtual channels see Section A 2 on page 46 The virtual channels are not stored in the output file but they are instead utilized in Maxwell filtering operations such as in composing interference free brain signals and in transforming data between different head positions Normally the program applies the vir tual channels to convert the input data to idealized sensors Besides inter ference suppression MaxFilter removes the distortions caused by imperfect calibration and gradiometer imbalance Sometimes it is however useful to reconstruct the signals b and b without correcting the above mentioned non idealities e g for compari son with the recorded signals 2006 10 31 31 El MaxFilter parameters You can apply the command line option reconst to compose the non idealized signals corresponding to spaces S and S us The program applies fine calibration data in determining the harmonic function ampli tudes but in contrast to idealized channels the reconstruction utilizes all virtual channels without optimizing the selection The program estimates the signals b and b using the standard calibration extracted from the input file Note that the cross talk correction is however applied Thus in order to compare the original and reconstructed data you should run le fif reconst both o rec
53. off Applies cross talk correction matrix in ct_matrix_file fif or switches the cross talk correction off magbad Marks all magnetometer channels bad uses only gradiometer chan nels in Maxwell filtering regularize on off Applies multipole component selection using signal to noise ratios maximized total information or applies all components with the option off iterate ri Applies iterative pseudo inverse and multipole amplitude estima tion set n iteration rounds default 10 Option n 0 forces direct pseudo inverse NM21993A 2006 10 31 61 E Command line arguments 62 ds factor Applies down sampling and low pass FIR filtering Default down sampling factor is 2 Maintenance options Furthermore some options have been reserved for maintenance use Please note that these options are not meant for regular data analysis reconst in out both Reconstructs inside outside or both field components magscale mult Applies scaling between magnetometer and gradiometer channels Default factor is 100 nosss Just copies input data to output Default output filename is input_file_nosss fif ctc only Applies cross talk correction but does not do any Maxwell filtering operations Default output filename is input_file_ctc fif T2 Changes all Elekta Neuromag sensors of type T3 to type T2 T3 Corrects Elekta Neuromag type T2 magnetometers to T3 if needed I ISt Shows more detailed outpu
54. ooommooo 57 Appendix E Command line arguments 59 Appendix F Revision history 63 CHAPTER 1 Introduction 1 1 Overview NM21993A This User s Guide gives detailed explanation of the Elekta Neuromag MaxFilter 2 0 software for MEG data analysis MaxFilter is intended to be used with Elekta Neuromag MEG products in suppressing magnetic interferences coming from inside and outside of the sensor array in reducing measurement artifacts in transforming data between different head positions and in compensating disturbances due to head movements This Chapter presents a general overview and main functionalities of the software Chapters 2 5 describe how to start the program and how to con trol the functionality and parameters Mathematical background and some further information are included in the Appendices 2006 10 31 5 EN Introduction 1 2 Maxwell filtering Signal Space Separation is a method that utilizes the fundamental proper ties of electromagnetic fields and harmonic function expansions in sepa rating the measured MEG data into three components Figure 1 1 bin The brain signals originating inside of the sensor array space S bout External disturbances arising outside of the sensor array space Sour n Noise and artifacts generated by the sensors and sources of interference located very close to the sensors space Sy b Din Dout n Figure 1 1 The geometry in Maxwell filtering One hypothetic
55. play appear ance changes according to Figure 5 1 Instead of showing MaxFilter parameter values and head position parameter drawing area the dialog is reserved for showing the progress of averaging After you have defined the output filename and optionally modified the parameters press Execute to start averaging The scale bar indicates the number of processed data buffers The text area shows how many epochs were found and were they added or rejected You can press STOP to can cel averaging if the program rejects too many or too few epochs Then you can modify the rejection limits and try again press Execute 2006 10 31 41 5 MaxAve parameters 42 MaxFilter 2 0 imix File Filter Move Averager Help Averager E Tag 40 4 events Epoch ok Epoch ok 74 responses averaged Tag 41 4 events Epoch ok Epoch ok 76 responses averaged Tag 42 5 events Epoch rejected large amp MEG1332 7 14111e 11 7e 11 Epoch ok 77 responses averaged Tag 43 5 events Epoch ok Epoch ok 79 responses averaged Tag 44 4 events Epoch ok Epoch ok Epoch ok 82 responses averaged Tag 45 4 events Epoch ok E La cae Wrote successfully the result file automount taavi root data chpi ph antom 060329 dip04_chpi_still_raw_new_mc_avwe fif No warnings 48 No errors Processing finished OK Figure 5 1 Averaging dialog You can view more detailed information of the MaxAve output by select ing Sho
56. r and stops head position estimation if it does find the HPI result from the input file In the outset MaxFilter estimates the consistency between fitted isotrak points and initial HPI fitting results The fitted coil positions are trans formed to head coordinates and the distances between them and the isotrak points are calculated If the mismatch of any coil is larger than 2 mm the program gives a warning After extracting HPI signal amplitudes the program estimates the position of each coil the location parameters x y z are searched with non linear Simplex minimization using the same interval as with HPI amplitude extraction The success of HPI fitting is judged in terms of Goodness of fit g value Measures the match of measured and modelled HPI amplitude data ranges between 0 and 1 Estimation error The distance of the isotrak point and the fitted point when fit ted HPI coil positions are transformed from the device to the head frame HPI fitting is considered successful if at least in three HPI coils g value gt 0 98 and error lt 5 mm In such cases the program reports that the fit was OK HPI fit fails if the acceptance criteria are met with less than three HPI coils In such cases the coordinate transformation cannot be defined and the program gives a warning Matching of the fitted and digitized HPI coil positions is performed using quaternion parameters resulting in a coordinate transformation from the device to the
57. re presented below Warning It is important that the user inspects both the input and the out put data visually to judge the quality of the Maxwell filtering result Warning If the fine calibration and cross talk correction data are not available the performance of Maxwell filtering may not be as good as with the fine calibrated system Warning Special care should be taken to ensure that right fine calibra tion data are used for imported or old data for which the default calibra tion does not apply Warning If the threshold of the automated bad channel detection is too small the program may classify good channels as bads and if it is too high some bad channels may remain undetected Warning When transforming data from one head position into another special care should be taken to ensure that both head positions are accu rately defined in respect to the sensor array Warning Head position calculation errors affect the data quality after movement compensation The user must inspect the head position fitting error and goodness before data analysis Warning If MaxShield was applied in the input file the user must not perform data analysis on MaxFilter output files obtained with the options nosss or ctc only gt gt gt ibbppb 2006 10 31 9 EN Introduction 10 2006 10 31 NM21993A CHAPTER 2 Using MaxFilter 2 1 Background NM21993A MaxFilter can be applied to a FIFF file with raw data or averaged ME
58. s one by one The channels that appear bad in more than one buffer are treated as static bad channels throughout the whole raw data file 2006 10 31 25 El vextiter parameters O MaxFilter parameters 26 3 3 4 If the default settings are not satisfying you can try one of the following options e Set the bad channels manually and disable autobad e Setnraw 1 or autobad on and select Force to ignore warnings the program tries then to detect bad channels from each raw data buffer separately e Set nraw n Note If the iterative method is used see Section 3 4 and Appendix A 3 automated bad channel scanning may become slow In such cases you can set bad channels manually and disable autobad to speed up the processing Sensor artifacts Sometimes the MEG data may contain saturated channels or interference that originates in the SQUID sensors or the electronics Such disturbances are usually manifested in few channels as spurious artifacts in the signal and the program can detect and discard such artifacts automatically If the sensor artifacts are present in a larger number of channels e g due to a strong interference coupled via the electronics the result may still contain unwanted interference contributions In such cases you can apply MaxST to reduce the interferences see Section 3 4 Note If the raw data has segments where there are too many artifact or saturated channels the program may not be able to do Max
59. t than option v mainly for tracing pos sible problems in processing the datafile Warning If MaxShield M was applied in the input file the user must not perform data analysis on MaxFilter output files obtained with the options nosss or ctc only 2006 10 31 NM21993A aPPENDIX F Revision history NM21993A This Appendix lists the changes made to the MaxFilter program and User s Guide NM21492A D e User s Guide for program MaxFilter 1 1 NM21993A e New software version MaxFilter 2 0 e The software has several important extensions Section 1 3 e Graphical User Interface Chapter 2 e MaxST Chapter 3 e MaxMove Chapter 4 e MaxAve Chapter 5 e Appendices B E 2006 10 31 63 F Revision history 64 NM21993A
60. w log from the File menu On command line you can run the program as neuro bin util maxave v i input_file fif o output_file fif e event_file fif p par_file dat where Switches on verbose logging H input_file fif Defines the FIFF file where the raw data are read O output_file fif Defines the FIFF file where the averaged results are written 0 event_file fif Optional FIFF file for the events p par_file dat Optional ascii file where the averaging parameters are read 2006 10 31 NM21993A MaxAve parameters g NM21993A 5 3 Selecting files MaxAve can be applied to FIFF files containing raw data Off line aver aging can however be performed only if the on line averaging parameters were set during data acquisition Otherwise the program reports an error and exits MaxAve accepts raw data FIFF files as input New data are loaded by selecting Load raw data from the Averager menu Figure 5 2 Besides file selection the dialog has a control button for optional saving of the averaging parameters Select file to load Filter idata chpi phantom 060329 fi Directc Files dip04_chpi_rotfif dip04_chpi_rot_raw fif dip04 chpi_rot_raw_mc fif dip04_chpi_rot_raw_quat fif dip04_chpi_rot_raw_quat_ave fif dip04_chpi_rot_raw_sss fif dipo4_chpi_still fif dip04_chpi_still_raw fif Ale a F Save parameters to file jave par Current file ahantom 060329
61. well filtering properly You may still be able to process the segments which show acceptable data The program gives a warning if there are such bad data segments they are indicated by setting the output data of all channels to zero If you for some reason want to exclude all magnetometers from Maxwell filtering you can use the command line option magbad The program determines the harmonic amplitudes from gradiometer channels only and utilizes the calibration adjustment and cross talk correction Both gradi ometer and magnetometer channels are then reconstructed from the har monic amplitudes 2006 10 31 NM21993A MaxFilter parameters Ej NM21993A 3 4 MaxST settings As briefly described in Section 1 2 MaxST can be regarded as a four dimensional filter besides the three spatial dimensions it also checks the time domain First normal spatial Maxwell filtering is applied to the data typically in blocks of four seconds The program reconstructs the waveforms b and b t and subtracts them from the measured data b 1 b t b t 0 0 D 4 0 If the interference is located very close to the sensor array residual wave forms exhibit very large disturbances and remaining disturbance is also evident in b ft The insight of the temporal extension is that if there are similar wave forms in b and b they must be artifacts Such waveforms can be easily recognized by computing correlations of the tempor
62. x icon labeled as MaxFilter Linux Select the menu Applications gt Other click icon Max Filter Command line neuro bin vue maxfilter_gui Upon launching the program checks available licenses and displays a welcome logo Figure 2 1 Click Hide to close the window Information Welcome to Elekta Neuromag MaxFilter User interface version 2 0 13 Oct 16 2006 14 02 06 Copyright C 2006 Elekta Neuromag Oy All rights reserved The product is protected by patents US2006031038 Signal Space Separation US6876196 Head position determination WO2005067789 DC fields W0O2005078467 SmartShield FI20050445 MaxST Figure 2 1 MaxFilter welcome window 2 2 2 Command line Alternatively you can start MaxFilter from a command line as neuro bin util maxfilter f input_file fif options If no arguments are given the program gives just a brief message usage maxfilter f lt infile gt options maxfilter help shows available options A comprehensive list of the options is given in Appendix E 2006 10 31 NM21993A Using MaxFilter E NM21993A 2 3 GUI main window The main window consists of the following areas MaxFilter 2 0 1 File Filter Move Averager Help Current settings Input filename automount taavi root data chpi phantorn 051006 dip4_chpi_3pos fif gt Output filename automount taavi root data chpi phantom 05 1006 dip4_chpi_3pos_sss fif Origin
63. zed material on the head or in the EEG cables In such cases you may need the capabilities of MaxST at the same time with movement compensation Note Disturbances on MEG channels are especially evident during rapid head movements Therefore they may deteriorate the HPI signal estima tion and position fitting Warning Head position calculation errors affect the data quality after movement compensation The user must inspect the head position fitting error and goodness before data analysis The program recognizes from each raw data block if the continuous HPI is on or off In the latter case the program reports that continous HPI was off and skips the data block from the result file The output data block shows zero values on all MEG channels On command line you can select the option movecomp to perform the movement compensation You can select the reference position with the option trans see above If the transformation is not specified the program uses the transformation stored in the original file when acquiring the data Sometimes the continuous HPI may be needed only periodically e g for few seconds every 1 or 2 minutes to check the head position during a long recording In such cases you can select the command line option move comp inter Then the program does head position estimation and move ment compensation during the periods where continuous HPI is on Instead of skipping data the program assumes that the head positio

MaxFilter 2.0 Manual

Contents

Download Pdf Manuals

Related Search

Related Contents