Method and device for recognizing and removing undesired artifacts in multichannel magnetic field or electric potential measurements
The present invention introduces a method, device and a computer program for removing artifacts present in individual channels of a multichannel measurement device. At first, a basis is generated defining an n-dimensional subspace of the N-dimensional signal space, where n is smaller than N, where using in the definition of the n-dimensional basis a physical model of a Signal Space Separation method, or a statistical model based on the statistics of recorded N-dimensional signals. Thereafter, a combined (n+m)-dimensional basis is formed by adding m signal vectors to the n-dimensional basis, each of these m signal vectors representing a signal present only in a single channel of the N-channel device. After this the recorded N-dimensional signal vector is decomposed into n+m components in the combined basis, and finally, components corresponding to the m added vectors in the combined basis are subtracted from the recorded N-dimensional signal vector.
1. A method for recognizing and removing undesired artifacts and sensor-specific interference signals in at least one measurement channel of a multichannel magnetic field or electric potential measurement, wherein a measurement device comprises a set of N measurement sensors and a data processing logic, and wherein the method comprises the step of:
receiving, at an electronic processor, an N-dimensional signal from the set of N measurement sensors;
recording the N-dimensional signal;
generating, via the electronic processor, a basis defining an n-dimensional subspace of the N-dimensional signal space, wherein n is smaller than N, using, in the definition of the n-dimensional basis, a physical model of a Signal Space Separation method, or a statistical model based on the statistics of the recorded N-dimensional signal;
characterized in that the method further comprises the steps of:
forming a combined (n+m)-dimensional basis by adding m signal vectors to the n-dimensional basis, each of these m signal vectors representing a signal present only in a single channel of the N-channel device,
decomposing the recorded N-dimensional signal into n+m components in the combined basis, and
subtracting, from the recorded N-dimensional signal, the components corresponding to the m added vectors in the combined basis;
producing a modified N-dimensional signal by removing the undesired artifacts and sensor-specific interference signals, via the electronic processor, based on the recorded N-dimensional signal; and
outputting, via the electronic processor, the modified N-dimensional signal;
wherein the N-dimensional signal is recorded at a temporal sampling rate exceeding the highest frequencies in a signal of interest and at a spatial sampling rate exceeding the highest spatial frequencies in a signal of interest, so that n+m is smaller than N.
2. The method according to claim 1 , characterized by,
adding only one signal vector, representing the N-dimensional signal in this one channel only, to the n-dimensional basis, where m is equal to 1.
3. The method according to claim 1 , further characterized by,
repeating the decomposition of the recorded N-dimensional signal into the n+1 components in the combined basis N times, using as the one additive basis vector a vector corresponding to each one of the channels at a time, and
subtracting the N signal components corresponding to the additive basis vectors in each decomposition from the original data.
4. The method according to claim 1 , further characterized by,
stabilizing numerically a solution of a linear problem related to the decomposition of the N−15 dimensional signal into the n+m components by using a regularization method.
5. The method according to claim 4 , further characterized by,
estimating statistical correlations of the sensor noise between the N channels from differences between consecutive samples in the recorded signals.
6. The method according to claim 5 , further characterized by,
estimating the statistical correlations of the sensor noise between the N channels from the correlations in high-pass filtered recorded signals.
7. The method according to claim 1 , characterized by the steps of forming, decomposing and subtracting are realized by a cross validation method where the signal of m channels are derived from the signals of the other N-m channels based on the n-dimensional physical model, or based on the n-dimensional statistical model of the original N-dimensional signal.
8. The method according to claim 1 , further characterized by,
applying the method to recordings made by a multichannel magnetoencephalography (MEG) device.
9. The method according to claim 1 , further characterized by,
applying the method to recordings made by a multichannel electroencephalography (EEG) device.
10. The method according to claim 1 , further characterized by,
applying the method to recordings of magnetic resonance signals obtained by a multichannel MRI device.
11. A multichannel measurement device arranged to recognize and remove undesired artifacts and sensor-specific interference signals in at least one measurement channel of a multichannel magnetic field or electric potential measurement, wherein the measurement device comprises
a set of N measurement sensors arranged to record the N-dimensional signal at a spatial sampling rate exceeding the highest spatial frequencies in a signal of interest, so that n+m is smaller than N; and
a data processing logic, which is arranged to receive and record, at an electronic processor, the N-dimensional signal from the set of N measurement sensors and record the signal at a temporal sampling rate exceeding the highest frequencies in a signal of interest and generate a basis defining an n-dimensional subspace of the N-dimensional signal space, wherein n is smaller than N, using, in the definition of the n-dimensional basis, a physical model of a Signal Space Separation method, or a statistical model based on the statistics of a recorded N-dimensional signal;
characterized in that the data processing logic is further arranged to:
form a combined (n+m)-dimensional basis by adding m signal vectors to the n-dimensional basis, each of these m signal vectors representing a signal present only in a single channel of the N-channel device,
decompose the recorded N-dimensional signal into n+m components in the combined basis, and
subtract, from the recorded N-dimensional signal, the components corresponding to the m added vectors in the combined basis;
produce a modified N-dimensional signal by removing undesired artifacts and sensor-specific interference signals, via the electronic processor, based on the recorded N-dimensional signal; and
output, via the electronic processor, the modified N-dimensional signal.
12. The device according to claim 11 , characterized in that the set of N measurement sensors is further arranged to:
record the N-dimensional signal at a spatial sampling rate exceeding the highest spatial frequencies in a signal of interest, so that n+m is smaller than N.
13. The device according to claim 11 , characterized in that the data processing logic is further arranged to:
add only one signal vector, representing the signal in this one channel only, to the n-dimensional basis, where m is equal to 1.
14. The device according to claim 11 , characterized in that the data processing logic is further arranged to:
repeat the decomposition of the N-dimensional recorded signal into the n+1 components in the combined basis N times, using as the one additive basis vector a vector corresponding to each one of the channels at a time, and
subtract the N signal components corresponding to the additive basis vectors in each decomposition from the original data.
15. The device according to claim 11 , characterized in that the data processing logic is further arranged to:
stabilize numerically a solution of a linear problem related to the decomposition of the N-dimensional signal into the n+m components by using a regularization method.
16. The device according to claim 15 , characterized in that the data processing logic is further arranged to:
estimate statistical correlations of the sensor noise between the N channels from differences between consecutive samples in the recorded signals.
17. The device according to claim 16 , characterized in that the data processing logic is further arranged to:
estimate the statistical correlations of the sensor noise between the N channels from the correlations in high-pass filtered recorded signals.
18. The device according to claim 11 , characterized in that the data processing logic is further arranged to:
realize the steps of forming, decomposing and subtracting by a cross validation method where the signal of m channels are derived from the signals of the other N-m channels based on the n-dimensional physical model, or based on the n-dimensional statistical model of the original N-dimensional signal.
19. The device according to claim 11 , characterized in that the device is a multichannel magnetoencephalography (MEG) device.
20. The device according to claim 11 , characterized in that the device is a multichannel electroencephalography (EEG) device.
21. The device according to claim 11 , characterized in that the device is a multichannel magnetic resonance imaging (MRI) device obtaining magnetic resonance signals.
22. A computer program for recognizing and removing undesired artifacts and sensor-specific interference signals in at least one measurement channel of a multichannel magnetic field or electric potential measurement, wherein a measurement device comprises a set of N measurement sensors and a data processing logic, and wherein the computer program comprises program code which, when run on an electronic processor, is arranged to execute the steps of:
receiving, at the electronic processor, an N-dimensional signal from the set of N measurement sensors;
recording the N-dimensional signal;
generating, via the electronic processor, a basis defining an n-dimensional subspace of the N-dimensional signal space, wherein n is smaller than N, using, in the definition of the n-dimensional basis, a physical model of a Signal Space Separation method, or a statistical model based on the statistics of a recorded N-dimensional signal;
characterized in that the computer program is further arranged to execute the steps of:
forming a combined (n+m)-dimensional basis by adding m signal vectors to the n-dimensional basis, each of these m signal vectors representing a signal present only in a single channel of the
N-channel device,
decomposing the recorded N-dimensional signal into n+m components in the combined basis, and
subtracting, from the recorded N-dimensional signal, the components corresponding to the m added vectors in the combined basis;
producing a modified N-dimensional signal by removing the undesired artifacts and sensor-specific interference signals, via the electronic processor, based on the recorded N-dimensional signal; and
outputting, via the electronic processor, the modified N-dimensional signal;
wherein the N-dimensional signal is recorded at a temporal sampling rate exceeding the highest frequencies in a signal of interest and at a spatial sampling rate exceeding the highest spatial frequencies in a signal of interest, so that n+m is smaller than N.
23. The computer program according to claim 22 , characterized in that the computer program is stored on a medium readable by the electronic processor.