NOISE REMOVAL IN MAGNETOMETER FOR MEDICAL USE

Abstract

A method of using a magnetometer system to analyse the magnetic field of a region of a subject's body is disclosed. The method comprises using one or more detectors to detect the time varying magnetic field of a region of a subject's body, filtering a signal or signals from the one or more detectors using a filter or filters, and using the filtered signal or signals to analyse the magnetic field generated by the region of a subject's body. The filter or filters is configured to attenuate noise in the signal or signals that is synchronised with motion of the region of the subject's body such as ballistocardiographic noise.

Claims

1. A method of using a magnetometer system to analyse the magnetic field of a region of a subject's body, the method comprising: using one or more detectors to detect the time varying magnetic field of a region of a subject's body; filtering a signal or signals from the one or more detectors using a filter or filters, wherein the filter or filters is configured to attenuate noise in the signal or signals that is synchronised with motion of the region of the subject's body; and using the filtered signal or signals to analyse the magnetic field generated by the region of a subject's body.

2. The method of claim 1, wherein the filter or filters is configured to attenuate signals having frequencies below a low frequency cut-off frequency, wherein the low frequency cut-off frequency is optionally between around 8 and 12 Hz.

3. (canceled)

4. (canceled)

5. The method of claim 1, wherein the filter or filters is configured to attenuate signals having frequencies above a high frequency cut-off frequency.

6. The method of claim 5, wherein the high frequency cut-off frequency is between around 45 and 60 Hz.

7. The method of claim 1, wherein the filter or filters comprises at least one windowed sinc filter.

8. The method of claim 7, wherein the windowed sine filter is formed using a Blackman window.

9. The method of claim 1, comprising using the magnetometer system to detect the time varying magnetic field of a region of a subject's body in a non-magnetically shielded environment.

10. The method of claim 1, comprising using the magnetometer system to detect the time varying magnetic field of a region of a subject's body when the subject is supported by a structure that comprises electrically conductive and/or ferrous material.

11. The method of claim 1, wherein the region of the subject's body comprises one of: the abdomen, bladder, heart, head, brain, chest, womb, one or more foetuses, or a muscle.

12. A magnetometer system for medical use, comprising: one or more detectors for detecting the time varying magnetic field of a region of a subject's body; and a filter or filters configured to filter a signal or signals from the one or more detectors, wherein the filter or filters is configured to attenuate noise in the signal or signals that is synchronised with motion of the region of the subject's body; wherein the magnetometer system is configured to provide the filtered signal or signals for use to analyse the magnetic field generated by the region of the subject's body.

13. The system of claim 12, wherein the filter or filters is configured to attenuate signals having frequencies below a low frequency cut-off frequency.

14. The system of claim 13, wherein the low frequency cut-off frequency is between around 8 and 12 Hz.

15. A magnetometer system for medical use, comprising: one or more detectors for detecting the time varying magnetic field of a region of a subject's body; and a filter or filters configured to filter a signal or signals from the one or more detectors, wherein the filter or filters is configured to attenuate signals having frequencies below a low frequency cut-off frequency, wherein the low frequency cut-off frequency is between around 8 and 12 Hz; wherein the magnetometer system is configured to provide the filtered signal or signals for use to analyse the magnetic field generated by the region of the subject's body.

16. The system of claim 12, wherein the filter or filters is configured to attenuate signals having frequencies above a high frequency cut-off frequency.

17. The system of claim 15, wherein the high frequency cut-off frequency is between around 45 and 60 Hz.

18. The system of claim 12, wherein the filter or filters comprises at least one windowed sinc filter.

19. The system of claim 12, wherein the windowed sinc filter is formed using a Blackman window.

20. The system of claim 12, wherein the region of the subject's body comprises one of: the abdomen, bladder, heart, head, brain, chest, womb, one or more foetus, or a muscle.

21. The system of claim 12, wherein the magnetometer system is configured to detect the time varying magnetic field of a region of a subject's body in a non-magnetically shielded environment.

22. The system of claim 12, further comprising a support structure for supporting the subject's body, wherein the support structure comprises electrically conductive and/or ferrous material.

Description

[0187] A number of preferred embodiments of the present invention will now be described by way of example only and with reference to the accompanying drawings, in which:

[0188] FIG. 1 shows schematically the use of an embodiment of the present invention for detecting the magnetic field of a subject's heart;

[0189] FIGS. 2-5 show further exemplary arrangements of the use of an embodiment of the present invention when detecting the magnetic field of a subject's heart;

[0190] FIG. 6A shows schematically a coil arrangement in accordance with an embodiment of the present invention, and FIG. 6B shows schematically another coil arrangement in accordance with an embodiment of the present invention;

[0191] FIG. 7 shows a typical healthy ECG trace;

[0192] FIG. 8 shows a further exemplary arrangement of the use of an embodiment of the present invention when detecting the magnetic field of a subject's heart;

[0193] FIG. 9A shows cycle averaged MCG data for a healthy subject captured by a 37-channel magnetometer device in an unshielded environment on a wooden bed, and FIG. 9B shows cycle averaged MCG data for a healthy subject captured by a 37-channel magnetometer device in an unshielded environment on a bed comprising ferrous (magnetic) material;

[0194] FIG. 10 A shows a log periodogram of MCG data captured by a 37-channel magnetometer device in an unshielded environment without a subject present, FIG. 10B shows a log periodogram of MCG data for a healthy subject captured by a 37-channel magnetometer device in an unshielded environment on a wooden bed, and FIG. 100 shows corresponding data for a bed comprising ferrous (magnetic) material;

[0195] FIG. 11 illustrates an ideal band-pass filter in the frequency domain;

[0196] FIG. 12A shows a windowed-sinc filter kernel with a cut-off frequency at 45 Hz and M=2400, and FIG. 12B shows the frequency response of the filter;

[0197] FIG. 13A shows a filter kernel formed from the difference between two windowed-sinc filters with cut-off frequencies at 8 Hz and 45 Hz, and M=2400, and FIG. 13B shows the frequency response of the filter;

[0198] FIG. 14A shows an averaged MCG signal recorded in a non-shielded room, FIG. 14B shows the Fourier spectrum of the signal of FIG. 14A, FIG. 14C shows a windowed-sinc filter kernel with cut-off frequencies at 8 Hz and 45 Hz, FIG. 14D shows the corresponding frequency response of the filter kernel of FIG. 14C, FIG. 14E shows the result of applying the filter in the time domain to the signal of FIG. 14A (solid line) and the result of applying a filter with cut-off frequencies at 2 Hz and 45 Hz to the signal of FIG. 14A (dashed line), and FIG. 14F shows the result of applying the filter in the frequency domain to the signal of FIG. 14A;

[0199] FIG. 15A again shows the cycle averaged MCG data for a healthy subject captured by a 37-channel magnetometer device in an unshielded environment on a bed comprising ferrous (magnetic) material of FIG. 9B, and FIGS. 15B and 15C show the data after a windowed sinc filter kernel with a Blackman window and cut-off frequencies at 8 Hz and 45 Hz has been applied to the data;

[0200] FIG. 16 illustrates a process in accordance with an embodiment of the present invention.

[0201] Like reference numerals are used for like components where appropriate in the Figures.

[0202] Magnetocardiography (MCG) is the measurement of magnetic fields emitted by the heart caused by the electrical current within myocardium heart cells. The measurement of these fields provides diagnostically significant information which is complimentary to that obtained by electrocardiography (ECG), and can be used to diagnose abnormalities of heart function.

[0203] FIG. 1 shows schematically the basic arrangement of a preferred embodiment of a magnetometer system that may be operated in accordance with the present invention. This magnetometer system is specifically intended for use as a cardiac magnetometer (for use to detect the magnetic field of a subject's heart). However, the same magnetometer design can be used to detect the magnetic field produced by other body regions, for example for detecting and diagnosing bladder conditions, pre-term labour, foetal abnormalities and for magnetoencephalography. Thus, although the present embodiment is described with particular reference to cardio-magnetometry, it should be noted that the present embodiment (and the present invention) extends to other medical uses as well.

[0204] The magnetometer system comprises a detector 40 coupled to a detection circuit 41 that may contain a number of components. The detector 40 may be an induction coil 40. The detection circuit 41 may comprise a low impedance preamplifier, such as a microphone amplifier, that is connected to the coil 40.

[0205] The current output from the coil 40 is processed and converted to a voltage by the detection circuit 41 and provided to an analogue to digital converter (ADC) 42 which digitises the analogue signal from the coil 40 and provides it to a data acquisition system 43.

[0206] A biological signal that is correlated to the heartbeat, e.g. an ECG or Pulse-Ox trigger from the test subject may be used as a detection trigger for the digital signal acquisition, and the digitised signal over a number of trigger pulses is then binned into appropriate signal bins, and the signal bins overlaid or averaged, by the data acquisition unit 43. Other arrangements would, however, be possible.

[0207] The coil 40 and detection circuit 41 may be arranged such that the coil 40 and the preamplifier of the detection circuit 41 are arranged together in a sensor head or probe which is then joined by a wire to a processing circuit that comprises the remaining components of the detection circuit 41. Connecting the sensor head (probe) and the processing circuit by wire allows the processing circuit to be spaced from the sensor head (probe) in use.

[0208] With this magnetometer, the sensor head (probe) will be used as a magnetic probe by placing it in the vicinity of the magnetic fields of interest.

[0209] FIG. 2 shows an improvement over the FIG. 1 arrangement, which uses in particular the technique of gradient subtraction to try to compensate for background noise. (Other techniques could, however, be used). In this case, an inverse coil 44 is used to attempt to subtract the effect of the background noise magnetic field from the signal detected by the probe coil 40. The inverse coil 44 will, as is known in the art, be equally sensitive to any background magnetic field, but only weakly sensitive to the subject's magnetic field. The inverse coil 44 can be accurately matched to the pickup coil 40 by, for example, using a movable laminated core to tune the performance of the inverse coil to that of the pickup coil 40.

[0210] FIG. 3 shows an alternative gradient subtraction arrangement. In this case, both coils 40, 44 have the same orientation, but their respective signals are subtracted using a differential amplifier 45. Again, the best operation is achieved by accurately matching the coils and the performance of the detection circuits 41. Again, a movable laminated core can be used to tune the performance of one coil to match the performance of the other.

[0211] FIG. 4 shows a further preferred arrangement. This circuit operates on the same principle as the arrangement of FIG. 3, but uses a more sophisticated method of field cancellation, and passive coil matching. In particular, a known global magnetic field 44 is introduced to both coils 40, 44 to try to remove background magnetic field interference.

[0212] In this circuit, the outputs from the detection circuits 41 are passed through respective amplifiers 47, 48, respectively, before being provided to the differential amplifier 45. At least one of the amplifiers 47, 48 is tuneable. In use, a known global field 46, such as 50 or 60 Hz line (and harmonics) noise, or a signal, such as a 1 kHz signal, applied by a signal generator 49, is introduced to both coils 40, 44. The presence of a signal on this frequency on the output of the differential amplifier 45, which can be observed, for example, using an oscilloscope 50, will then indicate that the coils 40, 44 are not matched. An amplifier control 51 can then be used to tune the tuneable voltage controlled amplifier 48 to eliminate the global noise on the output of the differential amplifier 45 thereby matching the outputs from the two coils appropriately.

[0213] Most preferably in this arrangement, a known global field of 1 kHz or so is applied to both coils, so as to achieve the appropriate coil matching for the gradient subtraction.

[0214] FIG. 5 shows a further variation on the FIG. 4 arrangement, but in this case using active coil matching. Thus, in this arrangement, the outputs of the coils 40, 44 are again channelled to appropriate detection circuits 41, and then to respective amplifiers 47, 48, at least one of which is tuneable. However, the tuneable amplifier 48 is tuned in this arrangement to remove the common mode noise using a lock in amplifier 52 or similar voltage controller that is appropriately coupled to the output from the differential amplifier 45 and the signal generator 49.

[0215] The above embodiments of the present invention show arrangements in which there is a single pickup coil that may be used to detect the magnetic field of the subject's heart. In these arrangements, in order then to make a diagnostic scan of the magnetic fields generated by a subject's heart, the single pickup coil can be moved appropriately over the subject's chest to take readings at appropriate spatial positions over the subject's chest. The readings can then be collected and used to compile appropriate magnetic field scans of the subject's heart.

[0216] It would also be possible to arrange a plurality of coil and detection circuit arrangements, e.g. of the form shown in FIG. 1, in an array, and to then use such an array to take measurements of the magnetic field generated by a subject's heart. In this case, the array of coils could be used to take readings from plural positions over a subject's chest simultaneously, thereby, e.g., avoiding or reducing the need to take readings using the same coil at different positions over the subject's chest.

[0217] FIGS. 6A and 6B show suitable coil array arrangements that have an array 60 of 16 detection coils 61, which may be then placed over a subject's chest to measure the magnetic field of a subject's heart at 16 sampling positions over the subject's chest. FIG. 6A shows a regular rectangular array and FIG. 6B shows a regular hexagonal array. In these cases, each coil 61 of the array 60 should be configured as described above and connected to its own respective detection circuit (i.e. each individual coil 61 will be arranged and have a detection circuit connected to it as shown in FIG. 1). The output signals from the respective coils 61 can then be combined and used appropriately to generate a magnetic scan of the subject's heart.

[0218] Other array arrangements could be used, if desired, such as circular arrays, irregular arrays, etc.

[0219] More (or less) coils could be provided in the array, e.g. up to 50 coils, or more than 50 coils. For example, where it is desired to measure the magnetic field of a different region of a subject's body (i.e. other than the heart), then an increased number of coils may be provided so as to provide an appropriate number of sampling points and an appropriate spatial coverage for the region of the subject's body in question.

[0220] It would also be possible in this arrangement to use some of the coils 61 to detect the background magnetic field for the purposes of background noise subtraction, rather than for detecting the wanted field of the subject's heart. For example, the outer coils 62 of the array could be used as background field detectors, with the signals detected by those coils then being subtracted appropriately from the signals detected by the remaining coils of the array. Other arrangements for background noise subtraction would, of course, be possible.

[0221] It would also be possible to have multiple layers of arrays of the form shown in FIG. 6, if desired. In this case, there could, for example, be two such arrays, one on top of each other, with the array that is closer to the subject's chest being used to detect the magnetic field generated by the subject's heart, and the array that is further away being used for the purposes of background noise detection.

[0222] To measure the magnetic fields generated by the heart, the above arrangements can be used to compile magnetic field scans of a subject's heart by collecting magnetic field measurements at intervals over the subject's chest. False colour images, for example, can then be compiled for any section of the heartbeat, and the scans then used, for example by comparison with known reference images, to diagnose various cardiac conditions. Moreover this can be done for significantly lower costs both in terms of installation and on-going running costs, than existing cardiac magnetometry devices.

[0223] FIG. 7 shows a typical ECG trace and the conventional labelling of the typical elements present in the ECG trace. Similar elements also occur in the MCG trace and the correspondence between the two has led to researchers using the same labelling convention.

[0224] As shown in FIG. 7, the ECG trace comprises a repeating P-P interval comprising a so-called P-wave, followed by a P-R (or P-Q) segment (where the combination of the P-wave and the P-R (or P-Q) segment is referred to as the P-R (or P-Q) interval), followed by a QRS complex, followed by an S-T segment, followed by a T-wave (where the combination of the S-T segment and the T-wave is referred to as the S-T interval, and the combination of the QRS complex and the S-T interval is referred to as the Q-T interval), followed by a T-P segment. Each of the features within the ECG can have diagnostic importance.

[0225] FIG. 8 shows an exemplary arrangement of the magnetometer as it is envisaged it may be used in a hospital, for example. The magnetometer 30 is a portable device that may be wheeled to a patient's bedside 31 where it is then used to take a scan of the patient's heart (e.g.). There is no need for any magnetic shielding, cryogenic cooling, etc. The magnetometer 30 can be used in the normal (non-shielded) ward environment. (Magnetic shielding and/or cooling could, however, be provided if desired.)

[0226] As used herein, a magnetometer or other apparatus in a magnetically shielded environment may comprise a magnetometer or other apparatus that is either arranged in a specially designed room or enclosure. In such arrangements, both the subject being measured and the equipment doing the measuring are contained within the same shielded enclosure. By contrast, as used herein, a magnetometer or other apparatus in a non-magnetically shielded comprises a magnetometer or other apparatus for which no external piece or pieces of apparatus are used to protect the subject being measured, nor the equipment doing the measuring.

[0227] The magnetometer system can be used in an analogous manner to detect and analyse other medically useful magnetic fields produced by other regions of the body, such as the bladder, abdomen, chest, head, brain, one or more foetuses, a muscle, etc.

[0228] Despite the success of MCG as a diagnostic aid, the hospital environment (such as an emergency department) can present challenges which interfere with the acquisition of acceptable MCG data.

[0229] Three main types of noise can cause such interference: homogenous noise (e.g. the earth's magnetic field), stochastic noise (e.g. white noise, coloured noise), and background noise (e.g. power line disturbances with power spectrum peaks at 50 or 60 Hz (and harmonics), vibrations of the system itself, etc.). Background noise can often exceed the MCG signal by orders of magnitudes and can vary in time, which makes its removal a challenging problem. The present embodiment is directed, in particular, to the removal of background noise components.

[0230] Background noise components may be characterised as being either low, medium or high frequency noise. Low frequency noise (0.1 to 1 Hz) is typically due to moving elevators, metal doors, metal chairs or other metallic objects. High frequency noise (>20 Hz) is mostly due to power supplies, monitor frequencies, or other electronic devices. Vibrations of the system itself cause disturbances in the middle frequency noise range (1 Hz to 20 Hz).

[0231] The Applicants have recognised that, where it is desired to take a scan of a patient's heart while the patient is on a hospital bed 31, noise can be produced due to coupling of residual magnetism in the steel frame of the hospital bed 31, which can vibrate due to the motion of the patient as their heart beats.

[0232] These transient noise pulses consist of a relatively short sharp initial pulse followed by decaying low-frequency oscillations. The initial pulse is due to the heart beat (systole), whereas the oscillations are due to the resonance of the system (bed and patient) excited by the initial pulse.

[0233] Once picked up by the magnetometer device 30, this transient noise can make it difficult to assess the quality of the captured data during the scanning process. It also makes it difficult to extract the useful undistorted magnetocardiograph data that is required for an accurate diagnostic.

[0234] Ballistocardiographic noise may be caused by vibration of the bed 31, where the vibration is correlated with the recoil forces of the body in reaction to cardiac ejection of blood into the vasculature.

[0235] Other sources of synchronised biological noise include, for example, seismocardiographic noise which may be caused by local vibrations of the chest wall in response to the heartbeat, as well as breathing, and changes in the position of the subject's body on the bed 31, e.g. due to talking, fidgeting, etc., that can cause vibration of the bed 31 which in turn can produce synchronous magnetic noise in the magnetocardiograph.

[0236] These biological sources of synchronous noise should be contrasted with other sources of noise such as nearby vibrations (e.g. vibrating lift shafts, large objects being dropped or moved, etc.). Although these other noise sources can cause vibration of the bed 31 which in turn can produce magnetic noise in the magnetocardiograph, generally such noise is not synchronised with the motion of region of the subject's body in question (e.g. heartbeat) and can therefore be reduced using averaging (over a long enough period of time).

[0237] Other support structures such as beds, chairs, etc., may also give rise to biological synchronous magnetic noise, e.g. where the support structure comprises a material that can produce a magnetic field, such as high permeability materials, and/or high conductivity materials.

[0238] High permeability materials include, for example, iron, steel, nickel, and various alloys thereof. High permeability materials comprise materials that can be magnetised and/or that can attract a magnet, e.g. ferrous materials that can generally hold and maintain a permanent magnetic field of their own (i.e. that are ferromagnetic). High permeability materials react strongly to applied magnetic fields, and are typically electrically conductive.

[0239] Low permeability, high conductivity materials comprise highly conductive materials that do not have a magnetic field of their own but can produce a response in reaction to changes in applied fields (e.g. paramagnetic or diamagnetic). Low permeability, high conductivity materials comprise, for example, stainless steel, aluminium, graphene, etc.

[0240] With respect to electrically conductive materials, if a conductive material is stationary and an applied field is stationary, it will not produce a magnetic field. However, if a conductive material is stationary and an applied magnetic field is moving, electric (eddy) currents may be induced in the material, which have their own corresponding magnetic fields. Equally, if a conductive material is moving and the magnetic field is stationary, electric (eddy) currents may be induced in the material, which have their own corresponding fields.

[0241] By contrast, low permeability, low conductivity materials may comprise, for example, wood, most plastics, ceramics, fiberglass, etc. Low permeability, low conductivity materials are (non-conductive) electrical insulators with low permeability and conductivity and do not provide any magnetic interference, i.e. do not produce magnetic fields even if they vibrate.

[0242] FIG. 9A shows cycle averaged MCG data for a healthy subject captured by a 37-channel magnetometer device in an unshielded environment on a wooden bed, and FIG. 9B shows corresponding data for a bed made of ferrous (magnetic) material. The averaged signal was filtered using a notch filter to remove power line noise followed by a finite impulse response (FIR) low pass filter.

[0243] The peaks visible in the middle of FIG. 9A correspond to the QRS section of the cardiac cycle (more specifically the time derivative of the magnetic field of the cardiac muscle and not the static field). These represent the highest signal to noise ratio part of the signal.

[0244] Distortions to the MCG data caused by the magnetic bed material are evident in FIG. 9B, making it impossible to extract useful information even from the QRS section.

[0245] FIG. 10A shows a log periodogram of raw MCG signals captured by a 37-channel magnetometer device in a non-shielded environment for a noise scan (i.e. without a subject present under the scan head). FIG. 10B shows a log periodogram of MCG signals for a healthy subject captured by a 37-channel magnetometer device in an unshielded environment for a subject on a wooden bed, and FIG. 100 shows a corresponding signal for a subject on a bed having ferrous (magnetic) material. A 8192-point Welch periodogram was used with a hamming window and a 4096-point overlap for the spectral calculations.

[0246] The noise peaks visible in FIG. 10A are due to the mains power supply and its subharmonics (50 Hz, 25 Hz, 16 Hz etc.).

[0247] The contribution of the healthy subject MCG signal to the spectral content in FIG. 10B appears at approximately 4 Hz, 10 Hz and 33 Hz, while the contribution of the ballistic effects due to the bed material are clearly visible in the spectral content of FIG. 100.

[0248] These ballistic effects fall in the frequency range <10 Hz, making it difficult to extract useful information from the MCG signal.

[0249] The Applicants attempted to use a number of techniques to try to diminish or remove such unwanted noise from the MCG signal.

[0250] One such technique is nonlinear denoising (NLD) in state space. Nonlinear denoising operates on the reconstructed state space of the time series which represents the dynamical properties of the observed system. Background noise such as powerline disturbances fills a subspace in the state space which can be separated from the MCG manifold. This is done by recording the disturbances using an additional sensor, followed by a projection onto the noise subspace, followed by a subtraction from the original signal.

[0251] This approach requires the noise to be at least approximately deterministic, and therefore works well in removing powerline disturbances. However, the Applicants have found that this approach fails to remove transient noise caused by the bed vibrations, i.e. due to their nondeterministic nature.

[0252] In contrast with this, and in accordance with the present invention, the Applicants have found that a particular band-pass filter arrangement (as described further below) can be used to successfully separate the ballistic effects caused by the use of magnetic beds from the QRS complex. This allows a useful signal to be extracted from the corrupted MCG signal.

[0253] The Applicants have found, in particular, that a bandpass filter having a passband around 8-45 Hz can be used to separate the MCG signal from the ballistocardiographic (BCG) noise and background noise. The filter is designed to significantly reduce the impact of the ballistic effects from the measured signal, specifically the QRS region. The filter is a band pass filter constructed as combination of a high pass filter (removing ballistic effects <10 Hz), and a low pass filter (removing background noise >50 Hz).

[0254] FIG. 11 illustrates an ideal band-pass filter. An idealised filter is one that removes all frequency components above a given cutoff frequency, without affecting lower frequencies, and has linear phase response. All frequencies within the passband 10-50 Hz, are passed with unity amplitude, while all other frequencies are blocked. The passband is perfectly flat, the attenuation in the stopband is infinite, and the transition between the two is infinitely small. The filter's impulse response is a sinc function in the time domain, and its frequency response is a rectangular function. It is an ideal low-pass filter in the frequency sense, perfectly passing low frequencies, perfectly cutting high frequencies, and thus may be considered to be a brick-wall filter.

[0255] In the present embodiment, in order to approximate such an ideal filter, two windowed-sinc filters are combined to construct a band-pass filter that can separate the MCG signal from the BCG signal and background noise. This allows for an efficient separation of the QRS-complex from the ballistic effects and other background noise interferences, without phase distortions.

[0256] The filter is configured such that it removes all frequency components below a cut-off frequency f.sub.c1 and above a cut-off frequency f.sub.c2 without affecting frequencies in between. The filter is designed as the difference of two windowed-sinc filters whose cut-off frequencies are f.sub.c1 and f.sub.c2. The filter is able to significantly reduce the impact of the ballistocardiographic effects (BCG), on the MCG signal, specifically the depolarisation (QRS) section.

[0257] In the present embodiment, the signal from the detector is firstly digitised, e.g. using a 4-bit 37-channel 2400 kS/s ADC. MCG signals are baseline corrected and averaged to increase signal-to-noise level. Data are averaged centring on the R wave peak, which is obtained using an accompanying ECG signal. The averaged signal can be windowed, using a suitable window function, to reduce abruptness.

[0258] FIG. 12A shows a filter kernel for a windowed-sinc filter, and FIG. 12B shows the frequency response (with a cut-off frequency of 45 Hz and M=2400). The filter acts as a low pass filter.

[0259] In the time domain, the filter kernel is a modification of the sinc function. The frequency response of the windowed-sinc filter is rectangular. This corresponds to the fact that the sinc filter is the ideal (brick-wall, i.e. rectangular frequency response) low-pass filter.

[0260] The windowed-sinc filter kernel with cut-off frequency f.sub.c1 is given by:

[00002]$h_{f_c\left[i\right]}=K.Math..Math.\frac{\sin \left(2.Math..Math..Math.f_c\left(i-M/2\right)\right)}{i-M/2}.Math.w\left[i\right]$

where w[t] is a window function centred on t=0, and where i ranges from 0 to M. The constant K is a normalisation factor chosen to provide a unity gain at zero frequency. The cut-off frequency f.sub.c is expressed as a fraction of the sampling rate (a value between 0 and 0.5). The length of the filter kernel is determined by M, which must be an even integer.

[0261] The Applicants have found that, for the purposes of the present embodiment, the choice of window function is important. This involves a trade-off between roll-off and stop-band attenuation. Possible choices for the window function include the Hamming window, the Blackman window, the Bartlett window, and the Hanning window.

[0262] The Applicants have found in particular that, for the purposes of the present embodiment, the Blackman window is particularly suitable. This window has a slower roll-off compared with other window functions such as the Hamming window. However, the Blackman window has an improved stopband attenuation, and a lower passband ripple.

[0263] A Fourier transform may be performed to convert a signal in the time domain to its frequency domain counterpart. To calculate the output of the filter in the time domain a convolution may be performed, and in the frequency domain a point-by-point multiplication may be performed.

[0264] The length of the filter kernel M determines the transition bandwidth of the filter in the frequency domain, BW (the transition bandwidth is measured from where the curve leaves a value of one, to where it almost reaches zero), and is expressed as a fraction of the sampling rate (i.e. a value between 0 and 0.5). The trade-off between the computation time (which depends on the value of M) and the filter sharpness (the value of BW) can be expressed through the approximation:

[00003]$M\frac{4}{\mathrm{BW}}$

[0265] As such, the sharper the filter is (the smaller the transition bandwidth BW), the longer is the time required to perform convolution in the time domain.

[0266] In the present embodiment, the signal is averaged into a single 1s cycle (e.g. 2400 samples at 2400 Hz sampling rate, i.e. one second of data which is approximately equivalent to a single heartbeat). The length of M is then maximised M=2400 to minimise the transition bandwidth BW. This means that BW 4/24000.00167, or BW2 samples.

[0267] The filter of the present embodiment is constructed as the difference of two windowed-sinc filters whose cut-off frequencies are f.sub.c1 and f.sub.c2. Since taking linear combinations in the frequency domain is equivalent to taking the same linear combinations in the time domain, the filter is constructed as the difference of two windowed-sinc filters:

g[i]=h.sub.f.sub.c2[i]h.sub.f.sub.c1[i]

[0268] This provides a band pass filter which only passes frequencies between f.sub.c1 and f.sub.c2. If f.sub.c2 is set as 0.5, a high pass filter is obtained, and if f.sub.c1 is set as 0.0, a low pass filter is obtained.

[0269] FIG. 13A shows the filter kernel and FIG. 13B shows the frequency response of the difference of two windowed-sinc filters with cut-off frequencies f.sub.c1=0.0033 (8.0 Hz), f.sub.c2=0.01875 (45.0 Hz) and M=2400. The filter acts as a band pass filter.

[0270] The filter can be applied in either the time domain or the frequency domain to effectively separate the repolarisation (QRS section) of the MCG signal from the BCG effects and background noise.

[0271] FIG. 14 illustrates example MCG data for a healthy subject on a metal bed obtained in a non-shielded room.

[0272] FIG. 14A shows the obtained averaged MCG signal, and FIG. 14B shows the frequency spectrum (Fourier spectrum) of the signal. FIGS. 14C and 15D respectively show the filter kernel and frequency response of the difference of two windowed-sinc filters with cut-off frequencies of f.sub.c1=0.0033 (8.0 Hz) and f.sub.c2=0.01875 (45.0 Hz) with M=2400.

[0273] FIGS. 14E and 14F show the time series and its corresponding Fourier spectrum resulting from the present filtering method (solid line). FIG. 14E also shows the result of applying a filter with cut-off frequencies at 2 Hz and 45 Hz to the signal of FIG. 14A (dashed line), where the presence of ballistic noise is evident.

[0274] FIG. 15 illustrates the effectiveness of the windowed sinc filter in removing the ballistocardiographic effects due to the ferrous (magnetic) material of the bed. FIG. 15A shows the data of FIG. 9B, and FIGS. 15B and 15C show the data after the filter has been applied. A windowed-sinc filter kernel with a Blackman window and cut-offs at 8 Hz and 45 Hz was used. The ballistocardiographic effects have been removed from the signal and the useful MCG features (namely the QRS section) is now visible and can be used to extract medically useful information.

[0275] FIG. 16 shows a sequence of data processing steps in accordance with the present embodiment.

[0276] A sensor 40 and a digitiser 42 are used to obtain a signal 101. The signal is then averaged 102 over plural periods. This involves using a trigger such as the ECG to determine the plural repeating periods of the signal. Data is taken from the target waveform in each of plural windows around each of plural triggers. Several subsequent windows are averaged to remove random noise.

[0277] Filtering 103 is then applied to remove the noise that cannot be removed by averaging, i.e. the bed noise and other background noise as described above.

[0278] Following any additional data processing 104, diagnostic parameter extraction 105 may be performed, and used for analysis 106.

[0279] Some examples of medically useful signals that may be analysed are (i) S-T baseline shifts (STEMI) e.g. S-T elevated myocardial infarction; and (ii) R-S transition rate, e.g. bundle branch block. However, in general any of the signal features described herein may have diagnostic importance and may be used for analysis.

[0280] It can be seen from above that the present invention provides an improved magnetometer system for medical use. This is achieved, in the preferred embodiments of the present invention at least by filtering a signal or signals from using a filter that is configured to attenuate synchronised, e.g. ballistocardiographic noise.