Title

Adaptive Real-time Line Noiss Suppression for Electrical or Magnetic Physiological Signals

Field of the Invention

This invention relates to methods for analysis of outputs of sensors, in particular, physiological sensors, and more particularly, sensors for electroencephalogram (EEG) and magnetoencephalogram (MEG) measurements.

Background

Currently, most, if not all, commercially available devices for recording physiological signals are subject to line noise, the spurious electrical signals derived from the electrical supply to a sensor device, the noise signals often masking the electrical signals attributable to the physiological process of interest. Line noise appears at a frequency of 60 Hz in North America and 50 Hz in other regions throughout the world in accordance with the frequency of the local electrical current. Line noise may be conducted or radiated in origin. Examples of devices affected by line noise include, but are not limited to, devices and systems for recording EEG, MEG, electromyogram (EMG), electrocardiogram (EKG or ECG), ballistocardiogram (BKG), electrooculogram (EOG), electrodermalgram (EDG), electrodermal activity (EDA), αr eyelid movement (ELM).

It is Known in the art to minimise line noise from physiological signal measurements of interest simply by applying a notch filter, which essentially comprises of a combination of a steep low-pass filter and a high-pass filter. While effective, this type of signal filtration can distort the signal of, for example, an EEG in spectral proximity to the effective range of the' notch filter. Consequently, high-frequency cortical oscillations occurring in the upper gamma band range (50-60 Hz) are compromised by such notch filtering solutions. In addition to such standard filtering approaches, other attempts to remove line noise from EEG data, for example, include spatial implementations of principal and independent components analysis and wavelet de-noising. While such approaches can be used effectively to minimize line noise offline, they are typically not applied under online, real-time recording conditions.

What is needed is a method for line-noise suppression that removes line noise from sensor signals effectively but does not distort the remaining signals that represent the physiological signal of interest. Ideally, the line-noise suppression would enable the signal analysis to occur no later than a short time after the signals are collected, effectively, in "real-time" or "near real-time".

Brief Description of the Figures

Figure 1 shows a flow diagram of the steps used in the method of acquiring and analyzing biosignals.

Figure 2 shows a screen display of an example of electrophysiological signals acquired and analysed according to the invention.

Figure 3 shows a screen display of an example of EEG signals acquired and analysed according to the invention.

Summary of the Invention It is an object of the invention to provide a method and apparatus for acquiring physiological signals from a subject and removing line noise from the signals with minimal distortion of the signal or signals of interest. It is a further object of the invention to provide a method and apparatus that is operable in real-time or near real-time. Other objects will become evident on reading the detailed description of the invention. It will be understood that the scope of the invention is not limited to the embodiments described in the description but that the scope includes embodiments within the scope of the appended claims.

The present invention provides a method of overcoming the contamination of desired physiological signals with periodic, replicable signals, or noise, caused by inherent electrical characteristics of the electrical supply to electrical measuring devices. The method of the invention exploits the periodic and spectrally stationary nature of line noise, which is spectrally constant at its frequency of origin, but may vary over time with respect to location-specificity and time-varying amplitude. The method can be advantageously implemented in computer software for easy calculation and display of calculated results for interpretation and use of the resulting relatively uncontaminated signals, The method can be applied in applications wherein measurements are made of physiological parameters of humans or any other animal, as appropriate.

In one aspect, the invention provides a method for processing data acquired from physiological sensors, comprising the steps of collecting raw sensor data in a file, said data representing at least one electrophysiological signal; selecting a time interval that is a whole-number multiple of the period of the waveform of said at least one signal; calculating an average value, of the data for each of a series of consecutive time periods in the data file wherein said time period is a whole-number multiple of the time period of an artefact waveform; calculating a standard cross- correlation value for the calculated average from the sampling period according to the spectral peak and the raw data collected over the same time Interval; and subtracting the average calculated according to the sampling period from the raw data in each time period.

In another aspect, the invention provides a method for acquiring and processing physiological signals acquired from a subject, comprising the steps of locating at least one sensor to acquire a least one physiological signal from a subject; acquiring a least one physiological signal from said at least one sensor; selecting a time interval that is a whole-number multiple.of the period of the waveform of said at least one signal; transforming the at least one signal into raw data in a format suitable for data storage; storing the raw data at least one signal in at least one data storage means; and for each sensor, calculating an average value of the sensor output for each of a series of consecutive time periods in the data file wherein said time period is a whole-number multiple of the time period of an artefact waveform, providing a dynamic average value for the time periods; calculating a standard cross-correlation value for the calculated average for each sensor for each of the series of time periods and the raw data measured and stored over the same time interval; and subtracting the dynamic average from the raw data in each time period.

In a further aspect, the invention provides a method for processing data acquired from physiological sensors, comprising the steps of collecting raw sensor data in a file, said data representing at least one electrophysiological signal; identifying the spectra! peak of an artefactual waveform in the at least on electrophysiological signal: calculating a sampling period according to the spectral peak; calculating an average value of the data for each of a series of consecutive time periods in the data file wherein said time period is a whole-number multiple of the sampling period of the artefact waveform; calculating a standard cross-correlation value for the calculated

average from data for each of the series of consecutive time periods and the raw data collected in step from a sensor over th& same time interval; and subtracting the average calculated for each of the series time period from the raw data in each time period.

Preferably, the method inoludes.a step of determining the sampling period according to the time period during which the spectral peak exceeds a threshold. Preferably, the at least one physiological signal comprises of a continuous stream of measurable input. Preferably, the method includes the step of determining the shift delay at the maximum value in the cross-correlation function and tϊmeshifting the artefact average correspondingly. Preferably the method includes the step of creating and displaying a corrected data set. Preferably, the method includes the step of storing the calculated data in a computer file. Preferably, the method includes displaying the raw, uncorrected data. Preferably the waveform of the electrophysiological signal is any one of sinusoidal, square or triangular in graphical shape. Preferably, the steps of the method are carried out in real-time or near-real time.

In a still further aspect, the invention provides apparatus for acquiring and processing physiological signals from a subject including at least one sensor for acquiring at least one signal and at least one microprocessor means for processing said at least one signal, said at least one microprocessor means including means for storing a whole number multiple of an artefact waveform for calculating the line-noise component of data derived from said at least one sensor.

Detailed Description of the Figures and Preferred Embodiment of the Invention

The method of the invention includes the steps of acquiring a real-time (or near real- time) signal or signals using an electrical sensor device or devices having a source or sources of electrical power, followed by the transforming the acquired signal(s) according to the algorithm of the invention. It will be understood that the method can be used for one or more sensors simultaneously or in sequence.

Characteristics of the real-time data acquisition step include the following. Data acquisition should be a continuous stream of measurable impulses comprising the targeted physiological signal from a sensor located adjacent, or in proximity to, the subject. Data must be stored in raw (uncorrected) format. It may be displayed and stored in the modified (corrected) format. An underlying assumption for the use of

the algorithm to analyze signals is that the line noise or other such external continuous periodic source is defined to mean "a continuously or episodically present repetitive waveform" such as, for example, any one of a sinusoid, square wave, or triangular wave, or any other continuously repetitive ariefactual activity measured in the physiological parameter of interest, where "artefactual" is defined to mean any activity that is not the targeted signal of interest.

Figure 1 shows the steps in the method of the invention, including, but not limited to, the following steps. It will be understood that the method preferably includes the additional steps of storing and/or displaying raw data representing acquired signals and corrected data but those steps need not to be practised. The method includes the step of selecting a time interval that is a whole-number multiple of the period of the waveform of the target sensor signal 1. The sensor output is sampled, preferably continuously 2, resulting in a stream of raw data. The sensor output is recorded in a data file as it is sampled 3. An average value of the sensor output is calculated for each of a series of consecutive time periods in the data file 4, providing a dynamic average value for the time periods. A standard cross-correlation value is calculated from the calculated average from box 4 and the stored measured raw data 3 over the same time interval 5. Optionally, a shift delay at the maximum value in the cross- correlation function may be calculated and the artefact average is time-shifted correspondingly 6, if a shift is required. The dynamic average of the artefact is subtracted from the raw data in each time period 7. Preferably the method includes creating, displaying and storing a corrected data set 8 comprised of the results from calculating the average 4 concurrentiy with the raw data set 3. Preferably the raw, uncorrected data is displayed or stored 9.

The method of the invention includes the steps of real-time data acquisition using an electrical sensor device having a source of electrical power and the transformation of acquired data according to the algorithm of the invention.

The method of the invention includes that the steps in boxes 4 to 7 in Figure 1 may be repeated for successive time periods based on the period of the artefact waveform. The successive time periods must be whole-number multiples of the period of the artefact waveform. Low multiples of the time period of the artefact are preferable as they provide for rapid correction and rapid updating of the average artefact used for subtraction purposes. According to the invention for step 1 , for

example, for a source of electrical current at 50 Hz powering the electrical sensor device, this would be 20 ms (or any whole number multiple of 20). For a 60 Hz artefact, this interval would be a whole number multiple of 16.666 ms (or any whole- number multiple). Alternatively, a common value of 500 ms could be used, which would provide an interval. that is a whole number multiple of the period for either 50 or 60 Hz. The whole number may be hard-coded in hardware. Alternatively, it may be a user-controlled parameter in computer software. Alternatively, it may be dynamically adjusted to provide the maximum suppression of line (mains) based on a real-time spectral amplitude/power measure of the targeted artefact.

According to the invention, for steps in boxes 2-6 the physiological response is continuously sampled and an average signal is calculated for the physiological activity recorded from consecutive periods. The number of sampled periods used to generate the average may be fixed (e.g., 10 sampled periods) or it may be a user- determined value. The average calculated according to box 4 is dynamically updated so that as the number of sampled periods for the average is fulfilled, the first sample in a period is dropped and replaced by the next sampled period. Based on this procedure, activity that is time-locked and/or phase-locked to the spectral period of the artefact signal is maintained with amplitude equal to the average of the sampled epochs when in the average, while all other non-time-locked or phase- locked activity to the period of the artefact is diminished because of the absence of phase coherence.

The Invention includes that a dynamic average is continuously subtracted from the raw data (buffered or streamed in real-time) and sent to a corrected data set collected concurrently with the raw data set This corrected data set maybe used for display purposes only. Alternatively, it may be saved concurrently with, or instead of, the raw data set.

According the invention, as shown in box 7 of Figure 1 , to avoid single point offsets of the actual data to the sampled average used for correction, a standard cross- correlation method based on the best correlation match for the sampled average and the raw data over the same time interval, is used to time-shift the sampled average to obtain the best correction. If there ts no time-shift in the peak of the cross- correlation function, the average would be applied directly to the matching segment of buffered data. However, it there is a time-shift in the peak of the cross-correlation

function, the average is shift by the number of data points equivalent to the time shift in the cross-corralation analysis. Once this time-shift is performed, the averaged artefact is subtracted from the raw data.

Alternate embodiment of the invention may include a spectral detection method that automatically identifies the spectral peak of the artefact in the physiological data and then calculates the appropriate sampling period to apply the correction. Alternatively, the method may include a whole-number multiple of the sampling period for use in the calculation. This embodiment could also include a threshold detection measure for spectral amplitude and for a minimum time period before the "repetitive" activity would be regarded as artefactual to avoid removing, for example, real EEG signals, such as alpha oscillations, for example. Similar corrections can also be applied offline.

Other embodiments of the invention may include correction of electrical signals from sensor devices for any repetitive electrical source producing a well characterized or deterministic artefact signature. Such examples would specifically include the artefact produced by trans-cranial magnetic stimulation or electrical/mechanical somatosensory stimulation, electrical pump noise associated with the delivery of coolant in MRI (magnetic resonance imaging) environments or other similar sources of artefact signal.

Illustrations of the adaptive noise removal for simulated and real data are shown in

Figures 2 and 3.

Examples

The results for the simulated data are shown in Figures 2 and 3. The uncorreded data 10 is shown in Figure 2A, corrected (except for one channel 11) shown in

Figure 26, and the spectral peak 12 of the uncorrected versus corrected are shown in the graph Figure 2C With simulated data there is perfect correction, with no residual evidence of the continuously periodic waveform shown in Figure 2A. The reduction in the simulated line noise is essentially infinite.

Correction of real data, in this case from an EEG recording, is shown in Figure 3C. The raw data 13 is in Figure 3A, the corrected data 14 in Figure 3B, and the spectral comparison of the data 15 is shown in Figure 3C. The reduction in spectral energy at

50 Hz line frequency is from over 4000 microvolts to less than 100 microvolts. The residual 50 Hz line frequency noise is approximately 1/800 the amplitude of the uncorrected data or nearly 60 dB of line (mains) suppression.