ECG sensing with noise filtering

Abstract

Various embodiments are directed to signal processing. In accordance with example embodiments, methods and apparatuses involve using at least two electrodes that sense an ECG signal. A denoising module is communicatively coupled to the at least two electrodes, and receives the ECG signal sensed by the sensing electrodes. The denoising module includes circuitry that conditions and digitizes the ECG signal, and a computing circuit that processes the digitized ECG signal to denoise the ECG signal. A communications circuit generates a communication including the denoised ECG signal for access by a remote device.

Claims

1. An apparatus comprising: digitizing circuitry configured and arranged to generate a digitized ECG signal from an analog ECG signal; and a computing circuit coupled to receive the digitized ECG signal from the digitizing circuitry and configured and arranged to generate a denoised ECG signal from the digitized ECG signal, the denoised ECG signal having, relative to the digitized ECG signal, an improved signal-to-noise ratio of at least 15 dB as measured using the ANSI EC 57 standard, and a quality of signal reconstruction greater than 95%.

2. The apparatus of claim 1, wherein the computing circuit is configured and arranged to generate the denoised ECG signal from a digitized ECG signal that has a signal-to-noise ratio of 0 dB and that includes a substantially noise-free ECG with white noise added as prescribed in the ANSI EC 57 standard.

3. The apparatus of claim 2, wherein the substantially noise-free ECG is an ECG having a signal-to-noise ratio of greater than 30 dB according to the ANSI EC 57 standard.

4. The apparatus of claim 1, wherein the computing circuit is configured and arranged to generate the denoised ECG signal by: decomposing the digitized ECG signal from a first domain into subcomponents in a second domain that is different than the first domain, and selecting and combining ones of the subcomponents based upon a time-based distribution of the subcomponents.

5. The apparatus of claim 4, wherein the computing circuit is configured and arranged to select and combine the ones of the subcomponents by: identifying target subcomponents of the digitized ECG signal that are associated with a desired ECG signal based upon the time-based distribution of the subcomponents, and reconstructing a denoised ECG signal in the first domain from at least two of the identified target subcomponents.

6. The apparatus of claim 1, wherein the computing circuit is a portable or desktop computer.

7. The apparatus of claim 1, wherein the computing circuit is a mobile telephone configured and arranged to receive the digitized ECG signal.

8. The apparatus of claim 1, wherein the quality of signal reconstruction (QSR) is defined as $\mathrm{QSR}=100{\%}^{*}\left(1-\frac{\underset{i}{.Math.}{\left(x_{\mathrm{cl}}^i-x_{\mathrm{den}}^i\right)}^2}{\underset{i}{.Math.}{\left(x_{\mathrm{cl}}^i\right)}^2}\right),$ where x.sub.cl is the digitized ECG signal and x.sub.den is the denoised ECG signal.

9. The apparatus of claim 1, wherein the computing circuit is configured and arranged to generate the denoised ECG signal from an analog ECG signal that is obtained via at least two electrodes and subsequently digitized.

10. An apparatus comprising: at least two electrodes configured and arranged to sense an ECG signal; and a computing circuit configured and arranged to generate a denoised ECG signal from the sensed ECG signal having, relative to the sensed ECG signal, an improved signal-to-noise ratio of at least 15 dB as measured using the ANSI EC 57 standard, and a quality of signal reconstruction greater than 95%.

11. An apparatus comprising: an electrode circuit attached to an outer surface of a hand-held device, the electrode circuit configured and arranged to sense heart-related signals upon contact with a surface of a body of a living being and to convert the sensed heart-related signals to ECG electrical signals; and a converter circuit integrated with and electrically connected to the electrode circuit, said converter circuit configured and arranged to receive the ECG electrical signals generated by the electrode circuit and to generate a digitized ECG signal based upon the ECG electrical signals; and a computing circuit communicatively coupled to the converter circuit and configured and arranged to generate a denoised ECG signal from the digitized ECG signal, the denoised ECG signal having, relative to the digitized ECG signal, an improved signal-to-noise ratio of at least 15 dB as measured using the ANSI EC 57 standard, and a quality of signal reconstruction greater than 95%.

12. The apparatus of claim 11, wherein the hand-held device is a smart phone.

13. The apparatus of claim 11, wherein the computing circuit is included in the converter circuit and is configured and arranged to communicate a denoised ECG to said hand-held device.

14. The apparatus of claim 11, wherein the computing circuit is incorporated into the hand-held device.

15. The apparatus of claim 11, wherein the hand-held device is configured and arranged to diagnose or screen cardiac-related disease.

16. The apparatus of claim 10, wherein the at least two electrodes are configured and arranged to contact and adhere to a surface of a patient's chest, and include an adhesive-backed patch as part of the electrodes.

17. The apparatus of claim 10, wherein the at least two electrodes are configured and arranged to contact extremities of a patient's arms, and to sense the ECG signal therefrom.

18. An apparatus comprising: an electrode circuit attached to an outer surface of a wearable device, the electrode circuit configured and arranged to sense heart-related signals upon contact with a surface of a body of a living being and to convert the sensed heart-related signals to ECG electrical signals; and a converter circuit integrated with and electrically connected to the electrode circuit, said converter circuit configured and arranged to receive the ECG electrical signals generated by the electrode circuit and to generate a digitized ECG signal based upon the ECG electrical signals; and a computing circuit communicatively coupled to the converter circuit and configured and arranged to generate a denoised ECG signal from the digitized ECG signal, the denoised ECG signal having, relative to the digitized ECG signal, an improved signal-to-noise ratio of at least 15 dB as measured using the ANSI EC 57 standard, and a quality of signal reconstruction greater than 95%.

19. The apparatus of claim 18, wherein the electrode circuit is configured and arranged to attach to a patient's chest or arm extremities.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) The invention may be more completely understood in consideration of the following detailed description of various embodiments of the invention in connection with accompanying drawings, in which:

(2) FIG. 1 shows a subject wearing an ambulatory monitoring device employing an ECG lead with integral denoising module, consistent with an example embodiment of the present invention;

(3) FIG. 2 shows a block diagram of the sensing lead and functional components of a denoising module; consistent with an example embodiment of the present invention.

(4) FIG. 3 shows a block diagram of portions of a denoising module, consistent with an example embodiment of the present invention;

(5) FIG. 4 shows a high-level signal flow diagram of a denoising module, consistent with an example embodiment of the present invention;

(6) FIG. 5 shows a signal flow diagram of an adaptive filtering denoising algorithm as implemented in a denoising module, consistent with an example embodiment of the present invention;

(7) FIG. 6 shows a signal flow diagram of a principal component analysis denoising algorithm as implemented in a denoising module, consistent with an example embodiment of the present invention;

(8) FIG. 7 shows a signal flow diagram of the decomposition and thresholding denoising algorithm as implemented in a denoising module, consistent with an example embodiment of the present invention;

(9) FIG. 8 shows a signal flow diagram of a Multi-Domain Signal Processing (MDSP) denoising algorithm as implemented in the denoising module, consistent with an example embodiment of the present invention;

(10) FIG. 9 shows an example of a two-lead ECG signal that has been denoised using an MDSP embodiment, consistent with an example embodiment of the present invention;

(11) FIG. 10 shows a signal flow diagram for computing a signal-to-noise ratio for an ECG signal, consistent with an example embodiment of the present invention; and

(12) FIG. 11 shows an apparatus including a portable computing circuit and a chassis for sensing ECG signals, in accordance with another example embodiment.

(13) While the invention is amenable to various modifications and alternative forms, specifics thereof have been shown by way of example in the drawings and will be described in detail. It should be understood, however, that the intention is not to limit the invention to the particular embodiments described. On the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the scope of the invention including aspects defined in the claims.

DETAILED DESCRIPTION

(14) Various example embodiments of the present invention relate to an active lead for sensing an ECG at or near the surface of the body of a subject, removing in-band noise from the sensed ECG, and providing a denoised output signal similar in amplitude to the sensed ECG signal. The active ECG sensing lead with denoising capability, which is the subject invention, can be used with a large installed base of ambulatory monitoring devices already in existence and can be substituted for commonly used passive sensing leads. While the present invention is not necessarily limited to such applications, various aspects of the invention may be appreciated through a discussion of examples using this context.

(15) Certain embodiments of the present invention describe the use of Multi-Domain Filtering (MDF). The term MDF is used herein to refer to embodiments that use Multi-Domain Signal Processing (MDSP) to denoise an ECG signal. Various embodiments of the present invention are directed to denoising ECG signals sensed at the surface of the body of a human being or animal. In the context of this invention, denoising refers to removing in-band noise, or noise present within the bandwidth of the ECG signal.

(16) In the following discussion, reference is made to cited references listed in a numbered order near the end of this document, which are fully incorporated herein by reference. These references may assist in providing general information regarding a variety of fields that may relate to one or more embodiments of the present invention, and further may provide specific information regarding the application of one or more such embodiments.

(17) An example embodiment is directed to an apparatus having two or more electrodes that sense an ECG signal, a denoising module and a communication circuit. These components may, for example, be integrated in a common unit and/or with one of the electrodes being remote, such as for adhesion to a patient's skin. Aspects of the denoising module facilitate denoising that is sufficient for identifying ECG signal components, which can mitigate unnecessary communication of undesirable signals and enjoy lower processing and power overhead. The denoising module is communicatively coupled to the at least two electrodes, and receives the ECG signal sensed by the sensing electrodes. The denoising module also includes circuitry that conditions and digitizes the ECG signal, and a computing circuit that processes said digitized ECG signal to denoise the ECG signal based upon a time-based distribution of subcomponents of the signal. The communications circuit generates a communication including the denoised ECG signal for access by a device remote from the communications circuit.

(18) Another embodiment is directed to an apparatus having two or more electrodes that sense an ECG signal, digitizing circuitry that is communicatively coupled to the at least two electrodes and that generates a digitized ECG signal from the ECG signal sensed via the electrodes, and a computing circuit coupled to the digitizing circuitry. The computing circuit generates a denoised ECG signal from the digitized ECG signal, in which the denoised ECG signal has (relative to the digitized ECG signal) an improved signal-to-noise ratio of at least 15 dB as measured using the ANSI EC 57 standard, and a quality of signal reconstruction greater than 95%. Such a computing circuit may, for example, be a portable computer circuit such as a mobile telephone that receives the ECG signal sensed by the sensing electrodes. In some implementations, the quality of signal reconstruction (QSR) may is defined as

(19) $\mathrm{QSR}=100{\%}^{*}\left(1-\frac{\underset{i}{.Math.}{\left(x_{\mathrm{cl}}^i-x_{\mathrm{den}}^i\right)}^2}{\underset{i}{.Math.}{\left(x_{\mathrm{cl}}^i\right)}^2}\right),$
in which x.sub.cl is the digitized ECG signal and x.sub.den is the denoised ECG signal.

(20) In some implementations, the denoised ECG signal is generated from a digitized ECG signal that has a signal-to-noise ratio of 0 dB and that includes a substantially noise-free ECG with white noise added as prescribed in the ANSI EC 57 standard. In this context, substantially noise-free refers to a signal with less noise than desired signal, and in some instances, a signal having a signal-to-noise ratio of greater than 30 dB according to the ANSI EC 57 standard.

(21) The denoising is carried out in a variety of manners to suit particular embodiments. In some embodiments, the computing circuit decomposes the digitized ECG signal from a first domain into subcomponents in a second domain that is different than the first domain, and generates the denoised ECG signal by selecting and combining ones of the subcomponents based upon a time-based distribution thereof. In certain embodiments, the subcomponents are combined by first identifying target subcomponents of the input signal that are associated with a desired ECG signal based upon the time-based distribution of the subcomponents. A denoised ECG signal is then reconstructed in the first domain from at least two of the identified target subcomponents.

(22) In some embodiments, one or more apparatuses as discussed above are implemented for coupling to a patient, such as by adhesion of the apparatus to a patient's skin. The denoising characteristics facilitate the use of compact and lightweight circuitry that is amenable to such implementation. In a particular such embodiment, the apparatus includes an adhesive substrate that adheres to a patient, with the electrodes being coupled to the substrate and operate with the substrate to adhere to the patient and sense the ECG signal. The digitizing circuitry and computing circuit are also coupled to the substrate between the electrodes. In a more particular embodiment, the apparatus also includes a wireless communication circuit coupled to the substrate, and which wirelessly communicates the denoised ECG signal with a remote transceiver. A battery can be implemented for powering the digitizing circuitry, computing circuit and wireless communication circuit, with the substrate supporting the battery while adhered to the patient. In certain implementations, a housing connected to the substrate encloses the digitizing circuitry, computing circuit, wireless communication circuit and battery, with the substrate supporting the housing while adhered to the patient and the housing being disposed between the first and second electrodes. In some embodiments, the computing circuit and digitizing circuitry sample each ECG signal at 300 Hz or less, and the computing circuit reduces in-band noise of the digitized ECG signal using an average power consumption that is less than 500 micro-Watts per ECG signal.

(23) According to an example embodiment, and referring to FIG. 1, skin electrodes 103 are positioned on the chest of a body surface of a subject to be monitored, sense an ECG signal and are electrically connected to conductive wires via snaps 101. In this embodiment, there are three skin electrodes used to provide two ECG signals (two channels). In an alternate embodiment, two electrodes are placed on the skin of the patient to provide a single channel ECG signal. In yet other alternate embodiments, additional skin electrodes and corresponding conductive wires can be added to provide additional ECG signal channels. The conductive wires extend from each snap to a point of convergence where they consolidate into a lead body that connects to denoising module 100. Denoising module 100 outputs a low-level analog signal of an amplitude similar to the ECG signal sensed by skin electrodes 103. The output signal is conductively communicated by lead body 104 to connector 105 that plugs into ambulatory monitoring device 102 worn by the subject. In this embodiment, ambulatory monitoring device 102 is suspended from lanyard 106 worn around the neck of the monitored subject. In alternate embodiments, skin electrodes may be adhesive backed electrodes such as 3M (St. Paul, Minn.) Red Dot electrodes. Alternately, the skin electrodes may be dry electrodes in direct contact with surface tissue and incorporated into an elastic strap worn around the subject's thorax, such as from Polar Electro (Kempele, Finland). Alternately, the skin electrodes may be dry electrodes positioned or woven into the fabric of a garment worn by the subject and designed to be in contact with the subject's skin such as is available from Textronic, Inc. (Wilmington, Del.). In one embodiment, a garment has multiple ECG electrodes incorporated into the fabric, each electrically connected to the denoising module. The denoising module could be incorporated into a small pouch in the fabric with a lead body extending from the denoising module for connection to an ambulatory monitoring device. In some embodiments the denoising module can be wirelessly connected to a monitoring device such as a heart rate monitor. Alternately, the skin electrodes may be capacitive sensing electrodes such as those described in [1]. In an embodiment that employs capacitive sensing electrodes, additional signal conditioning circuitry can be used to convert the capacitance modulated signal to a voltage representing the ECG signal for input to denoising module 100.

(24) Referring to FIG. 2, skin electrodes 201 are conductively connected to denoising module 200 by snaps 202 attached to lead wires. Signal conditioning and defibrillation protection circuits 209 protect the electronics from damage in the event the patient is defibrillated as well as amplify the sensed ECG signal. In some embodiments, the circuits 209 may include an anti-aliasing filter that removes energy from the signal outside the band of the frequencies contained in the ECG signal.

(25) In one embodiment, the device shown in FIG. 2 is implemented for monitoring a human ECG, and the amplifier circuits in 209 increase the amplitude of the sensed ECG by a factor of 1,000 from 2.5 mV to 2.5 Volts peak-to-peak and will pass frequencies from 0.05 Hz to 100 Hz. In another embodiment, the device shown in FIG. 2 is used to monitor other species, and the gain and filter settings are different than those used for a human ECG.

(26) The conditioned ECG signal is digitized using an analog-to-digital converter (ADC) 216 in microprocessor chip 210. In one embodiment, chip 210 is a Texas Instruments (Dallas, Tex.) MSP-430 microcontroller. A 16-bit RISC processor executes algorithms to remove in-band noise. The denoised ECG signal is converted to an analog signal via digital-to-analog converter (DAC) 218 incorporated in chip 210. In some embodiments, ADC 216 and DAC 218 may be separate from microprocessor chip 210. The analog output signal is conditioned in circuit 211 to provide filtering and any signal level adjustments necessary to provide compatibility with ambulatory monitoring device 215. The analog output signal is conductively communicated to connector 213 via lead body 212. Connector 213 mates with connector 214 of ambulatory monitoring device 215.

(27) In one embodiment of the present invention, and referring to FIG. 2, an automatic amplitude control (AAC) function can be implemented in chip 210. The AAC attempts to maintain relatively constant QRS complex amplitude. This approach is useful, for example, when used with ambulatory monitoring devices 215 having limitations in QRS detection algorithm where changes in amplitude can be detrimental to performance.

(28) In some embodiments, ambulatory monitoring device 215 is configured to evaluate the integrity of the conductive coupling between skin electrodes 201 and the skin of the subject. This is performed by circuits within device 215 that measure impedance between skin electrodes 201. When an impedance level indicative of poor coupling is detected, device 215 may notify the subject of a problem via an alarm. It is therefore useful for this active sensing lead to detect poor quality ECG signals and modulate an impedance measured by device 215 when poor quality ECG signals are detected. In some embodiments, an algorithm implemented by microprocessor 210 monitors the quality of the signal sensed by skin electrodes 201. If skin electrodes 201 are not making adequate contact with the skin or the skin has not been properly prepared, or the electrodes have not been properly positioned, the signal quality may be poor. When the microprocessor 210 detects that signal quality is unacceptable, it sends a logic level signal to circuit 211 via digital output port 217 to modulate an impedance that is evaluated by ambulatory monitoring device 215. The impedance can be modulated, for example, by switching a high-value resistor in series with an analog signal line carrying the denoised ECG signal to ambulatory monitoring device 215. The switch used to insert the series resistor is activated by a logic-level signal provided from chip 210 via digital output port 217 when the algorithm has detected a poor quality signal.

(29) In some embodiments, denoising module 200 is powered by a primary cell battery that can be replaced via an access in the housing of module. In some embodiments, denoising module 200 is contained in a sealed housing. This may be useful in preventing ingress of moisture that may negatively impact the reliability of the denoising module. For embodiments where the housing of denoising module 200 is sealed, it is useful to use a rechargeable battery such as lithium ion battery or thin film battery. In this embodiment, battery 207 is recharged by recharging circuit 206. Circuit 206 includes inductor 205 to receive a varying magnetic field induced by inductor 204. Circuit 206 processes the alternating voltage produced by inductor 205 and converts it to a direct current suitable for charging battery 207. Recharging field generator circuit 203 provides a varying voltage to inductor 204 to produce the varying magnetic field. It is anticipated that inductor 204 will be placed near or directly adjacent to the location of inductor 205 in order to facilitate a faster charge time for battery 207. In some embodiments, it is anticipated that the charge time for battery 207 will be a few hours and that charging will be required every 1 to 4 weeks. Power regulation circuits 208 regulate the voltage to circuits 209 and 211 and to microprocessor 210. In another embodiment a supercapacitor capable of storing a large amount of charge is used to power the denoising module 200 (e.g., as an alternative to the battery 207).

(30) Referring to FIG. 3, the sensed ECG signal from electrodes 301 is conditioned by circuits 302 as described herein. The microprocessor 303 executes an algorithm to remove noise from the input ECG signal. Each input channel of the conditioned ECG signal is digitized by an analog-to-digital converter incorporated on the Texas Instruments (Dallas, Tex.) MSP-430 microcontroller chip. In some embodiments, and referring to FIG. 4, the digitized signal 401 is first processed to remove power line noise (e.g. either 60 Hz or 50 Hz, depending upon power line frequency used in individual country) using notch filter 402. In one embodiment, notch filter 402 is implemented using infinite impulse response (IIR) filtering techniques. In other embodiments, power line noise is removed by the denoising process executed by microprocessor 303 in process 403 or 405. In one embodiment, referring to FIG. 4, the possible algorithms used for denoising depends upon the number of channels in the ECG signal. If there is only one channel, denoising is directed by decision point 407 to process 405 and is accomplished using one of adaptive filtering, decomposition and thresholding, and Multi-Domain Signal Processing (MDSP). If there is more than one channel, denoising is directed by decision point 407 to process 403 and is accomplished using one of principal component analysis, adaptive filtering, decomposition and thresholding, and Multi-Domain Signal Processing (MDSP). Following denoising the signal is converted to an analog signal using a digital-to-analog converter at 404.

(31) In one embodiment, and referring to FIG. 5, the digitized ECG signal 502 is denoised using adaptive filtering techniques [4, 5]. The input ECG signal 502 is composed of a desired signal S1 that is corrupted with noise N1. Signal 502 is preprocessed to extract a noise reference signal N2 using reference generator 501. Noise reference signal N2 is correlated with noise signal N1. The adaptive filter algorithm estimates and updates the parameters of adaptive multiplier 503 to minimize the least squared error between the output Y of adaptive multiplier 503 and input signal 502. Summation node 504 computes the difference between the input signal 502 and the output of adaptive multiplier 503. The resulting difference signal e is an estimate of the desired signal S1 when the least squared error between the output Y of 503 and input signal 502 is minimized.

(32) In another embodiment reference signal N2 is acquired independently by a skin electrode, separate from the skin electrodes that produced input signal 502, rather than generating a reference signal from the input signal 502. When the reference signal N2 is a sensed ECG signal S1 is extracted by minimizing mean square error between input signal 502 and the reference signal N2. Examples of techniques that minimize mean square error include least-mean squares, normalized least mean squares, recursive least squares, adaptive recurrent filter, fast affine projection, and fast Euclidean direction [6].

(33) In the process 505, the output signal Y of adaptive multiplier 503 is used as an estimate of noise for computing noise power and the denoised signal e output from summation node 504 is used to compute signal power. The SNR is then computed in process 505 according to formula:

(34) $\begin{array}{cc}{\mathrm{SNR}}_{\mathrm{dB}}=10{\log }_{10}\left(\frac{P_{\mathrm{signal}}}{P_{\mathrm{noise}}}\right)& \left(1\right)\end{array}$

(35) where P.sub.signal and P.sub.noise are respective signal and noise energy. SNR is used to evaluate input signal quality. For example, if a skin electrode is not properly attached, the quality will be poor and would result in a low SNR value. SNR is evaluated relative to a predetermined threshold T1 in decision point 507. A logic-level 0 signal is generated in process 508 if SNR<T1 to signal to circuit 211 of FIG. 2 that a resistor should be inserted into the analog signal line in communication with ambulatory monitoring device 215. Insertion of the resistor effectively increases source impedance in the ECG signal line to ambulatory monitoring device 215, thereby emulating the situation of an increase in source impedance that would occur if a skin electrode of a passive lead were to have come loose. If the ambulatory monitoring device incorporates a mechanism based upon source impedance to signal to the subject that a lead has come loose, it will generate an alarm signal to that effect, as described herein. ECG signals are reconstructed at 506.

(36) In another embodiment, and referring to FIG. 6, the digitized ECG signal 601 is denoised using principal component analysis (PCA). In this embodiment the multi-channel ECG signal is projected onto the domain of its principal components [7]. The principal components are computed in process 602 by performing eigenvalue or singular value decomposition of the spatial covariance matrix computed from ECG signals. The projection of ECG signals onto the domain of their principal components is performed in process 603 by multiplying the ECG signals by the extracted unitary matrix of eigenvectors or singular vectors. This linear projection results in new signals that are uncorrelated with normalized variance. Geometrically, the projection involves rotation and scaling of the data in order to orthogonalize signal components. Among the orthogonalized components, the ones with low signal power are often associated with noise and can be removed to achieve denoising. The data are filtered by zeroing the principal components associated with smaller eigenvalues in process 604. In one embodiment the filtered data can be projected back onto the original domain at 605 using the unitary matrix of eigenvectors to retain original signal morphology.

(37) In one embodiment PCA is performed in combination with independent component analysis technique (ICA) [8] for denoising. In this embodiment, PCA is performed as a preliminary step to decorrelate the signals prior to applying ICA for noise removal. In various embodiments involving the use of an ICA technique for denoising the signal, noise sources are separated by achieving their mutual independence. In one embodiment the problem of maximizing independence of ECG recording from contaminating noise is found as a solution of an optimization problem that maximizes independence between the signal sources. For example, ICA techniques can use either higher-order statistics of signal components [9, 10] or information-theoretic criteria to maximize independence. Information-theoretic criteria that can be applied include maximization of negentropy or its approximation [11], minimization of mutual information [11], maximum likelihood estimation [12, 13], maximum a posteriori probability [14], or expectation-maximization of Gaussian mixture models of sources [15]. These solutions can be approximated via efficient numerical methods, such as FastICA [16] and JADE [11] algorithms.

(38) In some embodiments, it is useful to compute signal-to-noise ratio (SNR) of the input signal 601 when denoising using PCA. The parameters required in Formula 1, P.sub.signal and P.sub.noise, can be computed as a byproduct of the PCA denoising process. P.sub.signal can be estimated as the sum of the eigenvalues of the retained principal components in process 604. Likewise, P.sub.noise can be estimated as the sum of the eigenvalues of the principal components removed in process 604.

(39) In another embodiment, and referring to FIG. 7, the digitized ECG signal 701 is denoised using a signal decomposition and thresholding technique (SDTT), in accordance with another example embodiment of the present invention. In process 702 input signal 701 is decomposed into subcomponents in a second domain of larger dimension than the first domain. The dimension of the first domain is defined by the number of observed, or captured, signal channels. The dimension of the second domain is defined by the number of channels multiplied by the number of subcomponents in each channel. Decomposition step 702 is performed using one of a variety of transforms that result in a small signal reconstruction error. Such transforms may include, for example, a discrete cosine transform [17], a wavelet related transform [19], a Karhunen-Loeve transform [21], a Fourier transform [18], a Gabor transform [20], or a filter bank [19]. In one embodiment, denoising is facilitated by a decomposition whereby signal energy is concentrated in a small number of large subcomponent coefficients, while noise is spread out across either many decomposition levels or decomposition levels corresponding to higher frequency and is represented by small coefficients. In process 705, the signal quality is enhanced by discarding subcomponents below a threshold computed in process 703. The energy in the discarded subcomponents is used to estimate noise energy in process 705. Signal energy, for computing SNR in process 705 is estimated using the residual subcomponents representing the denoised ECG signal. SNR is computed from noise energy and signal energy estimates as described herein. In some implementations, instead of discarding subcomponents below a threshold, those subcomponents above a threshold (the residual subcomponents above) can be positively identified and used to estimate the signal energy, with the remaining energy being noise. Techniques similar to soft or hard thresholding [23] can be used to remove this noise in process 704. Examples of a threshold selection rule used to compute the noise threshold in process 703 include adaptive threshold selection using the principle of Stein's Unbiased Risk Estimate and fixed threshold based on signal statistics such as SD*sqrt(2*log(length(X)), where SD is standard deviation and X is a vector of subcomponents [22]. The residual subcomponents are combined in process 706 to reconstruct an ECG signal using the inverse of the transform used for signal decomposition in process 702.

(40) Referring to FIG. 8, the digitized ECG is denoised using an MDSP embodiment. Input signal 801 is decomposed in process 802 from a first domain into largely uncorrelated subcomponents in a second domain of higher dimension. Decomposition in process 802 is accomplished using one of a discrete cosine transform, a Karhunen-Loeve transform, a Fourier transform, a Gabor transform, and a filter bank. In another embodiment, decomposition is accomplished using a wavelet-related transform and the decomposition levels correspond to wavelet scales. In another embodiment, decomposition is achieved by representing the observed signals as a linear combination of basis functions. Signal decomposition embodiments and the use of subcomponents derived from the decomposition for denoising (removal of at least some of the in-band noise contained in the signal), extraction of information from the signal, and evaluation of the accuracy of extracted information is referred to as Multi-Domain Signal Processing (MDSP) by way of example, in the discussion herein. Use of MDSP techniques for removal of in-band noise from a signal is referred to as Multi-Domain Filtering (MDF).

(41) The dimension of the first domain is defined by the number of ECG channels. The dimension of the second domain is defined by the number of channels multiplied by the number of subcomponents in each channel. Following decomposition, signal flow is directed to either process 804 or 805 by decision point 803, depending upon the number of channels in the signal. If the ECG signal has only one channel, signal flow is directed to process 805, where either spatially selective filtering (SSF) [24, 28, 29] or periodic component analysis [25, 26] are used alone or in combination to identify subcomponents that contain more noise energy than signal energy. If the ECG signal has more than one channel, signal flow is directed to process 804, where either spatially selective filtering, principal component analysis [27], independent component analysis, or periodic component analysis are used either alone or in combination to identify subcomponents that contain more noise energy than signal energy. A subcomponent that contains more noise than signal energy is said to be associated with noise. Conversely, a subcomponent that contains more signal energy than noise energy is said to be associated with a signal.

(42) In some embodiments, processes 804 and 805 result in separation of subcomponents into two groups, one associated with noise and the other associated with the ECG signal. Determining whether a subcomponent is associated with noise or ECG signal within a targeted interval of time is accomplished by using one or more of principal component analysis (PCA), independent component analysis (ICA), periodic component analysis (CA) and spatially selective filtering (SSF). PCA and ICA are applicable to multi-lead ECG, while and SSF can be applied to either multi-lead or single-lead ECG.

(43) The PCA technique [7, 27] employed in processes 804 and 805 uses subcomponent covariance information to orthogonalize subcomponents. The orthogonalized subcomponents with low signal power are often associated with noise and can be removed to achieve denoising. PCA can be used as a preliminary step prior to applying an ICA technique. The ICA technique further separates signal and noise sources [8] as a solution of an optimization problem that maximizes independence between them. The CA technique computes and jointly diagonalizes covariance and autocorrelation matrices of subcomponents to separate them based on their periodicity or quasi-periodicity [25, 26]. The CA technique extracts most periodic subcomponents corresponding to ECG rhythm and, since noise is not generally periodic, it is left behind.

(44) SSF techniques [24, 28, 29] employed in processes 804 and 805 detect one or more signal-related features (e.g., QRS complex) and pass them across the subcomponents while blocking features inherent to noise. The technique relies on the differences of noise and signal distributions across decomposition levels. In one embodiment, spatially selective filtering is facilitated by a decomposition whereby signal energy is concentrated in a small number of large subcomponent coefficients while noise is spread out across many decomposition levels and is represented by small coefficients. Techniques similar to wavelet thresholding [23] can be used to remove this noise.

(45) In another embodiment of process 804 and 805, a spatially selective filtering approach exploits the fact that most noise subcomponents are confined to decomposition levels that represent high frequencies. In this embodiment the locations of signal features are identified by examining subcomponents corresponding to lower frequency. For example, a QRS wave location can be identified as high amplitude changes in peaks and valleys that occur simultaneously across multiple subcomponents associated with lower frequencies. To avoid signal distortion, the subcomponents associated with high frequency are preserved within the time window surrounding the identified peaks and valleys, and are zeroed at other times. By zeroing out the subcomponents or time segments within subcomponents associated with noise, and reconstructing the ECG signal using those subcomponents associated with the ECG signal, the in-band noise level in the ECG is reduced, or denoised, to create a denoised ECG.

(46) The subcomponents identified as associated with noise in processes 804 and 805 are removed and the residual subcomponents are used to reconstruct a denoised ECG signal for each ECG channel in process 806 using the inverse of the transform used to decompose the ECG in process 802. In some implementations, instead of removing subcomponents that are associated with noise, subcomponents associated with signal energy are identified (the residual subcomponents above) and used to estimate the signal energy, with the remaining energy being noise.

(47) In some embodiments involving an MDSP-based approach, a signal-to-noise ratio (dSNR) is computed as the ratio of the energies in signal and noise subcomponents. In one embodiment, referring to FIG. 8, following separation of subcomponents into two groups, as described herein in processes 804 and 805, the power in each of the groups, P.sub.signal and P.sub.noise, is independently computed and used to compute SNR as per formula 1.

(48) Referring to FIG. 9, an example of a two channel ECG signal denoised by an MDSP embodiment is provided. Input signals 901 and 902 are recordings of a human ambulatory ECG and were obtained from the MIT-BIH database. Output ECG traces 903 and 904 are shown for each channel. The improvement in SNR in this example is approximately 8 dB and noise amplitude is reduced by about 85% with no noticeable distortion.

(49) In some embodiments, and referring to FIG. 10, signal-to-noise ratio is computed for evaluating signal quality. As described herein, a measure of signal quality can be useful to change the effective source impedance of the ECG signal communicated to the ambulatory monitoring device when the device is configured to detect unusually high source impedance as an indicator that an ECG sensing electrode is in poor contact with the skin. In process 1001, signal and noise are separated. Some embodiments for separating signal and noise, described herein, include adaptive filtering, decomposition and thresholding, principal component analysis, and MDSP denoising embodiments. In process 1002, separated noise and signal are used to estimate signal energy and noise energy. SNR is computed in process 1003, as described herein. The approach used in process 1001 to separate signal and noise would typically match the process technique used for denoising, as the signal and noise values can be computed as a byproduct of the denoising process. Hence the technique used in 1001 will depend upon the method of denoising.

(50) Various embodiments as described herein can be used in connection with ambulatory monitoring devices. For instance, devices as discussed herein can be used as a substitute for existing passive ECG leads used with ambulatory monitoring devices, such as Holter monitors, event recorders, and Mobile Cardiac Outpatient Telemetry devices. Such devices may be used, for example, to improve performance of these devices in one or more of a variety of manners.

(51) In some embodiments, denoising module 200 includes a wireless communication link to communicate the denoised ECG to a location away from the body of the subject from which the ECG is obtained.

(52) The various computing components, circuits and signal processing methods described herein can be implemented using a variety of devices and methods. For example, computing, logic or processing circuits can be implemented using one or more of: discrete logic circuitry, fully-programmable and semi-programmable circuits such as PLAs (programmable logic arrays), specialized processors or general purpose processors that are specially programmed. Combinations of these and other circuit components are also possible and within the scope of various embodiments, including those discussed above. For example, the various components and processes shown in the figures can be implemented in a variety of circuit-based forms, such as through the use of data processing circuit modules. Such systems are exemplified by implementation in high-speed programmable computer/processor circuits, or in combination with discrete and or semi-programmable circuitry (e.g., as Field-Programmable Gate Arrays, Programmable Logic Devices/Arrays).

(53) FIG. 11 shows an apparatus 1100 including a portable computing circuit 1110 and a chassis 1120 for sensing ECG signals, in accordance with another example embodiment. The chassis 1120 includes ECG electrodes 1121 and 1122, as well as wireless communication circuitry 1123, which are shown on an inset back-side view 1105 by way of example. The electrodes 1121/1122 sense an ECG signal from a patient, and the wireless communication circuitry 1123 wirelessly communicates a wireless signal corresponding to the sensed ECG signal (e.g., via sound or radio frequency signals). The portable computing circuit 1100 (e.g., a mobile telephone) includes wireless communication circuitry 1111 that receives the wireless signal, digitizing circuitry 1112 that digitizes the wireless signal, and a computing circuit 1113 that generates a denoised ECG signal from the digitized wireless signal, as shown at inset 1104. In some embodiments, computing circuit 1110 and chassis 1120 are operable for coupling to one another, such as by snapping the computing circuit 1110 into the chassis 1120. In some implementations, the wireless communication circuits each have at least one of a speaker and a microphone that communicate the ECG signal via sound waves.

(54) Another embodiment, is directed to an ECG device having an electrode assembly that senses heart-related signals upon contact with a user's skin, and that converts the sensed heart-related signals to ECG electrical signals wherein the electrode assembly is positioned on an outer surface of a smartphone protective case. Such a device may, for example, be implemented with the apparatus shown in FIG. 1, with the following referring to FIG. 1 by way of example. A converter assembly (e.g., 1123) includes an audio transmitter for outputting frequency modulated ultrasonic signals, and is integrated with and electrically connected to the electrodes for receiving the ECG electrical signals generated thereby, and output ECG sound signals through the audio transmitter to a microphone in a computing device within range of the audio transmitter (e.g., 1110). For instance, the converter assembly may output the ECG signals as an ultrasonic FM sound signal, with the audio transmitter outputting the ultrasonic FM sound signal to a microphone in a computing device within range of the audio transmitter. This output may, for example, be detectable by a microphone in a smartphone (e.g., 1110) when the smartphone is positioned within the smartphone protective case (e.g., 1120).

(55) Such an approach, with digitizing and computing circuitry as discussed herein, can be implemented to achieve an improved signal-to-noise ratio of at least 15 dB as measured using the ANSI EC 57 standard, and a quality of signal reconstruction greater than 95%. For instance, the digitizing circuitry may be implemented to receive the ECG electrical signals via the ultrasonic FM sound signal. The converter assembly may include the digitizing circuitry and the computing circuit, and may output the ECG sound signals based upon the denoised ECG signal.

(56) For general information regarding apparatuses that may be implemented for sensing signals, and for specific information regarding such apparatuses as may be implemented above, such as via using one or more denoising approaches as discussed herein, reference may be made to U.S. Pat. No. 8,301,232 and U.S. Pat No. 8,460,189, which are fully incorporated herein by reference.

REFERENCES CITED

(57) For general information regarding a variety of fields that may relate to one or more embodiments of the present invention, and for specific information regarding the application of one or more such embodiments, reference may be made to the following documents, which are fully incorporated herein by reference.

(58) Various ones of these references are further cited above via corresponding numerals, and may be implemented as such.

(59) 1. Akinori Ueno, et al. Capacitive sensing of electrocardiographic potential through cloth from the dorsal surface of the body in a supine position: a preliminary study. IEEE Transactions on Biomedical Engineering, vol. 54, no. 4, April 2007, pp 759-766

(60) 4. N. V. Thakor and Y. Zhu, Applications of adaptive filtering to ECG analysis: noise Cancellation and arrhythmia detection, IEEE Trans. Biomedical Engineering, Vol. 38, No. 8, pp. 785-794, August 1991.

(61) 5. B. Widrow, et al., Adaptive noise cancelling: principles and applications, IEEE Proc., Vol. 63, No. 12, pp. 1692-1716, December 1975.

(62) 6. S. Hadei, M. lotfizad. A family of adaptive filter algorithms in noise cancellation for speech enhancement. International Journal of Computer and Electrical Engineering, Vol. 2, No. 2, April 2010. 1793-8163

(63) 7. L. Smith. A tutorial on Principal Components Analysis, http://users.ecs.soton.ac.uk/hbr03r/pa037042.pdf

(64) 8. P. Comon, Independent component analysis, a new concept?, Signal Process. Special Issue on Higher Order Statistics, vol. 36, no. 3, pp. 287-314, 1994.

(65) 9. J.-F. Cardoso, High-order contrasts for independent component analysis, Neural Comput., vol. 11, no. 1, pp. 157-192, 1999.

(66) 10. Blaschke, T.; Wiskott, L.; CuBICA: independent component analysis by simultaneous third- and fourth-order cumulant diagonalization IEEE Transactions on Signal Processing, Volume 52, Issue 5, May 2004 Page(s): 1250-1256

(67) 11. Hyvrinen, A. New approximations of differential entropy for independent component analysis and projection pursuit. In Advances in Neural Information Processing Systems, volume 10, pages 273-279.MIT Press.

(68) 12. Bell, A. and Sejnowski, T. An information-maximization approach to blind separation and blind deconvolution, Neural Computation, 7:1129-1159.

(69) 13. Cardoso, J.-F. Infomax and maximum likelihood for source separation. IEEE Letters on Signal Processing, 1997, 4, 112-114.

(70) 14. M. Zibulevsky, B. Pearlmutter, Blind source separation by sparse decomposition in a signal dictionary, Neural Computation 2001, v 13, pp 863-882.

(71) 15. Todros, K.; Tabrikian, J.; Blind Separation of Independent Sources Using Gaussian Mixture Model IEEE Transactions on Signal Processing Volume 55, Issue 7, Part 2, July 2007 Page(s): 3645-3658

(72) 16. Hyvrinen, A. Fast and robust fixed-point algorithms for independent component analysis. IEEE Transactions on Neural Networks, 10(3):626-634.

(73) 17. K. R. Rao and P. Yip, Discrete Cosine Transform: Algorithms, Advantages, Applications San Diego, Calif.: Academic, 1990.

(74) 18. Mallat, S. G., and Zhang, Z., Matching Pursuits with Time-Frequency Dictionaries, IEEE TSP(41), No. 12, December 1993, pp. 3397-3415.

(75) 19. Vaidyanathan. Multi-rate Systems and Filter Banks, Prentice Hall, 1993

(76) 20. J. Woods. Subband Coding, Kluwer Academic Press, 1990.

(77) 21. K. S. Ball, L. Sirovich, L. R. Keefe. Dynamical eigenfunction decomposition of turbulent channel flow. International Journal for Numerical Methods in Fluids Volume 12, Issue 6, Date: 5 Apr. 1991, Pages: 585-604

(78) 22. Donoho, D. L., I. M. Johnstone (1994), Ideal spatial adaptation by wavelet shrinkage, Biometrika, vol 81, pp. 425-455.

(79) 23. Donoho, D. L. (1995), De-noising by soft-thresholding, IEEE Trans. on Inf. Theory, 41, 3, pp. 613-627.

(80) 24. Xu, Yansun, et al. Wavelet transform domain filters: a spatially selective noise filtration technique, IEEE transactions on image processing 1994, vol. 3, no 6, pp. 747-758

(81) 25. L. K. Saul and J. B. Allen, Periodic component analysis: An eigenvalue method for representing periodic structure in speech. in NIPS, [Online]. 2000, pp. 807-813. Available: http://www.cs.cmu.edu/Groups/NIPS/00papers-pub-on-web/SaulAllen.pdf

(82) 26. R Sameni, et. al. Multichannel electrocardiogram decomposition using periodic component analysis. IEEE Transactions on Biomedical Engineering, 2008 vol 55, no 8 pp 1935-1940

(83) 27. Aminghafari, M.; Cheze, N.; Poggi, J-M. (2006), Multivariate de-noising using wavelets and principal component analysis, Computational Statistics & Data Analysis, 50, pp. 2381-2398

(84) 28. Mallat, S. G., Hwang, W. L., Singularity Detection and Processing with Wavelets, IEEE Transactions on Information Technology (38), 1991, pp. 617-643.

(85) 29. Pan Q., Zhang L. Dai G., et al. Two denoising methods by wavelet transform. IEEE Trans. on SP., 1999, 47(12): 3401-3406

(86) Based upon the above discussion and illustrations, those skilled in the art will readily recognize that various modifications and changes may be made to the present invention without strictly following the exemplary embodiments and applications illustrated and described herein. For example, various aspects are directed to signal processing such as denoising using one or more approaches as described in one or more of the priority documents referenced above and incorporated herein. One such aspect involves the denoising of an ECG signal using one or more approaches described in connection with multi-domain signal processing as described in these documents. Such modifications and changes may include, for example, incorporating one or more aspects described in the above references and/or applying one or more embodiments thereto, or combining embodiments. These and other modifications do not depart from the true spirit and scope of the present invention, including that set forth in the following claims.