FIELD OF INVENTION
[001] The present invention relates to signal processing system and more particularly to a method and system for peak detection in an electrocardiogram (ECG) signal processing and analysis system.

BACKGROUND OF INVENTION
[002] An electrocardiogram (ECG) waveform or signal generally includes a series of characteristic points conventionally designated by the letters P  Q  R  S  and T. The Q  R  and S portions of the wave when taken together are referred to as “QRS complex”. The R-wave of the QRS complex is the most prominent wave in each cardiac cycle of the ECG signal. Therefore  an efficient automatic detection of time instants of R-peaks is important in various ECG signal processing applications  such as Heart Rate Variability (HRV) analysis  computer-aided cardiac diagnostic system  Fetal Heart Rate (FHR) monitor  heart sound detection  ECG-based biometric system  ECG compression system  cardiac event change detector  wireless medical body area network  remote cardiac patient monitoring system  and other ECG signal processing applications.
[003] Different digital processing systems and methods implement techniques for detecting R-peaks or QRS complexes. Such techniques can include  for example  Digital Filter (DF)  filter-banks  Geometrical Matching (GM)  genetic algorithms  Hilbert Transform (HT)  Higher-Order Statistics (HOS)  Hidden Markov Model (HMM)  Linear Prediction (LP)  Maximum-a-Posteriori (MAP) estimation  matched filters  mathematical morphology  multi-scale mathematical morphology  3M and Empirical Mode Decomposition (EMD)  syntactical rules  neural networks  Support Vector Machine (SVM)  Template Matching (TM)  Two-pole recursive filter  Wavelet Transform (WT)  Zero-Crossing (ZC)  and various other techniques.
[004] Although the techniques described above are effective in detecting R-peaks in the ECG signal  but includes both the advantages and disadvantages in terms of the ECG signals with wide QRS complexes  low-amplitude QRS complexes  negative QRS polarities  sudden changes in RR intervals  sudden changes in QRS amplitudes  sudden changes in QRS morphologies  sharp P/T waves  and various kinds of noise (or artifacts) including baseline wander  power-line interference  muscle artifacts  electrosurgical noise  and motion artifacts. Thus  there remains a need of a robust system and method for automatically determining time instants of R-peaks in a received or recorded signal.
OBJECT OF INVENTION
[005] The principal object of the embodiments herein is to provide a system and method for automatically determining time instants of peaks in a received or recorded signal.
[006] Another object of the invention is to provide a method and system for enhancing QRS complexes (or large high slope regions) and suppressing small-amplitude high-frequency noises and artifacts of a received or recorded signal.
[007] Another object of the invention is to provide a mechanism for automatically detecting peaks in a received or recorded signal.
[008] Another object of the invention is to provide a method and system for improving detection accuracy of time instants of peaks in a received or recorded signal.
[009] Another aspect of the invention is to provide a method for constructing smooth envelope of processed signal which comprises large local maxima corresponding to desired QRS complexes in a received or recorded signal.

SUMMARY
[0010] Accordingly the invention provides a method for automatically determining time instants of peaks in a signal. The method includes determining a zero-mean data sequence of the signal and filtering the zero-mean data sequence of the signal. Further  the method includes determining envelope of the filtered data sequence of the signal and determining the time instants of the peaks in the entropy data sequence of the signal.
[0011] In an embodiment  the determination of the zero-mean data sequence further includes dividing the signal into processing blocks and performing a mean subtraction of the signal to determine the zero-mean data sequence.
[0012] Furthermore  the method includes generating an over-complete transform matrix including a set of elementary functions. The set of elementary functions include column vectors from an identity matrix and a cosine or sine matrix. Furthermore  the method includes estimating a transformation coefficient for the set of elementary function of the over-complete transform matrix and filtering the zero-mean data sequence of the signal using the estimated transform coefficient.
[0013] Furthermore  the method includes performing an amplitude normalization of the filtered data sequence of the signal and performing an absolute operation of the normalized filtered data sequence to transform bipolar filtered data sequence into unipolar filtered data of the signal. Furthermore  the method includes performing adaptive thresholding on the transformed data sequence of the signal and determining the entropy of the thresholded data sequence of the signal. In an embodiment  the determined entropy is Shannon entropy.
[0014] Furthermore  the method includes smoothing the determined entropy data sequence and convolving the smoothed entropy data sequence of the signal. The convolved data sequence of the signal includes positive zero-crossing point and negative zero-crossing point. The negative zero-crossing points indicate locations of the peaks in the entropy data sequence of the signal.
[0015] Furthermore  the method includes detecting the locations of the negative zero-crossing points in the convolved data sequence and using the detected location of the negative zero-crossing points to automatically determine the time instants of the peaks of the signal.
[0016] Accordingly the invention provides a system for automatically determining time instants of peaks in a signal. The system includes a blocking and mean subtraction module configured to determine a zero-mean data sequence of the signal and a sparsity filtering module configured to filter the zero-mean data sequence of the signal. Further  the system includes an envelope module configured to compute envelope of the filtered data sequence of the signal and an output detector module configured to automatically determine the time instants of the peaks in the entropy data sequence of the signal.
[0017] In an embodiment  the system processes the signal using multi-electrode leads. Further  the blocking and mean subtraction module is further configured to divide the signal into processing blocks and perform a mean subtraction of the signal to determine the zero-mean data sequence.
[0018] In an embodiment  the sparsity filtering module further includes a dictionary matrix generation module configured to construct an over-complete transform matrix including a set of elementary functions. The set of elementary functions include column vectors from an identity matrix and a cosine or sine matrix. Furthermore  the sparsity filtering module includes a sparse coefficient estimation module configured to estimate a transformation coefficient for the set of elementary function of the over-complete transform matrix. Furthermore  the sparsity filtering module is configured to filter the zero-mean data sequence of the signal using the estimated transform coefficient.
[0019] In an embodiment  the envelope module is further configured to perform an amplitude normalization of the filtered data sequence of the signal and perform an absolute operation of the normalized filtered data sequence to transform bipolar filtered data sequence into unipolar filtered data of the signal. Furthermore  the envelope module is further configured to perform adaptive thresholding on the transformed data sequence of the signal and compute the entropy of the thresholded data sequence of the signal. In an embodiment  the envelope module computes Shannon entropy of the thresholded data sequence of the signal.
[0020] Furthermore  the system is configured to include a smoothing filter configured to remove noise from the computed entropy data sequence using a rectangular impulse response of length L. Furthermore  the system is configured to include a Gaussian filtering module configured to convolve the smoothed entropy data sequence of the signal. The convolved data sequence of the signal includes positive zero-crossing point and negative zero-crossing point. In an embodiment  the negative zero-crossing points indicate locations of the peaks in the entropy data sequence of the signal.
[0021] Furthermore  the system is configured to include a zero-crossing detector module configured to detect the locations of the negative zero-crossing points in the convolved data sequence and an output detector module is configured to use the detected location of the negative zero-crossing points to automatically determine the time instants of the peaks of the signal.
[0022] These and other aspects of the embodiments herein will be better appreciated and understood when considered in conjunction with the following description and the accompanying drawings. It should be understood  however  that the following descriptions  while indicating preferred embodiments and numerous specific details thereof  are given by way of illustration and not of limitation. Many changes and modifications may be made within the scope of the embodiments herein without departing from the spirit thereof  and the embodiments herein include all such modifications.

BRIEF DESCRIPTION OF FIGURES
[0023] This invention is illustrated in the accompanying drawings  throughout which like reference letters indicate corresponding parts in the various figures. The embodiments herein will be better understood from the following description with reference to the drawings  in which:
[0024] FIG. 1 depicts a block diagram of applications of a R-peak detection system  according to embodiments as disclosed herein;
[0025] FIG. 2 depicts a detailed view of the R-peak detection system of the FIG. 1  according to embodiments as disclosed herein;
[0026] FIG. 3 depicts a detailed view of Sparsity filtering module of the FIG. 2  according to embodiments as disclosed herein;
[0027] FIG. 4 depict graphs representing an example of experimental waveforms obtained by the sparsity filtering module of the FIG. 3  according to embodiments as disclosed herein;
[0028] FIG. 5 depict graphs representing another example of experimental waveforms obtained by the sparsity filtering module of the FIG. 3  according to embodiments as disclosed herein;
[0029] FIG. 6 depicts a detailed view of envelope module of the FIG. 2  according to embodiments as disclosed herein;
[0030] FIG. 7 depicts a flow diagram illustrating operations performed by the R-peak detection system of the FIG. 2  according to embodiments as disclosed herein;
[0031] FIG. 8 depict graphs representing an example of experimental waveforms obtained by the R-peak detection system of the FIG. 2  according to embodiments as disclosed herein;
[0032] FIG. 9 depict graphs representing an example of experimental waveforms showing performance detection for ECG signal with large P-waves and muscle noise  according to embodiments as disclosed herein;
[0033] FIG. 10 depict graphs representing an example of experimental waveforms showing performance detection for ECG signal with noise and numerous long pauses  according to embodiments as disclosed herein;
[0034] FIG. 11 depict graphs representing an example of experimental waveforms showing performance detection for ECG signal with wide QRS complexes  according to embodiments as disclosed herein; and
[0035] FIG. 12 depicts a computing environment implementing the application  in accordance with various embodiments of the present invention.

DETAILED DESCRIPTION OF INVENTION
[0036] The embodiments herein and the various features and advantageous details thereof are explained more fully with reference to the non-limiting embodiments that are illustrated in the accompanying drawings and detailed in the following description. Descriptions of well-known components and processing techniques are omitted so as to not unnecessarily obscure the embodiments herein. The examples used herein are intended merely to facilitate an understanding of ways in which the embodiments herein can be practiced and to further enable those of skill in the art to practice the embodiments herein. Accordingly  the examples should not be construed as limiting the scope of the embodiments herein.
[0037] The embodiments herein achieve a system and method for automatically determining time instants of R-peaks in a received or recorded signal. The system and method includes a sparsity filtering module to filter the signal and suppress baseline wander  power-line interference  muscle artifacts  motion artifacts  sharp P/T waves  and electrosurgical noises of the signal. An envelope module performs an amplitude normalization of the filtered data sequence of the signal. An absolute operation is performed on the normalized data sequence to transform bipolar filtered data sequence into unipolar filtered data sequence of the signal. The envelope module performs an adaptive thresholding of the transformed data sequence and computes Shannon entropy of the thresholded data sequence of the signal.
[0038] Further  the system and method includes a smoothing filter to remove noise from the computed Shannon entropy data sequence of the signal. A Gaussian filtering module convolves the smoothed Shannon entropy data sequence of the signal with a Gaussian derivative function. The Gaussian filtering module provides a convolved Shannon entropy data sequence includes negative zero-crossing points indicating locations of R-peaks in the signal. A zero-crossing detector module detects the locations of the negative zero-crossing points in the convolved data sequence of the signal. An output detector module uses the detected locations to automatically determine the time instants of the R-peaks in the signal.
[0039] The method and system disclosed herein is simple  robust  reliable  inexpensive  and accurate in detecting the R-peaks in a received or recorded signal under different noisy conditions. Unlike the other existing detection methods and systems  the present invention provides a one-pass detection method  without the use of search-back mechanisms  implemented with different amplitude-dependent and RR-interval-dependent thresholds. The present invention uses a single adaptive threshold rule to improve the accuracy of detecting the time instants of the R-peaks in the signal under different noisy conditions. The proposed system and method uses an automated peak-finding logic to accurately detect low-amplitude QRS complexes and wide-QRS complexes.
[0040] Referring now to the drawings  and more particularly to FIGS. 1 through 12  where similar reference characters denote corresponding features consistently throughout the figures  there are shown preferred embodiments.
[0041] Throughout the description  the term sparsity filtering and L1-Sparsity filtering is used interchangeably.
[0042] Throughout the description  the term dictionary matrix and transform matrix (or representation matrix or sparse matrix) is used interchangeably. The dictionary matrix can be overcomplete or undercomplete or critical depending on applications.
[0043] Throughout the description  the term elementary functions and elementary waveform (or basis functions or elementary atoms or elementary waveforms) is used interchangeably.
[0044] Throughout the description  the term transform coefficient and coefficient vector is used interchangeably.
[0045] FIG. 1 depicts a block diagram 100 of applications of an R-peak detection system 102  according to embodiments as disclosed herein. Information about the physiological conditions of a patient can be monitored by positioning electrodes on the patient body in specific locations. In an example  different channels can be used to monitor electrical activity from different horizontal and frontal planes. In an embodiment  the R-peak detection system 102 can processes multi-channels (or multi-electrode leads) output received from the electrodes. The output of the electrodes can include  for example  cardiac related electrical signals such as electrocardiogram (ECG) waveform signals  pacemaker pulse signals acquired by the electrodes  or any other physiological parameter of the patient. The R-peak detection system 102 is configured to be coupled to a data acquisition module 104 to acquire the physiological signal output from the electrodes.
[0046] In an embodiment  the data acquisition module 104 is configured to receive cardiac bio-potentials of the patient. The R-peak detection system 102 is configured to receive the physiological signal using the data acquisition module 104. The R-peak detection system 102 is configured to automatically determine time instants of R-peaks of the signal. The determined time instants of the R-peaks can be used by different signal processing applications for further processing or analysis purposes. The signal processing applications described herein can include  for example  Heart-rate variability (HRV) analyzer 106  ECG signal compressor 108  ECG-wave delineator 110  Fetal heart-rate analyzer 112  Heart sound analyzer 114  ECG-biometric authenticator 116  Cardiac event change detector 118  and ECG arrhythmias recognizer 120. In an embodiment  the determined time instants of the R-peaks can be used by any cardiac signal processing applications.
[0047] In an embodiment  the R-peak detection system 102 can provide a connecting and controlling mechanism for communicating electronic data with the signal processing applications. The signal processing applications can be wired or wirelessly connected to the R-peak detection system 102 using the techniques such as cellular networks  Radio-frequency identification (RFID)  ZigBee  Bluetooth  Wi-Fi  Ultra-wideband (UWB)  Worldwide Interoperability for Microwave Access (WiMax)  wireless USB  wireless local area network  near filed communications  or any other techniques. In an embodiment  the signal processing applications can be embedded in the R-peak detection system 102 to use the determined time instants for further processing or analysis purposes.
[0048] FIG. 2 depicts a detailed view of the R-peak detection system 102 of the FIG. 1  according to embodiments as disclosed herein. The R-peak detection system 102 is configured to include blocking and mean subtraction module 202  sparsity filtering module 204  envelope module 206  Gaussian filtering module 208  Zero-crossing detector module 210  and output detector module 212.
[0049] In an embodiment  the blocking and mean subtraction module 202 is configured to receive an input signal from the data acquisition module 104. In an example  the input signal described herein can be an ECG signal including cardiac biopotentials. The blocking and mean subtraction module 202 is configured to divide the input signal into non-overlapping blocks of length N with certain time duration (for example  10 seconds). The blocking of the input signal is performed for effective suppression of different shapes of the baseline wander. Further  the blocking and mean subtraction module 202 is configured to perform a mean subtraction of the input signal to provide a zero-mean discrete-time data sequence of the input signal. The mean subtraction of the input signal is performed for better estimation of a transform coefficient of the signal.
[0050] In an example  the sparsity filtering module 204 is configured  to be coupled to the blocking and mean subtraction module 202  to receive the zero-mean discrete-time data sequence of the input signal. The sparsity filtering module 204 is configured to filter the zero-mean discrete-time data sequence to enhance QRS complex portions of the input signal. The sparsity filtering module 204 is configured to suppress the baseline wander  power-line interference  muscle artifacts  motion artifacts  sharp P/T waves  electrosurgical noises  and any other noise (or artifact) of the input signal. Further  the sparsity filtering module 204 is configured to implement a L1-sparsity filtering method to filter the input signal based on an over-complete set of elementary functions or waveforms. More detailed information about the L1-sparsity filtering method is described in conjunction with FIG.3.
[0051] In an embodiment  the envelope module 206 is configured  to be coupled to the sparsity filtering module 204  to receive the filtered data sequence of the input signal. The envelope module 206 is configured to compute entropy of the filtered data sequence of the input signal. In an embodiment  the entropy described herein is Shannon entropy. In an embodiment  the entropy can be Kolmogorov entropy or any other entropy. The R-peak detection system 102 is configured to compute Shannon entropy to improve accuracy of detecting the time instants of the R-peaks in the input signal with low-amplitude and wide QRS complexes. Further  the envelope module 206 is configured is configured to implement a Shannon entropy method to compute the Shannon entropy of the input signal. More detailed information about the Shannon entropy method is described in conjunction with FIG. 4.
[0052] In an embodiment  the Gaussian filtering module 208 is configured  to be coupled to the envelope module 206  to receive the Shannon entropy data sequence of the input signal. The Gaussian filtering module 208 is configured to implement a peak-finding logic or scheme to identify locations of peaks in the input signal. In an embodiment  the Gaussian filtering module 208 is configured to convolve the Shannon entropy data sequence with a Gaussian derivative function to identify the R-peaks locations of the input signal. The convolved data sequence of the input signal includes positive zero-crossing points and negative zero-crossing points. In an example  the negative zero-crossing points described herein indicate locations of the R-peaks in the Shannon entropy data sequence of the input signal.
[0053] In an embodiment  the zero-crossing detector module 210 is configured  to be coupled to the Gaussian filtering module 208  to detect the locations of the negative zero-crossing points in the Shannon entropy data sequence. In an example  these locations can be used to determine the time instants of the R-peaks in the data sequence of the input signal.
[0054] In an embodiment  the output detector module 212 is configured  to be coupled to the zero-crossing detector module 210. The output detector module 212 is configured to use the detected locations of the negative zero-crossing points (indicating the locations of the R-peaks) to automatically determine the time instants of the R-peaks in the input signal. Further  the output detector module 212 is configured to output or provide the determined time instants of the R-peaks to different signal processing applications (may be for further processing or analysis purposes). Further  the output detector module 212 is configured to combine the detection R-peaks in order to reduce false positive and false negative of the input signal.
[0055] FIG. 3 depicts a detailed view of the sparsity filtering module 204 of the FIG. 2  according to embodiments as disclosed herein. The sparsity filtering module 204 enhances the QRX complex portions and suppresses any type of specific or complete noise (or artifact) from a recorded or received signal. For this purpose  the sparsity filtering module 204 is configured to initiate design parameters. The design parameters described herein can include  an input signal (x[n])  a block length (N)  a regularization parameter (?)  an over-complete transform matrix ( )  a Gaussian window length (P)  and a rectangular window length (L). The blocking and mean subtraction module 202 divides the input signal x[n] into non-overlapping processing blocks of length N for effective suppression of different shapes of the baseline wander. The blocking and mean subtraction module 202 performs a mean subtraction of the input signal x[n] to provide a zero-mean discrete-time signal for better estimation of the transform coefficient (a) of the input signal x[n]. The sparsity filtering module 204 specifies a value to the regularization parameter to control fidelity and sparse constraint of the input signal x[n]. The Gaussian window is used to provide locations of peaks in the input signal x[n].
[0056] The sparsity filtering module 204 is configured to implement a L1-sparsity filtering method to filter the input signal x[n]. The method enhances the QRX complex portions and suppresses the baseline wander  power-line interference  muscle artifacts  motion artifacts  sharp P/T waves  and electrosurgical noises of the input signal x[n]. The sparsity filtering module 204 is configured  to include a dictionary matrix generation module 302  to generate or construct the over-complete transform matrix ( ) for the input signal x[n]. In an embodiment  the dictionary matrix generation module 302 can generate under-complete or critical transform matrix for the input signal x[n]. The over-complete transform matrix ( ) described herein includes a set of elementary functions from column vectors of an identity matrix (I) and a cosine matrix (C). In an embodiment  the column vectors of the identity matrix (I) can be used to extract high-frequency components of the input signal x[n] and the column vectors of the cosine matrix (c) can be used to extract-low frequency components of the input signal x[n].
[0057] The dictionary matrix generation module 302 constructs the over-complete transform matrix ? RNxM with size of N X M (where  N < M that contains M prototype waveforms for columns of ?). The over-complete transform matrix ( ) includes a set of elementary functions from two matrixes  which is computed as  where I is the N ×N identity matrix and C is the N ×K cosine matrix. In an example  the input signal x ? RNxN can be represented as a linear combination of the prototype waveforms (as the column vectors = { ?1| ?2| ?3|..........| ?M}):

Where  is the transform coefficients vector that is computed as .
[0058] In an embodiment  for the input signal x[n] and the over-complete transform matrix ?  the transform coefficient (a) needs to be computed. The sparsity filtering module 204 is configured  to include a sparse coefficient estimation module 304  to compute the transform coefficient (a) by using a L1-norm minimization algorithm. For the input signal x[n] and the over-complete transform matrix ?  the transform coefficient (a) can be computed by solving the following L1-norm minimization problem:

[0059] Where  is fidelity term  is a sparsity term  x is the input signal to be decomposed  and ? is the regularization parameter that controls the relative importance of the fidelity and sparseness terms.
[0060] In an example  for a pre-defined over-complete elementary function set ? = [{i1| i2| i3|..... iN| cN+1| cN+2| cN+3|.... cN+K]  the estimated transform coefficient (a) is given by   where denotes the coefficients vector for the elementary functions from the column vectors of the identity matrix (I) and denotes the discrete cosine transform (DCT) coefficients vector for the elementary functions from the column vectors of the cosine matrix (C). In an embodiment  the spike-like waveforms in the column vectors of the identity matrix I ? RN×N can be used as a basis to extract QRS complex portions or high-frequency components of the input signal x[n]. The column vectors of the cosine matrix C ? RN×K can be used to extract the slowly-varying components or low-frequency components of the input signal x[n]. Further  the sparsity filtering module 204 is configured to filter the input signal x[n] by extracting the QRS complex portions from the transformation coefficient (a).
[0061] In an embodiment  the filtered signal d[n] can be computed as   where column vector ii ? RN×1 from the identity matrix (I) includes only one non-zero entry. The output of the sparsity filtering module 204 is the filtered signal d[n] of the input signal x[n]  which is further processed by the R-peak detection system 102 to locate the QRS complexes.
[0062] The R-peak detection system 102 can configure sparsity filtering module 204 by selecting an appropriate filtering technique including derivative filter  band-pass filter  wavelet decomposition  empirical mode decomposition  L1-sparsity filter  or any other filtering technique. In an embodiment  the R-peak detection system 102 is configured to select the appropriate filtering technique for enhancing QRS complex portions and reducing any type of noise and artifacts associated with the input signal. Depending upon the exemplary application(s) and computing resources requirements  the R-peak detection system 102 can implement the appropriate filtering technique for the QRS detection.
[0063] FIG. 4 depict graphs 400 representing an example of experimental waveforms obtained by the sparsity filtering module 204 of the FIG. 3  according to embodiments as disclosed herein. The performance of the proposed L1-sparsity filtering method as described in the FIG. 3 is evaluated using an exemplary noisy or corrupted ECG signal 402. The estimated coefficients vector for the ECG signal 402 is shown in graph 404. The filtered signal d[n] is shown in graph 406. The filtered signal d[n] shows that the L1-sparsity filtering method emphasizes the QRS complex portions and simultaneously suppresses the baseline wander  power-line interference  muscle artifacts  motion artifacts  sharp P/T waves  and electrosurgical noises of the input signal x[n]. The low-frequency component constructed from the 77 × 1 discrete cosine transformed coefficients vector obtained for the discrete cosine functions is shown in graph 408. Thus  the experimental result shows that the spike-like waveforms from columns of the identity matrix (I) can capture the QRS complex portions of the ECG signal 402.
[0064] FIG. 5 depict graphs 500 representing another example of experimental waveforms obtained by the sparsity filtering module 204 of the FIG. 3  according to embodiments as disclosed herein. The performance of the proposed L1-sparsity filtering method as described in the FIG. 3 is evaluated using an exemplary noisy or corrupted ECG signal 502. The estimated coefficients vector for the ECG signal 502 is shown in graph 504. The filtered signal d[n] is shown in graph 506. The filtered signal d[n] shows that the L1-sparsity filtering method emphasizes the QRS complex portions and simultaneously suppresses the baseline wander  power-line interference  muscle artifacts  motion artifacts  sharp P/T waves  and electrosurgical noises in the input signal x[n]. The low-frequency component constructed from the 77 × 1 discrete cosine transformed coefficients vector obtained for the discrete cosine functions of is shown in graph 508. Thus  the experimental result shows that the spike-like waveforms from columns of the identity matrix (I) can capture the QRS complex portions of the ECG signal 502.
[0065] FIG. 6 depicts a detailed view of the envelope module 206 of the FIG. 2  according to embodiments as disclosed herein. The envelope module 206 is configured to normalize the filtered data sequence d[n] of the input signal x[n] in terms of peak amplitude. In an example  the normalization is achieved by multiplying the ECG waveform by normalization  or scaling  function. The amplitude normalization is performed using the following function:

[0066] The envelope module 206 is configured to perform an absolute operation on the normalized data sequence of the input signal x[n]. In an example  the envelope module 206 is configured to perform an absolute operation using a non-linear transformation technique. The non-linear transformation is performed to convert a bipolar data sequence into a unipolar data sequence of the input signal x[n]. In an example  the non-linear transformation  including the absolute operation  is performed to convert the bipolar filtered ECG sequence into a positive-valued ECG sequence that eliminates detection problems in case of negative QRS complexes. The absolute value of the filtered data sequence d[n] is computed as:

[0067] Further  the present invention converts or transforms the normalized data sequence of the input signal using a linear transformation technique or the non-linear transformation technique. The linear or non-linear transformation technique described herein can include the absolute operation  squaring operation  Shannon entropy operation  Shannon energy operation  or any other technique to obtain a positive-valued signal from the filtered signal that eliminates detection problems in the case of negative QRS complexes. The R-peak detection system 102 is configured to select an appropriate linear or non-linear transformation operation based on peak-finding method  characteristics of noise components  computing resources  and level of detection accuracy required for targeted exemplary application(s).
[0068] In an embodiment  the R-peak detection system 102 can implement or compute the amplitude normalization and the absolute operation as a single logic or scheme to reduce the computational complexity of the system 102.
[0069] The envelope module 206 is configured to perform adaptive thresholding on the transform data sequence of the input signal x[n]. In an example  the R-peak detection system 102 can determine the applicability of thresholding function based on the filtering technique used. In an example  the thresholding function is defined as:

[0070] Where the absolute values a[n] is smaller than a threshold parameter ? and the threshold parameter are set to zero. In an embodiment  the adaptive-threshold parameter ? is computed for each input signal processing block. The threshold parameter ? is determined based on the standard deviation of the absolute values of the normalized filtered data sequence of the input signal x[n]. The threshold process effectively eliminates spurious noise spikes and reduces the number of false positive detections under noisy ECG signals and ECG signal with long pauses.
[0071] The envelope module 206 is configured to compute the Shannon of the thresholded data sequence of the input signal x[n]. The envelope module 206 implements Shannon entropy based method for producing small deviations for the successive local maxima. The Shannon entropy of the sequence is computed as .
[0072] Further  the R-peak detection system 102 is configured to remove noise from the computed Shannon entropy data sequence s[n] of the input signal x[n]. In an example  the thresholded absolute values are smoothed using the zero-phase filtering with a rectangular impulse response (or rectangular window) of length L. Generally  the L is approximately the same as duration of possible wider QRS complex. In an example  the average of lower and upper duration limits of QRS complex portions is considered based on the length L of the rectangular window. The smoothing process reduces the effect of multiple peaks around QRS complex regions and provides sharp large local maxima around QRS complex portions. The smoothing process provides smoothed energy envelopes with isolated peaks corresponding to the QRS-complex portions in the input signal x[n]. The locations of the candidate R-peaks in the smoothed Shannon entropy data sequence s[n] correspond to approximate locations of the R-peaks in the input signal x[n].
[0073] The locations of peaks in the smoothed entropy data sequence s[n] are identified using the Gaussian filtering module 208 and the Zero-crossing detector module 210. The Gaussian filtering module 208 is configured to implement a peak-finding logic or scheme to identify locations of peaks in the input signal. The Gaussian filtering module 208 provides an output sequence that is the convolution of the smoothed Shannon entropy s[n] with a Gaussian derivative kernel function g[n]. In an embodiment  the output of the Gaussian filtering module 208 includes positive and negative zero-crossing points of the input signal x[n]. In an embodiment  the negative zero-crossing points indicate locations of the peaks in the Shannon entropy data sequence s[n].
[0074] In an example  the P-point Gaussian window g[p] is computed as:

[0075] The first order Gaussian derivative sequence is computed as   which gives the slope at each sample.
[0076] The zero-crossing detector module 210 is configured to detect locations of negative zero-crossing points in the s[n] using the peak-finding logic. The output detector module 212 is configured to automatically determine the locations of true R-peaks in input signal x[n]. The output detector module 212 uses the detected locations of the negative zero-crossings (indicating the location of R-peaks) to automatically determine the time instants of the R-peaks in the input signal x[n]. In an example  the output detector module 212 combines the detection R-peaks to reduce the false positive and false negative detections of the input signal x[n]. Further  the output detector module 212 is configured to output the determined time instants of the R-peaks in the input signal x[n].
[0077] Further  present invention implements different implementing peak-finding logic or method and to detecting locations of negative zero-crossing points in the Shannon entropy data sequence. The peak-finding logic or method described herein can include Hilbert-Transform (HT) and Moving Average (MA) ?lter  the first-order Gaussian differentiator  peak-amplitude thresholding and peak-searching window  or other peak-finding logic. The R-peak detection system 102 is configured to select an appropriate peak-finding method based on the computing resources targeted exemplary application(s).
[0078] FIG. 7 depicts a flow diagram 700 illustrating operations performed by the R-peak detection system 102 of the FIG. 2  according to embodiments as disclosed herein. At step 702  the R-peak detection system 102  in communication with the data acquisition module 104  receives an input data sequence x[n]. In an example  the input data sequence x[n] described herein can be an ECG signal including cardiac biopotentials  or any other type of signal. In an example  the R-peak detection system 102 initializes the design parameters  for example  the input signal x[n]  a block length (N)  a regularization parameter (?)  an over-complete transform matrix ( )  a Gaussian window length (P)  and a rectangular window length (L).
[0079] At step 704  the blocking and mean subtraction module 202 performs blocking and means subtraction on the input data sequence x[n]. In an example  the blocking and mean subtraction module 202 divides the input data sequence x[n] into non-overlapping processing blocks of length N for effective suppression of different shapes of the baseline wander. The blocking and mean subtraction module 202 performs a mean subtraction of the input data sequence x[n] to provide a zero-mean discrete-time data sequence for better estimation of the transform coefficient (a) of the input signal x[n].
[0080] At step 706  the sparsity filtering module 204 filters the zero-mean data sequence of the input signal x[n]. In an example  the sparsity filtering module 204 implements a L1-sparsity filtering method to enhance the QRX complex portions and suppress the baseline wander  power-line interference  muscle artifacts  motion artifacts  sharp P/T waves  and electrosurgical noises of the input data sequence x[n]. In an example  the sparsity filtering module 204 sparsity filtering module 204  in communication with the dictionary matrix generation module 302  generates the over-complete transform matrix ? RNxM with size of N X M (where  N < M that contains M prototype waveforms for columns of ?) to filter the input data sequence x[n]. The over-complete transform matrix ( ) includes a set of elementary functions from two matrixes computed as  where I is the N ×N identity matrix and C is the N ×K cosine matrix. In an example  the input data sequence x ? RNxN can be represented as a linear combination of the prototype waveforms as the column vectors = { ?1| ?2| ?3|..........| ?M}:

[0081] In an example  the the sparsity filtering module 204  in communication with the sparse coefficient estimation module 304  computes the transform coefficient (a) by using a L1-norm minimization algorithm. For the input data sequence x[n] and the over-complete transform matrix ?  the transform coefficient (a) can be computed by solving the following L1-norm minimization problem:

[0082] In an example  the sparsity filtering module 204 is configured to filter the input signal x[n] by extracting the QRS complex portions from the transformation coefficient (a).
[0083] At step 708  the R-peak detection system 102 performs amplitude normalization of filtered data sequence d[n] of the input signal x[n]. In an example  the amplitude normalization is performed using the following function:

[0084] At step 710  the R-peak detection system 102 performs an absolute operation on the normalized data sequence of the input data sequence x[n]. In an example  a non-linear transformation  including the absolute operation  is performed to convert a bipolar filtered data sequence d[n] into a unipolar data sequence of the input signal x[n]. The absolute operation is performed to eliminate detection problems in case of negative QRS complexes. The absolute value of the filtered data sequence d[n] is computed as .
[0085] At step 712  the R-peak detection system 102 performs adaptive thresholding of the transformed data sequence of the input data sequence x[n]. In an example  the thresholding process is performed to eliminate spurious noise spikes and to reduce the number of false positive detections of the input signal x[n]. In an example  the thresholding function is defined as:

Where  the absolute values a[n] is smaller than a threshold parameter ?. In an embodiment  the adaptive-threshold parameter ? is computed for each input signal processing block. The threshold parameter ? is chosen based on the standard deviation of the absolute values of the normalized filtered data sequence of the input signal x[n].
[0086] At step 714  the envelope module 206 determines Shannon entropy of the thresholded data sequence of the input signal x[n]. In an example  the envelope module 206 uses the Shannon entropy based method to produce small deviations for the successive local maxima of the input signal x[n]. The Shannon entropy of the sequence is computed as .
[0087] At step 716  the R-peak detection system 102 applies smooth filtering on the Shannon entropy data sequence s[n] of the input signal x[n]. In an example  the Shannon entropy data sequence s[n] is smoothed using the zero-phase filtering with the rectangular impulse response (or rectangular window) of length (L). The smooth filtering reduces the effect of multiple peaks around QRS complex regions and provides sharp large local maxima around QRS complex portions. The smooth filtering provides smoothed Shannon entropy data sequence with isolated peaks corresponding to the QRS-complex portions in the input signal x[n].
[0088] At step 716  the Gaussian filtering module 208 convolves the smoothed Shannon entropy data sequence and first-order Gaussian derivative sequence. In an example  the Gaussian filtering module 208 convolves the smoothed Shannon entropy data sequence s[n] with a Gaussian derivative function g[n]. The output of the Gaussian filtering module 208 includes the positive and negative zero-crossing points of the input signal x[n]. In an example  the negative zero-crossing points indicate locations of the peaks in the Shannon entropy data sequence s[n].
[0089] At step 720  the zero-crossing detector module 210 detects the locations of the negative zero-crossings in the convolved data sequence of the input signal x[n].
[0090] At step 722  the output detector module 212 determines the locations of true peaks in input data sequence x[n]. In an example  the output detector module 212 uses the detected location of the negative zero-crossings (indicating the location of R-peaks) to automatically determine the time instants of the peaks in the input data sequence x[n]. In an example  the output detector module 212 combines the detection R-peaks to reduce false positive and false negative detections of the input signal x[n]. At step 724  the output detector module 212 outputs the determined information vector including the time instants of the peaks in the input data sequence x[n].
[0091] FIG. 8 depicts graphs 800 representing an example of experimental waveforms obtained by the R-peak detection system 102 of the FIG. 2  according to embodiments as disclosed herein. The performance of the R-peak detection system 102 is evaluated using an exemplary noisy or corrupted ECG signal x[n] including low-amplitude QRS  narrow QRS  and wide QRS complexes is shown in graph 802. The filtered signal d[n] obtained using the sparsity filtering module 204 is shown in graph 804. The Shannon entropy data sequence s[n] obtained using the envelope module 206 is shown in graph 806. The output signal z[n] obtained using the Gaussian filtering module 208 is shown in graph 808. Graph 810 shows the detected R-peaks of the input signal x[n].
[0092] FIG. 9 depicts graphs 900 representing an example of experimental waveforms showing performance detection for ECG signal with large P-waves and muscle noise  according to embodiments as disclosed herein. An exemplary ECG signal x[n] including with large P-waves and muscle noise is shown in graph 902. The filtered signal d[n] obtained using the sparsity filtering module 204 is shown in graph 904. The Shannon entropy data sequence s[n] obtained using the envelope module 206 is shown in graph 906. The output signal z[n] obtained using the Gaussian filtering module 208 is shown the graph 908. Graph 910 shows the detected R-peaks of the input signal x[n]. The R-peak detection system 102 produces 07 false positive beats and 02 false negative beats for a total of 1763 true beats.
[0093] FIG. 10 depicts graphs 1000 representing an example of experimental waveforms showing performance detection for ECG signal with noise and numerous long pauses  according to embodiments as disclosed herein. An exemplary ECG signal x[n] including noise and numerous long pauses up to 6 seconds is shown in graph 1002. The filtered signal d[n] obtained using the sparsity filtering module 204 is shown in graph 1004. The Shannon entropy data sequence s[n] obtained using the envelope module 206 is shown in graph 1006. The output signal z[n] obtained using the Gaussian filtering module 208 is shown in graph 1008. Graph 1010 shows the detected R-peaks of the input signal x[n]. The R-peak detection system 102 produces 02 false positive beats and 0 false negative beats for a total of 1780 true beats.
[0094] FIG. 11 depicts graphs 1100 representing an example of experimental waveforms showing performance detection for ECG signal with wide QRS complexes  according to embodiments as disclosed herein. An exemplary ECG signal x[n] including wide QRS complexes (premature ventricular contractions) is shown in graph 1102. The filtered signal d[n] obtained using the sparsity filtering module 204 is shown in graph 1104. The Shannon entropy data sequence s[n] obtained using the envelope module 206 is shown in graph 1106. The output signal z[n] obtained using the Gaussian filtering module 208 is shown in graph 1108. The detected R-peaks of the input signal x[n] is shown in Graph 1110. The R-peak detection system 102 produces 13 false positive beats and 0 false negative beats for a total of 2955 true beats.
[0095] The performance of the proposed method and system is evaluated using the noisy ECG signals taken from the standard MIT-BIH arrhythmia database at “Moody GB  Mark RG  The impact of the MIT-BIH Arrhythmia Database” www.physionet.org/physiobank/database/mitdb/” (Please refer more details). The preliminary experimental results of the method are shown in Figs. 4  5  and 8-11. The visual inspection of the experimental results shows that the method automatically determines the time instants of R-peaks in an ECG signal. The results also shows that the method captures the QRS complex portions of the ECG signal and increases detection accuracy of R-peaks in an ECG signal.
[0096] Further  unlike the existing systems  the experimental results shown in the below table illustrates that the proposed system and method includes lower false positive and false negative detection rates in the terms of ECG signals with the sharp P/T waves  negative QRS complex  small QRS complex  wider QRS complex  muscle noise  baseline wander  power-line interference  baseline drift  sudden changes in QRS amplitudes  sudden changes in QRS morphology  multiform PVCs  long pauses  and irregular heart rhythms.

[0097] Through the above description is described with respect to the ECG monitoring system  the person skilled in art can quickly identify that the invention can be used in any DSP systems. The present invention is capable of automatically detecting time instants of peaks in a recorded or received signal. Further  the present invention is capable of enhancing quality of a signal and suppressing any type of specific or complete noise or artifact from a recorded or received signal  without departing from the scope of the invention.
[0098] FIG. 12 depicts a computing environment implementing the application  in accordance with various embodiments of the present invention. As depicted the computing environment comprises at least one processing unit that is equipped with a control unit and an Arithmetic Logic Unit (ALU)  a memory  a storage unit  a clock chip  plurality of networking devices  and a plurality Input output (I/O) devices. The processing unit is responsible for processing the instructions of the algorithm. The processing unit receives commands from the control unit in order to perform its processing. Further  any logical and arithmetic operations involved in the execution of the instructions are computed with the help of the ALU.
[0099] The overall computing environment can be composed of multiple homogeneous and/or heterogeneous cores  multiple CPUs of different kinds  special media and other accelerators. The processing unit is responsible for processing the instructions of the algorithm. The processing unit receives commands from the control unit in order to perform its processing. Further  any logical and arithmetic operations involved in the execution of the instructions are computed with the help of the ALU. Further  the plurality of process units may be located on a single chip or over multiple chips.
[00100] The algorithm comprising of instructions and codes required for the implementation are stored in either the memory unit or the storage or both. At the time of execution  the instructions may be fetched from the corresponding memory and/or storage  and executed by the processing unit. The processing unit synchronizes the operations and executes the instructions based on the timing signals generated by the clock chip. The embodiments disclosed herein can be implemented through at least one software program running on at least one hardware device and performing network management functions to control the elements. The elements shown in FIGS. 1-3  6  and 7 include various units  blocks  modules  or steps described in relation with methods  processes  algorithms  or systems of the present invention  which can be implemented using any at least one of a general purpose processor  Digital Signal Processor (DSP)  multi-core application processor  Graphics Processing Unit (GPU)  Advanced RISC Machine (ARM) processor  multi-core processor or parallel processors  Field Programmable Gate Array (FPGA)  Application Specific Integrated Circuit (ASIC)  microcontroller  Software Defined Radio (SDR) tool  discrete hardware and analog circuit  and any combination of programming language  applications and embedded processor.
[00101] The foregoing description of the specific embodiments will so fully reveal the general nature of the embodiments herein that others can  by applying current knowledge  readily modify and/or adapt for various applications such specific embodiments without departing from the generic concept  and  therefore  such adaptations and modifications should and are intended to be comprehended within the meaning and range of equivalents of the disclosed embodiments. It is to be understood that the phraseology or terminology employed herein is for the purpose of description and not of limitation. Therefore  while the embodiments herein have been described in terms of preferred embodiments  those skilled in the art will recognize that the embodiments herein can be practiced with modification within the spirit and scope of the embodiments as described herein.

CLAIMS
We claim:
1. A method for automatically determining time instants of peaks in a signal  the method comprising:
determining a zero-mean data sequence of said signal;
filtering said zero-mean data sequence of said signal;
determining entropy of said filtered data sequence of said signal; and
determining said time instants of said peaks in said entropy data sequence of said signal.
2. The method of claim 1  wherein said method further comprises:
dividing said signal into at least one processing block; and
performing a mean subtraction of said signal to determine said zero-mean data sequence of said signal.
3. The method of claim 1  wherein filtering said zero-mean data sequence of said signal further comprises:
generating an over-complete transform matrix  wherein said over-complete transform matrix comprises a set of elementary functions of said signal;
estimating a transformation coefficient for said set of elementary function of said over-complete transform matrix; and
filtering said zero-mean data sequence of said signal using said estimated transform coefficient.
4. The method of claim 3  wherein said set of elementary functions of said signal comprises at least one column vector from at least one of an identity matrix  a cosine matrix  and a sine matrix.
5. The method of claim 4  wherein said at least one column vector of said identity matrix is used to extract high-frequency component of said signal.
6. The method of claim 4  wherein said at least one column vector of said at least one of cosine matrix and sine matrix is used to extract low-frequency component of said signal.
7. The method of claim 4  wherein size of said at least one of cosine matrix and sine matrix is less than size of said identity matrix.
8. The method of claim 1  wherein said entropy is Shannon entropy.
9. The method of claim 1  wherein determining said entropy of said filtered data sequence of said signal further comprises:
performing an amplitude normalization of said filtered data sequence of said signal 
performing an absolute operation of said normalized filtered data sequence of said signal  wherein said absolute operation is performed to transform bipolar filtered data sequence into unipolar filtered data of said signal 
performing adaptive thresholding on said transformed data sequence of said signal  and
determining said entropy of said thresholded data sequence of said signal.
10. The method of claim 9  wherein said absolute operation is performed using at least one of linear transformation and non-linear transformation.
11. The method of claim 1  wherein said method further comprises smoothing said determined entropy data sequence of said signal.
12. The method of claim 1  wherein said method further comprises:
convolving said smoothed entropy data sequence of said signal  wherein said convolved data sequence of said signal comprises at least one of positive zero-crossing point and negative zero-crossing point; and
detecting at least one location of said at least one negative zero-crossing point in said convolved data sequence of said signal.
13. The method of claim 9  wherein said at least one negative zero-crossing point indicate locations of said peaks in said entropy data sequence of said signal.
14. The method of claim 1  wherein said method further comprises using said at least one detected location of said at least one negative zero-crossing point to automatically determine said time instants of said peaks of said signal.
15. A system for automatically determining time instants of peaks in a signal  the system comprising:
a blocking and mean subtraction module configured to determine a zero-mean data sequence of said signal;
a sparsity filtering module configured to filter said zero-mean data sequence of said signal;
an envelope module configured to compute entropy of said filtered data sequence of said signal; and
an output detector module configured to automatically determine said time instants of said peaks in said entropy data sequence of said signal.
16. The system of claim 15  wherein said blocking and mean subtraction module is further configured to:
divide said signal into at least one processing block; and
perform a mean subtraction of said signal to determine said zero-mean data sequence of said signal.
17. The system of claim 15  wherein said signal is processed using at least one electrode lead.
18. The system of claim 15  wherein said sparsity filtering module further comprises:
a dictionary matrix generation module configured to construct an over-complete transform matrix  wherein said over-complete transform matrix comprises a set of elementary functions of said signal; and
a sparse coefficient estimation module configured to estimate a transformation coefficient for said set of elementary function of said over-complete transform matrix  wherein said sparsity filtering module is configured to use said estimated transform coefficient to filter said zero-mean data sequence of said signal.
19. The system of claim 18  wherein said set of elementary functions of said signal comprises at least one column vector from at least one of an identity matrix  a cosine matrix  and a sine matrix.
20. The system of claim 19  wherein said at least one column vector of said identity matrix is used to extract high-frequency component of said signal.
21. The system of claim 19  wherein said at least one column vector of said at least one of cosine matrix and sine matrix is used to extract low-frequency component of said signal.
22. The system of claim 19  wherein size of said at least one of cosine matrix and sine matrix is less than size of said identity matrix.
23. The system of claim 15  wherein said entropy is Shannon entropy.
24. The system of claim 15  wherein said envelope module is further configured to:
perform an amplitude normalization of said filtered data sequence of said signal 
perform an absolute operation of said normalized filtered data sequence of said signal  wherein said absolute operation is performed to transform bipolar filtered data sequence into unipolar filtered data of said signal 
perform adaptive thresholding on said transformed data sequence of said signal  and
compute said entropy of said thresholded data sequence of said signal.
25. The system of claim 24  wherein said absolute operation is performed using at least one of linear transformation and non-linear transformation.
26. The system of claim 15  wherein said system further comprises a smoothing filter configured to filter noise from said computed entropy data sequence of said signal.
27. The system of claim 26  wherein said system smoothing filter is configured to perform zero-phase filtering on said transformed data sequence of said signal.
28. The system of claim 26  wherein said smoothing filter comprises a rectangular impulse response of length L.
29. The system of claim 15  wherein said system further comprises:
a Gaussian filtering module configured to convolve said smoothed entropy data sequence of said signal  wherein said convolve data sequence of said signal comprises at least one of positive zero-crossing point and negative zero-crossing point; and
a zero-crossing detector module configured to detect at least one location of said at least one negative zero-crossing point in said convolved data sequence of said signal.
30. The system of claim 29  wherein said at least one negative zero-crossing point indicate locations of said peaks in said entropy data sequence of said signal.
31. The system of claim 29  wherein said zero-crossing detector module is configured to use at least one peak-finding logic to detect at least one negative zero-crossing point in said convolved data sequence of said signal.
32. The system of claim 15  wherein said output detector module is configured to use said at least one detected location of said at least one negative/positive zero-crossing point to automatically determine said time instants of said peaks/troughs of said signal.
33. The system of claim 15  wherein said output detector module is further configured to combine said detected peaks of said signal to reduce at least one of false positive detection and false negative detection.
34. The system of claim 15  wherein said system is further configured to select at least one filtering technique based on at least one digital signal processing application.