Method and system for detecting inter-beat intervals from a cardiac signal
By transforming cardiac signals into a 2D cepstrogram and using a predefined geometrical structure to track cepstral peaks, the method addresses inaccuracies in inter-beat interval detection, achieving enhanced accuracy in heart rate monitoring.
Patent Information
- Application Number
- PCT/SG2026/050294
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-09
- Filing Date
- 2026-05-09
- Publication Date
- 2026-11-12
AI Technical Summary
Existing methods for detecting inter-beat intervals from cardiac signals, particularly Ballistocardiography (BCG) signals, face challenges due to noise, motion artifacts, and dynamic signal morphology, leading to inaccuracies in heartbeat detection.
A method and system that transform cardiac signals into a 2D cepstrogram and utilize a predefined 2D geometrical structure to track cepstral peaks, defining a relationship between consecutive inter-beat intervals, thereby improving detection accuracy.
The method achieves significantly improved accuracy in detecting inter-beat intervals by robustly identifying correct peaks in the 2D cepstrogram, enhancing heart rate monitoring reliability.
Smart Images

Figure SG2026050294_12112026_PF_FP_ABST
Abstract
Description
METHOD AND SYSTEM FOR DETECTING INTER-BEAT INTERVALS FROM A CARDIAC SIGNALCROSS-REFERENCE TO RELATED APPLICATION
[0001] This application claims the benefit of priority of Singapore Patent Application No.10202501200P filed on 9 May 2025, the content of which being hereby incorporated by reference in its entirety for all purposes.TECHNICAL FIELD
[0002] The present invention generally relates to a method of detecting (or estimating) inter-beat intervals from a cardiac signal, and a system thereof.BACKGROUND
[0003] There are a variety of types of cardiac signal, such as a Ballistocardiography (BCG) signal, an electrocardiogram (ECG) signal and a photoplethysmogram (PPG) signal. Inter-beat interval (IB I) detection (or estimation) from, for example, a Ballistocardiography (BCG) signal offers valuable insights into cardiac activity by tracking the mechanical forces produced by the heart’s contractions. A key benefit of inter-beat interval detection from a cardiac signal is its non-invasive nature, for example, making it accessible for continuous cardiac monitoring in various settings, from clinical environments to everyday applications such as sleep studies and fitness tracking. Moreover, it provides crucial information about heart rate variability, which is an important indicator of cardiovascular health and autonomic nervous system balance. However, challenges arise in inter-beat interval detection from cardiac signals due to the inherently noisy and indirect nature of cardiac signals, and in particular, BCG signals. Factors such as motion artifacts, sensor placement, and individual variations in body composition can affect the accuracy of inter-beat interval detection from a cardiac signal. For example, challenges associated with BCG signals include non-consistency in signal waveform (morphology) and unpredictable signal quality. There have been disclosed sophisticated signal processing and machine learning techniques in an attempt to mitigate these challenges in interbeat interval detection from a cardiac signal for improving reliability and accuracy in heart rate monitoring.
[0004] Existing inter-beat interval detection (or estimation) methods may be broadly categorized into two main types: time-domain beat detection methods and signal self-similarityfunction methods. Each of these approaches has its own strengths and weaknesses when applied to detect or estimate inter-beat intervals, particularly from non-stationary signals such as BCG, ECG, or PPG signals.
[0005] For example, time-domain beat detection methods focus on identifying individual heartbeats by detecting significant points in the signal, typically peaks or zero crossings, which correlate to cardiac cycles. The time intervals between these detected beats are then calculated to estimate the inter-beat interval. However, one of the main problems of time-domain methods is that they are highly dependent on the signal's morphology (or the shape of the waveform). In practice, the morphology or shape of the signal can change dynamically due to a variety of factors, such as movement artifacts, physiological changes, and noise. These morphological changes can distort the signal, thereby making it difficult for simple peak detection algorithms to consistently identify heartbeats accurately. As a result, these time-domain beat detection methods may suffer from missed beats, false detections, and irregularities in the inter-beat interval detection or estimation.
[0006] For example, signal self-similarity function methods take a different approach by leveraging the repetitive nature of physiological signals. Instead of focusing on individual beats, these methods analyze the similarity between different segments of the signal to estimate or detect inter-beat intervals. Techniques such as autocorrelation or cross-correlation are commonly used, where the signal is compared against itself over time to detect periodic patterns. However, a significant challenge associated with self-similarity function methods is detecting / identifying the correct peaks in either the lag domain (time-shifted versions of the signal) or the time domain. For example, the correlation function may produce multiple peaks due to various periodicities or noise within the signal, thereby making it difficult to detect / identify which peak corresponds to the correct inter-beat interval. Choosing an incorrect peak can lead to inaccuracies in the inter-beat interval detection or estimation, especially when the signal contains harmonics, irregular heartbeats, or is affected by noise. This challenge is exacerbated in non-stationary signals, where periodicity is not constant.
[0007] A need therefore exists to provide a method of detecting (or estimating) inter-beat intervals from a cardiac signal, as well as a system thereof, that seek to overcome, or at least ameliorate, one or more deficiencies in existing methods or systems for detecting inter-beat intervals from a cardiac signal, and more particularly, with improved inter-beat interval detection (or estimation) accuracy. It is against this background that the present invention has been developed.SUMMARY
[0008] According to a first aspect of the present invention, there is provided a method of detecting inter-beat intervals from a cardiac signal using at least one processor, the method comprising:obtaining a cardiac signal of a subject, wherein the cardiac signal is in a time domain; transforming the cardiac signal into a two-dimensional (2D) cepstrogram comprising a 2D array of cepstral values in a 2D time-quefrency domain having a quefrency dimension and a time dimension;tracking a series of cepstral peaks in the 2D cepstrogram along the time dimension using a predefined 2D geometrical structure comprising a plurality of lines to respectively generate a plurality of cepstral peak tracks along the time dimension, wherein the predefined 2D geometrical structure defines a predefined geometrical relationship between consecutive interbeat intervals in the 2D cepstrogram; anddetecting inter-beat intervals from the cardiac signal based on the plurality of cepstral peak tracks.
[0009] According to a second aspect of the present invention, there is provided a system for detecting inter-beat intervals from a cardiac signal, the system comprising:at least one memory; andat least one processor communicatively coupled to the at least one memory and configured to:obtain a cardiac signal of a subject, wherein the cardiac signal is in a time domain; transform the cardiac signal into a 2D cepstrogram comprising a 2D array of cepstral values in a 2D time-quefrency domain having a quefrency dimension and a time dimension;track a series of cepstral peaks in the 2D cepstrogram along the time dimension using a predefined 2D geometrical structure comprising a plurality of lines to respectively generate a plurality of cepstral peak tracks along the time dimension, wherein the predefined 2D geometrical structure defines a predefined geometrical relationship between consecutive interbeat intervals in the 2D cepstrogram; anddetect inter-beat intervals from the cardiac signal based on the plurality of cepstral peak tracks.
[0010] According to a third aspect of the present invention, there is provided a computer program product, embodied in one or more non-transitory computer-readable storage mediums,comprising instructions executable by at least one processor to perform a method of detecting inter-beat intervals from a cardiac signal according to the above-mentioned first aspect of the present invention.BRIEF DESCRIPTION OF THE DRAWINGS
[0011] Embodiments of the present invention will be better understood and readily apparent to one of ordinary skill in the art from the following written description, by way of example only, and in conjunction with the drawings, in which:FIG. 1 depicts a schematic diagram of a method of detecting (or estimating) inter-beat intervals from a cardiac signal using at least one processor, according to various embodiments of the present invention;FIG. 2 depicts a schematic block diagram of a system for detecting inter-beat intervals from a cardiac signal, according to various embodiments of the present invention;FIG. 3 depicts a schematic flow diagram of an example method of detecting (or estimating) inter-beat intervals (IBIs) from a cardiac signal, according to various example embodiments of the present invention;FIG. 4 illustrates variable window size for cepstrum computation performed at an example single time point along the time dimension, according to various example embodiments of the present invention;FIG. 5 depicts an example of a high precision 2D cepstrogram transformed from a Ballistocardiogram (BCG) signal of 1 minute at 250 Hz sampling rate, according to various example embodiments of the present invention;FIG. 6 illustrates an example smoothing technique for the 2D cepstrogram along the time dimension using variable smoothing window size, according to various example embodiments of the present invention;FIG. 7 depicts an example smoothed 2D cepstrogram from the original (or sharp) 2D cepstrogram shown in FIG. 5, according to various example embodiments of the present invention;FIG. 8 shows a quefrency value curve (or a lag value curve) of the extracted cepstral peaks along the time dimension, according to various example embodiments of the present invention;FIG. 9 depicts a segmented region of interest (RO I) of the original (sharp) cepstrogram shown in FIG. 5, according to various example embodiments of the present invention;FIG. 10 illustrates an example cepstral peak tracking using a predefined 2D geometrical structure having three beaming orientations, namely, backward, middle and forward, according to various example embodiments of the present invention;FIG. 11 A depicts an example segmented ROI of the 2D cepstrogram,FIG. 1 IB shows extracted cepstral value curves (or cepstral energy value curves) from using the 2D geometrical structure comprising the three lines for cepstral peak tracking in the segmented ROI, along with the detected peaks and troughs, according to various example embodiments of the present invention;FIG. 11C shows extracted quefrency (or time lag) value curves, along with detected cepstral peaks corresponding to IBIs, according to various example embodiments of the present invention;FIG. 12A shows the detected IBIs (represented by horizonal white bars) from the backward beaming, according to various example embodiments of the present invention;FIG. 12B shows the detected IBIs (represented by horizonal white bars) from the forward beaming, according to various example embodiments of the present invention; and FIG. 13 depicts experimental results of the IBI detection estimation from BCG and electrocardiogram (ECG) signals.DETAILED DESCRIPTION
[0012] Various embodiments of the present invention provide a method of detecting (or estimating) inter-beat intervals from a cardiac signal, and a system thereof.
[0013] As discussed in the background, challenges arise in inter-beat interval detection in cardiac signals due to the inherently noisy and indirect nature of cardiac signals, and in particular, BCG signals. Factors such as motion artifacts, sensor placement, and individual variations in body composition can affect the accuracy of inter-beat interval detection from a cardiac signal. There are existing inter-beat interval detection (or estimation) methods for improving reliability and accuracy in heart rate monitoring, such as conventional time-domain beat detection methods and convention signal self-similarity function methods. However, conventional time-domain beat detection methods experience challenges associated with the dynamic morphology of cardiac signals (e.g., due to movement artifacts, physiological changes, and noise), and thus are prone to heartbeat detection inaccuracies, such as missed beats, false detections, and irregularities in the inter-beat interval detection (or estimation). Conventional signal self-similarity function methods experience challenges in identifying correct peakscorresponding to inter-beat intervals, and thus are prone to inter-beat interval detection inaccuracies. In this regard, various embodiments of the present invention provide a method of detecting (or estimating) inter-beat intervals from a cardiac signal, as well as a system thereof, that seek to overcome, or at least ameliorate, one or more deficiencies in existing methods or systems for detecting inter-beat intervals from a cardiac signal, and more particularly, with improved inter-beat interval detection (or estimation) accuracy.
[0014] FIG. 1 depicts a schematic diagram of a method 100 of detecting inter-beat intervals from a cardiac signal using at least one processor, according to various embodiments of the present invention. The method 100 comprises: obtaining (at 106) a cardiac signal (e.g., a BCG signal) of a subject, wherein the cardiac signal is in a time domain; transforming (at 108) the cardiac signal into a 2D cepstrogram comprising a 2D array of cepstral values in a 2D time-quefrency domain having a quefrency dimension (corresponding to a quefrency axis (e.g., a vertical axis)) and a time dimension (corresponding to a time axis (e.g., a horizontal axis)); tracking (at 110) a series of cepstral peaks in the 2D cepstrogram along the time dimension using a predefined 2D geometrical structure comprising a plurality of lines to respectively generate a plurality of cepstral peak tracks along the time dimension, wherein the predefined 2D geometrical structure defines a predefined geometrical relationship between consecutive inter-beat intervals in the 2D cepstrogram; and detecting (at 112) inter-beat intervals from the cardiac signal based on the plurality of cepstral peak tracks.
[0015] The method 100 of detecting inter-beat intervals from a cardiac signal advantageously has improved inter-beat interval detection (or estimation) accuracy. In particular, the method 100 transforms the cardiac signal into a 2D cepstrogram and then utilizes a predefined 2D geometrical structure, that is configured to define a particular or specific geometrical relationship between consecutive inter-beat intervals in the 2D cepstrogram, for tracking a series of cepstral peaks in the 2D cepstrogram along the time dimension (which may also be referred to as cepstral peak tracking). In this regard, the predefined 2D geometrical structure, which comprises a plurality of lines, is utilized to track the series of cepstral peaks in the 2D cepstrogram along the time dimension to generate a plurality of cepstral peak tracks along the time dimension based on the plurality of lines, respectively, of the predefined 2D geometrical structure. In this regard, various embodiments of the present invention found that by generating the plurality of cepstral peak tracks using such a predefined 2D geometrical structure, inter-beat intervals from the cardiac signal can be detected based on the plurality of cepstral peak tracks with improved (e.g., significantly improved) inter-beat interval detectionaccuracy. These advantages or technical effects, and / or other advantages or technical effects, will become more apparent to a person skilled in the art as the method 100 of detecting interbeat intervals from a cardiac signal, as well as the corresponding system for detecting inter-beat intervals, is described in more detail according to various embodiments and example embodiments of the present invention.
[0016] In various embodiments, the above-mentioned tracking (at 110) the series of cepstral peaks in the 2D cepstrogram along the time dimension comprises scanning (or traversing) the predefined 2D geometrical structure across the time dimension for the plurality of lines to respectively intersect with the series of cepstral peaks along the time dimension to generate the plurality of cepstral peak tracks respectively corresponding to where the plurality of lines respectively intersects with the series of cepstral peaks along the time dimension.
[0017] In various embodiments, the above-mentioned transforming (at 108) the cardiac signal into the 2D cepstrogram comprises, for each sliding time window (i.e., for each time point of a plurality of time points along the time dimension, the sliding time window at or associated with the time point) of the cardiac signal along the time dimension of the 2D cepstrogram, determining a column of cepstral values associated with the sliding time window of the cardiac signal based on a plurality of quefrency values (e.g., corresponding to a plurality of time lags) along the quefrency dimension, wherein the column of cepstral values comprises a plurality of cepstral values associated with the plurality of quefrency values, respectively, and the plurality of cepstral values is determined based on the plurality of quefrency values, respectively.
[0018] In various embodiments, for each sliding time window (i.e., for each time point of a plurality of time points along the time dimension, the sliding time window at or associated with the time point) of the cardiac signal, each of the plurality of cepstral values of the column of cepstral values associated with the sliding time window is determined based on a time window of the sliding time window, the time window having a length (or a size) determined based on the quefrency value associated with the cepstral value being determined.
[0019] In various embodiments, the length of the time window of the sliding time window determined based on the quefrency value (e.g., corresponding to the time lag) associated with the cepstral value being determined is double the quefrency value associated with the cepstral value being determined.
[0020] In various embodiments, the method 100 further comprises segmenting a region of interest (RO I) of the 2D cepstrogram in relation to the series of cepstral peaks along the timedimension to obtain a segmented ROI of the 2D cepstrogram. In this regard, the above-mentioned tracking (at 110) the series of cepstral peaks in the 2D cepstrogram is performed with respect to the segmented ROI of the 2D cepstrogram.
[0021] In various embodiments, the above-mentioned segmenting the ROI of the 2D cepstrogram comprises: applying, for each of the plurality of quefrency values along the quefrency dimension of the 2D cepstrogram, a sliding smoothing time window to the 2D cepstrogram at the quefrency value and along the time dimension to obtain a smoothed 2D cepstrogram; extracting a quefrency value curve corresponding to the series of cepstral peaks along the time dimension from the smoothed 2D cepstrogram; and segmenting the ROI of the 2D cepstrogram based on the extracted quefrency value curve.
[0022] In various embodiments, for each of the plurality of quefrency values along the quefrency dimension of the 2D cepstrogram, the sliding smoothing time window applied to the 2D cepstrogram at the quefrency value and along the time dimension has a length (or a size) equal to the quefrency value.
[0023] In various embodiments, the plurality of lines of the predefined 2D geometrical structure comprises a backward oriented straight line, a middle oriented straight line and a forward oriented straight line arranged such that the middle oriented straight line is along the quefrency dimension, the forward oriented straight line is slanted forward in the time dimension and makes a relative acute angle with the middle oriented straight line and the backward oriented straight light is slanted backward in the time dimension and makes a relative acute angle with the middle oriented straight line. Furthermore, the plurality of cepstral peak tracks comprising a first cepstral peak track, a second cepstral peak track and a third cepstral peak track generated based on the backward oriented straight line, the middle oriented straight line and the forward oriented straight line, respectively.
[0024] In various embodiments, the relative acute angle which the forward oriented straight line makes with the middle oriented straight line and the relative acute angle which the backward oriented straight line makes with the middle oriented straight line are equal. In various embodiments, the relative acute angle which the forward oriented straight line makes with the middle oriented straight line and the relative acute angle which the backward oriented straight line makes with the middle oriented straight line are each arctan(0.5).
[0025] In various embodiments, the above-mentioned detecting (at 112) the inter-beat intervals from the cardiac signal comprises: determining a plurality of aligned sets, each aligned set comprising a local maximum cepstral peak of the first cepstral peak track, a local minimumcepstral peak of the second cepstral peak track and a local maximum cepstral peak of the third cepstral peak track that are substantially aligned in the time dimension (close proximity in the time dimension); and for each of the plurality of aligned sets, the quefrency value associated with the local maximum cepstral peak of the first cepstral peak track of the aligned set corresponds to an inter-beat interval detected based on the backward oriented straight line, and the quefrency value associated with the local maximum cepstral peak of the third cepstral peak track of the aligned set corresponds to an inter-beat interval detected based on the forward oriented straight line.
[0026] In various embodiments, the above-mentioned detecting (at 112) the inter-beat intervals from the cardiac signal further comprises: generating a first cepstral value curve, a second cepstral value curve and a third cepstral value curve based on the first cepstral peak track, the second cepstral peak track and the third cepstral peak track, respectively; and determining, for each of the plurality of aligned sets, the local maximum cepstral peak of the first cepstral peak track of the aligned set based on a corresponding local maximum cepstral value of the first cepstral value curve, the local minimum cepstral peak of the second cepstral peak track of the aligned set based on a corresponding local minimum cepstral value of the second cepstral value curve and the local maximum cepstral peak of the third cepstral peak track of the aligned set based on a corresponding local maximum of the third cepstral value curve. In this regard, the corresponding local maximum cepstral value of the first cepstral value curve, the corresponding local minimum cepstral value of the second cepstral value curve and the corresponding local maximum of the third cepstral value curve are substantially aligned in the time dimension (close proximity in the time dimension).
[0027] In various embodiments, the above-mentioned detecting (at 112) the inter-beat intervals from the cardiac signal further comprises: obtaining a first set of inter-beat intervals across the time dimension comprising a plurality of inter-beat intervals detected based on the backward oriented straight line; obtaining a second set of inter-beat intervals across the time dimension comprising a plurality of inter-beat intervals detected based on the forward oriented straight line; and fusing the first and second sets of inter-beat intervals to obtain an output set of inter-beat intervals across the time dimension.
[0028] FIG. 2 depicts a schematic block diagram of a system 200 for detecting inter-beat intervals from a cardiac signal, according to various embodiments of the present invention, corresponding to the above-mentioned method 100 of detecting inter-beat intervals from a cardiac signal as described hereinbefore with reference to FIG. 1 according to variousembodiments of the present invention. The system 200 comprises: at least one memory 202; and at least one processor 204 communicatively coupled to the at least one memory 202 and configured to perform the method 100 of detecting inter-beat intervals from a cardiac signal according to various embodiments of the present invention. Accordingly, the at least one processor 204 is configured to: obtain a cardiac signal of a subject, wherein the cardiac signal is in a time domain; transform the cardiac signal into a 2D cepstrogram comprising a 2D array of cepstral values in a 2D time-quefrency domain having a quefrency dimension and a time dimension; track a series of cepstral peaks in the 2D cepstrogram along the time dimension using a predefined 2D geometrical structure comprising a plurality of lines to respectively generate a plurality of cepstral peak tracks along the time dimension, wherein the predefined 2D geometrical structure defines a predefined geometrical relationship between consecutive inter-beat intervals in the 2D cepstrogram; and detect inter-beat intervals from the cardiac signal based on the plurality of cepstral peak tracks.
[0029] It will be appreciated by a person skilled in the art that the at least one processor 204 may be configured to perform various functions or operations through set(s) of instructions (e.g., software modules) executable by the at least one processor 204 to perform various functions or operations. Accordingly, as shown in FIG. 2, the system 200 may comprise: a cardiac signal obtaining module (or a cardiac signal obtaining circuit) 206 configured to obtain a cardiac signal of a subject, wherein the cardiac signal is in a time domain; a cepstrogram transformation module (or a cepstrogram transformation circuit) 208 configured to transform the cardiac signal into a 2D cepstrogram comprising a 2D array of cepstral values in a 2D time-quefrency domain having a quefrency dimension and a time dimension; a cepstral peak tracking module (a cepstral peak tracking circuit) 210 configured to track a series of cepstral peaks in the 2D cepstrogram along the time dimension using a predefined 2D geometrical structure comprising a plurality of lines to respectively generate a plurality of cepstral peak tracks along the time dimension, wherein the predefined 2D geometrical structure defines a predefined geometrical relationship between consecutive inter-beat intervals in the 2D cepstrogram; and an inter-beat interval detection module (or an inter-beat interval detection circuit) 212 configured to detect inter-beat intervals from the cardiac signal based on the plurality of cepstral peak tracks.
[0030] It will be appreciated by a person skilled in the art that the above-mentioned modules of the system 200 for detecting inter-beat intervals are not necessarily separate modules, and two or more modules may be realized by or implemented as one functional module (e.g., acircuit or a software program) as desired or as appropriate without deviating from the scope of the present invention. For example, two or more of the cardiac signal obtaining module 206, cepstrogram transformation module 208, the cepstral peak tracking module 210 and the interbeat interval detection module 212 may be realized (e.g., compiled together) as one executable software program (e.g., software application), which for example may be stored in the at least one memory 202 and executable by the at least one processor 204 to perform the corresponding functions or operations as described herein according to various embodiments of the present invention.
[0031] In various embodiments, the system 200 for detecting inter-beat intervals corresponds to the method 100 of detecting inter-beat intervals as described hereinbefore with reference to FIG. 1, therefore, various operations, functions or steps configured to be performed by the at least one processor 204 may correspond to various operations, functions or steps of the method 100 of detecting inter-beat intervals described hereinbefore according to various embodiments, and thus need not be repeated with respect to the system 200 for detecting interbeat intervals for clarity and conciseness. In other words, various embodiments described herein in context of methods (e.g., the method 100 of detecting inter-beat intervals) are analogously valid for the corresponding systems or devices (e.g., the system 200 for detecting inter-beat intervals), and vice versa. For example, in various embodiments, the at least one memory 202 may have stored therein the cardiac signal obtaining module 206, cepstrogram transformation module 208, the cepstral peak tracking module 210 and / or the inter-beat interval detection module 212, which respectively correspond to various operations, functions or steps of the method 100 of detecting inter-beat intervals as described hereinbefore according to various embodiments, which are executable by the at least one processor 204 to perform the corresponding operations, functions or steps as described herein.
[0032] A computing system, a controller, a microcontroller or any other system providing a processing capability may be provided according to various embodiments in the present invention. Such a system may be taken to include one or more processors and one or more computer-readable storage mediums. For example, the system 200 for detecting inter-beat intervals for a subject described hereinbefore includes at least one processor 204 and at least one computer-readable storage medium (or memory) 202 which are for example used in various processing carried out therein as described herein. A memory or computer-readable storage medium used in various embodiments may be a volatile memory, for example a DRAM (Dynamic Random Access Memory) or a non-volatile memory, for example a PROM(Programmable Read Only Memory), an EPROM (Erasable PROM), EEPROM (Electrically Erasable PROM), or a flash memory, e.g., a floating gate memory, a charge trapping memory, an MRAM (Magnetoresistive Random Access Memory) or a PCRAM (Phase Change Random Access Memory).
[0033] In various embodiments, a “circuit” may be understood as any kind of a logic implementing entity, which may be special purpose circuitry or a processor executing software stored in a memory, firmware, or any combination thereof. Thus, in an embodiment, a “circuit” may be a hard-wired logic circuit or a programmable logic circuit such as a programmable processor, e.g., a microprocessor (e.g., a Complex Instruction Set Computer (CISC) processor or a Reduced Instruction Set Computer (RISC) processor). A “circuit” may also be a processor executing software, e.g., any kind of computer program, e.g., a computer program using a virtual machine code, e.g., Java. Any other kind of implementation of various functions or operations may also be understood as a “circuit” in accordance with various other embodiments. Similarly, a “module” may be a portion of a system according to various embodiments in the present invention and may encompass a “circuit” as above, or may be understood to be any kind of a logic-implementing entity therefrom.
[0034] Some portions of the present disclosure may be explicitly or implicitly presented in terms of algorithms and functional or symbolic representations of operations on data within a computer memory. These algorithmic descriptions and functional or symbolic representations are the means used by those skilled in the data processing arts to convey most effectively the substance of their work to others skilled in the art. An algorithm may be, and generally, conceived to be a self-consistent sequence of steps leading to a desired result.
[0035] The present specification also discloses systems (e.g., which may also be embodied as devices or apparatuses), such as the system 200 for detecting inter-beat intervals, for performing various operations, functions or steps of various methods described herein. Such systems may each be specially constructed for the required purposes or may comprise a general purpose computer system selectively activated or reconfigured by a computer program stored in the computer system. In general, various algorithms that may be presented herein are not limited to being implemented or executed by any particular computer system. Alternatively, the construction of more specialized computer system (e.g., a high-performance computer system) to perform various operations, functions or steps of various methods described herein may be provided as desired or as appropriate without going beyond the scope of the present invention.
[0036] In addition, the present specification also at least implicitly discloses computer program(s) or software / functional module(s), in that it would be apparent to a person skilled in the art that various operations, functions or steps of various methods described herein may be put into effect by computer code. The computer program(s) is not intended to be limited to any particular programming language and implementation thereof, and it will be appreciated by a person skilled in the art that a variety of programming languages and coding thereof may be used to implement the computer program(s). Moreover, the computer program(s) is not intended to be limited to any particular control flow as there are a variety of programming languages which can use different control flows. It will be appreciated by a person skilled in the art that a computer program may be stored on any computer-readable storage medium (non-transitory computer-readable storage medium), such as but not limited to, a magnetic disk, an optical disk or a memory chip. For example, a computer program stored on a computer-readable storage medium may be loaded and executed on a computer system to implement various operations, functions or steps of various methods described herein according to various embodiments of the present invention.
[0037] Accordingly, in various embodiments, there is provided a computer program product, embodied in one or more computer-readable storage mediums (non-transitory computer-readable storage medium), comprising instructions (e.g., the cardiac signal obtaining module 206, cepstrogram transformation module 208, the cepstral peak tracking module 210 and / or the inter-beat interval detection module 212) executable by one or more computer processors to perform the method 100 of detecting inter-beat intervals as described hereinbefore with reference to FIG. 1 according to various embodiments of the present invention. Accordingly, various computer programs or software modules described herein may be stored in a computer program product receivable by a system therein for execution by at least one processor of the system to perform various operations, functions or steps of various methods described herein according to various embodiments of the present invention.
[0038] It will be appreciated by a person skilled in the art that various modules of systems described herein (e.g., the cardiac signal obtaining module 206, cepstrogram transformation module 208, the cepstral peak tracking module 210 and / or the inter-beat interval detection module 212) may be software module(s) realized by computer program(s) or set(s) of instructions executable by a computer processor to perform various functions or operations. Various modules described herein (e.g., the cardiac signal obtaining module 206, cepstrogram transformation module 208, the cepstral peak tracking module 210 and / or the inter-beat intervaldetection module 212) may also be implemented as hardware module(s) being functional hardware unit(s) designed to perform various functions or operations. More particularly, in the hardware sense, a module is a functional hardware unit designed for use with other components or modules. For example, a module may be implemented using discrete electronic components, or it can form a portion of an entire electronic circuit such as an Application Specific Integrated Circuit (ASIC) or a Field Programmable Gate Array (FPGA). Numerous other possibilities exist. It will also be appreciated by a person skilled in the art that a combination of hardware and software modules may be implemented. Furthermore, various operations, functions or steps of various methods described herein may be performed in parallel rather than sequentially as desired or as appropriate (e.g., as long as it does not render the method(s) inoperable or unsatisfactory for its intended purpose).
[0039] It will be appreciated by a person skilled in the art that the terminology used herein is for the purpose of describing various embodiments only and is not intended to be limiting of the present invention. As used herein, the singular forms “a”, “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises” and / or “comprising,” when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.
[0040] Any reference to an element or a feature herein using a designation such as “first”, “second” and so forth does not limit the quantity or order of such elements or features, unless stated or the context requires otherwise. For example, such designations may be used herein as a convenient way of distinguishing between two or more elements or instances of an element. Thus, a reference to first and second elements does not necessarily mean that only two elements can be employed, or that the first element must precede the second element, unless stated or the context requires otherwise. In addition, a phrase referring to “at least one of’ a list of items refers to any single item therein or any combination of two or more items therein.
[0041] In order that the present invention may be readily understood and put into practical effect, various example embodiments of the present invention will be described hereinafter by way of examples only and not limitations. It will be appreciated by a person skilled in the art that the present invention may, however, be embodied in various different forms or configurations and should not be construed as limited to the example embodiments set forth hereinafter. Rather, these example embodiments are provided so that this disclosure will bethorough and complete, and will fully convey the scope of the present invention to those skilled in the art.
[0042] Inter-beat interval (IBI) detection (or estimation) in cardiac signals, such as Ballistocardiogram (BCG) signals, has great benefits in many medical and healthcare applications but is also a challenging task such as due to the noisy and non-stationary nature of BCG signals. As discussed in the background, existing IBI detection methods can be broadly categorized into two main types: time-domain beat detection methods and signal self-similarity function methods. The existing time-domain beat detection methods suffer from the dynamic morphology of the signal waveform, while the existing self-similarity function methods face challenges in detecting / identifying the correct peak either in the lag domain or time domain. In this regard, various example embodiments of the present invention provide a method of detecting (or estimating) IBIs from a cardiac signal with improved IBI detection (or estimation) accuracy, and more particularly as will be described later below, that is based on detecting correct cepstral peaks corresponding to IBIs in the 2D time-quefrency domain. In particular, as will be described later below, the method relies on a predefined geometrical relationship between two consecutive IBIs (every two consecutive IBIs) in the 2D cepstrogram, which is defined by a predefined 2D geometrical structure. Various example embodiments advantageously found that, by utilizing the predefined 2D geometrical structure, correct cepstral peaks in the 2D time-quefrency domain, each corresponding to two consecutive IBIs can be robustly and accurately detected / identified. To demonstrate the improved effectiveness and accuracy of the method of detecting (or estimating) IBIs from a cardiac signal according to various example embodiments of the present invention, the method is evaluated with experiment data of example cardiac signals (e.g., BCG and ECG signals) and will be discussed later below.
[0043] Accordingly, various example embodiments advantageously provide a unique method of detecting (or estimating) IBIs from a cardiac signal based on geometrical features associated every two consecutive IBIs in the 2D cepstrogram transformed from the time-domain cardiac signal.
[0044] FIG. 3 depicts a schematic flow diagram of an example method 300 of detecting (or estimating) IBIs from a cardiac signal, according to various example embodiments of the present invention. By way of an illustrative example only and without limitation, the example method 300 will be described with respect to a BCG signal.
[0045] In an initial step 308, after obtaining a ID time domain BCG signal of a subject, the ID time domain BCG signal is transformed to a 2D cepstrogram with high time-quefrency precision. In this regard, the ID time domain BCG signal is transformed into a 2D cepstrogram comprising a 2D array of cepstral values in a 2D time-quefrency domain having a quefrency dimension (corresponding to a quefrency axis (e.g., a vertical axis)) and a time dimension (corresponding to a time axis (e.g., a horizontal axis)). In various example embodiments, a variable window size is used for the computation of the cepstrum as illustrated in FIG. 4. In particular, for each sliding time window (i.e., for each time point of a plurality of time points along the time dimension, the sliding time window at or associated with the time point) of the cardiac signal along the time dimension of the 2D cepstrogram, a column of cepstral values associated with the sliding time window of the cardiac signal is determined based on a plurality of quefrency values (e.g., corresponding to a plurality of time lags, respectively) along the quefrency dimension of the 2D cepstrogram. In this regard, the column of cepstral values comprises a plurality of cepstral values associated with the plurality of quefrency values, respectively, and the plurality of cepstral values is determined based on the plurality of quefrency values, respectively. In various example embodiments, for each sliding time window of the cardiac signal, each of the plurality of cepstral values of the column of cepstral values associated with the sliding time window is determined based on a time window of the sliding time window, the time window having a length (or a size) determined based on the quefrency value associated with the cepstral value being determined. In various example embodiments, the length of the time window of the sliding time window determined based on the quefrency value (e.g., corresponding to the time lag) associated with the cepstral value being determined is double the quefrency value associated with the cepstral value being determined.
[0046] Accordingly, a plurality of time lags corresponds to a plurality of quefrency values, respectively, along the quefrency dimension of the 2D cepstrogram, and a plurality of time points is along the time dimension of the 2D cepstrogram. In this regard, each sliding time window corresponds to a time window (e.g., a center thereof) at a certain time point of the plurality of time points along a time dimension (or a time axis). In stark contrast to the conventional fixed-window cepstrum, in various example embodiments of the present invention, each cepstral value of a plurality of cepstral values of a column of cepstral values associated with a given sliding time window is computed from a time window (e.g., a subwindow) whose length is determined based on the quefrency value (or the time lag) associated with that cepstral value being computed. In this regard, in various example embodiments, for atarget quefrency value M, the cepstral value at M for a column of cepstral values associated with a given sliding time window is computed from a time window (e.g., sub-window) of length 2M extracted from (or defined relative to) a sliding time window.
[0047] Accordingly, a single sliding time window position (e.g., the time point along the time dimension corresponding to a center of the sliding time window) yields a plurality of cepstral values (forming a column of cepstral values), each cepstral value derived from a time window (e.g., sub-window) of a different length based to its corresponding quefrency value. This process of generating a column of cepstral values is repeated for each sliding time window position at each of the plurality of time points along the time dimension (e.g., according to window sliding steps) to populate or form the full 2D cepstrogram across both the quefrency and time dimensions. This per-quefrency adaptation of the time window length according to various example embodiments advantageously enables the transformation of the ID time domain cardiac signal (e.g., a BCG signal) to a 2D cepstrogram with high time-quefrency precision.
[0048] By forming a 2D cepstrogram with high time-quefrency precision, for example, various embodiments of the present invention advantageously enables the ability to accurately localize and resolve cepstral features for non-stationary signals where the periodic structure of interest may persist for only a single cycle. In the conventional cepstrum computation, c[n] = IFFT(log|FFT(x[n])|), a single fixed window of length L is applied, and cepstral values are obtained for all quefrencies up to L / 2 (values above L / 2 are discarded due to the even symmetry of the cepstrum for real signals). While computationally efficient, this conventional fixed-window approach is suboptimal for non-stationary signals. For example, various example embodiments found that when L is substantially larger than the quefrency of interest, the underlying signal characteristics may vary across the window, causing the cepstral peak corresponding to a single-cycle repetition to be smeared or attenuated. To address this technical problem, various example embodiments of the present invention adapt the analysis window for determining the cepstral value to the quefrency being evaluated, as described above. In this regard, in various example embodiments, for each target quefrency M (where M < L / 2), the cepstral value at M is computed using a time window of length exactly 2M, rather than the fixed length L. This per-quefrency window adaptation is repeated for every target quefrency M of interest, which yields markedly higher precision. Accordingly, various example embodiments found that using exactly two times the target quefrency for the window size provide the best or optimal cepstral precision for the 2D cepstrogram.
[0049] Without wishing to be bound by theory but for better understanding, a rationale for the variable window size will now be explained. First, because an objective is to detect a single cycle of repetition in a non-stationary signal, the analysis window may preferably be just long enough to contain that cycle and no longer as any excess length introduces signal content from outside the cycle of interest, which may smear the cepstral peak. Second, determining the window length based on the quefrency value associated with the cepstral value being determined (e.g., matching the window length to 2M) ensures that the spectral estimate underlying the cepstrum reflects only the local signal behavior associated with the target quefrency, preserving the sharpness of the cepstral peak. Accordingly, in various example embodiments, high time-quefrency precision may be achieved by computing each cepstral value using a window of length based on the target quefrency value M associated with the cepstral value being determined (e.g., precisely 2M tailored to the quefrency value M under evaluation), rather than relying on a single fixed window length for the entire cepstrum.
[0050] In various example embodiments, the target or candidate quefrencies (or target or candidate time lags) (corresponding to the above-mentioned plurality of quefrency values) are determined by the physiologically plausible range of beat-to-beat intervals corresponding to an expected heart rate range. For example, each target time lag M represents a hypothesized period of cardiac repetition, so the set of target time lags may be chosen to span the inter-beat intervals associated with the lowest and highest heart rates of interest.
[0051] On time precision for the 2D cepstrogram, according to various example embodiments, a small window sliding step size is used. In various example embodiments, the window sliding step size may be as small as one sample size. In this regard, one sample size corresponds to the reciprocal of the sampling rate of the signal, expressed in time units. Specifically, for a signal sampled at rate fs(in Hz), the duration of one sample is: Ts— / fs. For example, for a BCG signal sampled at 250 Hz, one sample corresponds to 1 / 250 = 0.004seconds, i.e., 4 milliseconds.
[0052] Accordingly, FIG. 4 illustrates the variable window size for cepstrum computation performed at a single time point along the time dimension, according to various example embodiments of the present invention. In various example embodiments, the single time point corresponds to one position of the sliding time window centered at a particular sample of the cardiac signal. At this single time point along the time dimension, the transformation of the corresponding sliding time window of the cardiac signal (e.g., illustrated in FIG. 4) yields a plurality of cepstral values (which may be referred to as a column of cepstral values), one foreach target quefrency value (or each time lag) M in a predetermined quefrency value range (or time lag range) along the quefrency (or time lag) dimension, where each cepstral value is computed from a time window (e.g., a sub-window) of length 2M as described hereinbefore. Accordingly, the plurality of cepstral values obtained at one sliding time window position is arranged along the quefrency axis according to their associated quefrency values. This ordered set of cepstral values thus constitutes a single column of the 2D cepstrogram being formed, where: the horizontal position of the column corresponds to the time point (or time index) of the sliding time window center, and the vertical position within the column corresponds to the quefrency value (or time lag) M. The pixel intensity (or color) at each 2D point (time, quefrency) coordinate of the 2D cepstrogram represents the cepstral value (or the magnitude thereof) at that quefrency value for that time point. Accordingly, the 2D cepstrogram comprises a 2D array of cepstral values in the 2D time-quefrency domain.
[0053] In various example embodiments, to generate successive columns of the 2D cepstrogram to produce the full 2D cepstrogram (e.g., as illustrated in FIG. 5), the center of the sliding time window is successively or iteratively advanced to the next sample along the cardiac signal according to a predefined window sliding step along the time dimension, and at each iteration, the sliding time window of the sample of cardiac signal at the new center position associated with the iteration is transformed in the manner as described hereinbefore with reference to FIG. 4. Therefore, each successive center position of the sliding time window at each successive time point along the time-dimension yields the next column of cepstral values (or cepstrogram). Accordingly, the transformation is repeated at each of a plurality of time points for the sliding time window centered at the time point, across the plurality of time points of a duration of the cardiac signal, and the columns of cepstral values obtained are assembled or arranged side by side to form the complete 2D cepstrogram, with the horizontal axis representing the time axis and the vertical axis representing the quefrency (or time lag) axis.
[0054] The time step size (window sliding step size) between columns of cepstral values will now be described according to various example embodiments of the present invention. The time step size by which the sliding time window center advances in the time dimension determines the temporal resolution of the 2D cepstrogram generated. A time step size of one sample produces the finest temporal resolution and a column of cepstral values for every signal sample. A larger step reduces computational load at the cost of coarser temporal granularity. Accordingly, the time step size may be selected based on the desired temporal resolution of the 2D cepstrogram and the desired / available computational resources. Therefore, it will beappreciated by a person skilled in the art that the time step size is not limited to any particular or specific size. Accordingly, each column of the 2D cepstrogram is a stack of per-quefrency cepstral values computed at a corresponding one sliding time window position in the manner as described hereinbefore with respect to FIG. 4, and the 2D cepstrogram is constructed by sliding the sliding time window center across the cardiac signal and concatenating the resulting columns of cepstral values along the time axis.
[0055] FIG. 5 shows an example of a high precision 2D cepstrogram transformed from a BCG signal of 1 minute at 250 Hz sampling rate. In this regard, each individual cepstral peak in the 2D cepstrogram corresponds to the presence of one IBI from two consecutive heart beats. For example, the time lags M along the quefrency axis of the 2D cepstrogram shown in FIG. 5 span 0.5 to 1.5 seconds (corresponding to roughly 40-120 bpm). In this regard, using the 2M rule described hereinbefore according to various example embodiments of the present invention, the time window (e.g., sub-window) used to compute the cepstral value at each time lag M has a length exactly 2M. Therefore, across the time lag range shown in FIG. 5, the time window lengths range from 1.0 second (for M = 0.5 seconds) to 3.0 seconds (for M = 1.5 seconds).
[0056] In a subsequent step 318, a region of interest (ROI) of the 2D cepstrogram is segmented such that the following subsequent processes or steps (including cepstral peak tracking which will be described later below) for detecting the IBIs are performed with respect to the segmented ROI of the 2D cepstrogram. In this regard, a ROI of the 2D cepstrogram is segmented in relation to the series of cepstral peaks (e.g., as shown in FIG. 5) along the time dimension to obtain a segmented ROI of the 2D cepstrogram. Accordingly, the ROI may be a narrower region (with a narrower margin) that contains cepstral peaks corresponding a target IBI range. In various example embodiments, the target IBI range is derived directly from the target heart rate range via the relationship IBI = 60 / HR (with IBI in seconds and HR in bpm). For example, the heart rate range itself may be either: a fixed, wide range that covers the full physiologically plausible span for an intended population (e.g., 40-200 bpm for general adult monitoring, corresponding to an IBI range of 0.3-1.5 seconds), or a customized, narrower range tailored to the application scenario, such as 40-100 bpm for resting / sleep monitoring (IBI 0.6-1.5 s), 100-220 bpm for neonatal monitoring, or a subject-specific range derived from prior measurements of a subject. For example, narrowing the range based on the application reduces computation and may improve robustness by excluding implausible lags from consideration. An advantage of the ROI segmentation is to eliminate ad hoc noisy spike that may be presentin a region / area far from relevant / plausible heart rates and to reduce the computation complexity as well.
[0057] In various example embodiments, the ROI segmentation comprises: applying, for each of the plurality of quefrency values along the quefrency dimension of the 2D cepstrogram, a sliding smoothing time window to the 2D cepstrogram at the quefrency value and along the time dimension to obtain a smoothed 2D cepstrogram; extracting a quefrency value curve corresponding to the series of cepstral peaks along the time dimension from the smoothed 2D cepstrogram; and segmenting the ROI of the 2D cepstrogram based on the extracted quefrency value curve.
[0058] FIG. 6 illustrates an example smoothing process according to various example embodiments of the present invention. In particular, FIG. 6 shows an example of smoothing the 2D cepstrogram along the time dimension using variable smoothing window size. The smoothing process is conducted in time-domain. In various example embodiments, for each of the plurality of quefrency values along the quefrency dimension of the 2D cepstrogram, the sliding smoothing time window (e.g., hamming) applied to the 2D cepstrogram (which may be referred to as the original or sharp 2D cepstrogram) at the quefrency value and along the time dimension has a length (or a size) equal to the quefrency value (or the time lag value). For example, as shown in FIG. 6, for a given time lag ‘a’ at time ‘t’, the size of the sliding smoothing time window may be from t-a / 2 to t+a / 2. For example, the step size of the plurality of quefrency values (or lag values) may be one sample size, and the step size of the sliding smoothing time window may also be one sample size in both the quefrency (time lag) dimension and time dimension. By applying the smoothing process, the resultant smoothed cepstrogram exhibits more continuity in the time-domain, and thus the cepstral peak can be detected in the cepstral domain more steadily. FIG. 7 shows an example smoothed 2D cepstrogram computed from the original (or sharp) 2D cepstrogram shown in FIG. 5.
[0059] The quefrency values (or time lag values) of the series of cepstral peaks along the time domain detected in the cepstral domain as shown in FIG. 8 represent candidates of IBIs from the cardiac signal. In particular, FIG. 8 shows the quefrency value curve (or the lag value curve) of the extracted cepstral peaks along the time dimension. In various example embodiments, the segmentation of a ROI is performed by applying a narrow margin (e.g., 50ms) in the quefrency dimension to both upper and lower bounds of the quefrency value curve (or the lag value curve). The ROI may then be applied to original (sharp) 2D cepstrogram as shownin FIG. 9. In particular, FIG. 9 shows a segmented ROI of the original sharp cepstrogram shown in FIG. 5.
[0060] In a subsequent step 322, cepstral peak tracking of the series of cepstral peaks is performed in the segmented ROI of the original 2D cepstrogram. In various example embodiments, the tracking of the series of cepstral peaks along the time dimension is performed using a predefined 2D geometrical structure comprising a plurality of lines to respectively generate a plurality of cepstral peak tracks along the time dimension. In this regard, the predefined 2D geometrical structure defines a predefined geometrical relationship between consecutive inter-beat intervals in the 2D cepstrogram. The IBIs from the cardiac signal are then detected based on the plurality of cepstral peak tracks. In various example embodiments, the tracking of the series of cepstral peaks along the time dimension comprises scanning (or traversing) the predefined 2D geometrical structure across the time dimension for the plurality of lines to respectively intersect with the series of cepstral peaks along the time dimension to generate the plurality of cepstral peak tracks respectively corresponding to where the plurality of lines respectively intersects with the series of cepstral peaks along the time dimension.
[0061] In various example embodiments, the plurality of lines of the predefined 2D geometrical structure comprises a backward oriented straight line, a middle oriented straight line and a forward oriented straight line arranged such that the middle oriented straight line is along the quefrency dimension, the forward oriented straight line is slanted forward in the time dimension and makes a relative acute angle with the middle oriented straight line and the backward oriented straight light is slanted backward in the time dimension and makes a relative acute angle with the middle oriented straight line. Furthermore, the plurality of cepstral peak tracks comprising a first cepstral peak track, a second cepstral peak track and a third cepstral peak track generated based on the backward oriented straight line, the middle oriented straight line and the forward oriented straight line, respectively.
[0062] In various example embodiments, the relative acute angle which the forward oriented straight line makes with the middle oriented straight line and the relative acute angle which the backward oriented straight line makes with the middle oriented straight line are equal. In various example embodiments, the relative acute angle which the forward oriented straight line makes with the middle oriented straight line and the relative acute angle which the backward oriented straight line makes with the middle oriented straight line are each arctan(0.5).
[0063] As an illustrative example, an example cepstral peak tracking will now be described with reference to FIG. 10, which uses a predefined 2D geometrical structure having three beaming orientations, namely, backward, middle and forward. Accordingly, the example cepstral peak tracking utilises a technique of multiple beaming orientations according to various example embodiments of the present invention. Each beaming orientation refers to a direction of a line cutting through the cepstrogram (more specifically, the segmented ROI) in the 2D space, as shown in FIG. 10. For example, a line in the quefrency (or lag) axis, which is perpendicular to the time axis, corresponds to a middle beaming orientation, and may thus be referred to as a middle oriented line (or a middle oriented straight line). A line at a slanted angle slanted backward in time corresponds to a backward beaming orientation, and may thus be referred to as a backward oriented line (or a backward oriented straight line). Similarly, a line at a slanted angle slanted forward in time corresponds to a forward beaming orientation, and may thus be referred to as a forward oriented line (or a forward oriented straight line). In various example embodiments, the slanted angle for both the forward and backward oriented lines is arctan(0.5). In this regard, various example embodiments found that setting the slanted angle for both the forward and backward oriented lines to be arctan(0.5) facilitates the detection of a plurality of aligned sets, each aligned set comprising a local maximum cepstral peak of a first cepstral peak track generated based on the backward oriented straight line, a local minimum cepstral peak of the second cepstral peak track generated based on the middle oriented straight line and a local maximum cepstral peak of the third cepstral peak track generated based on the forward oriented straight line that are substantially aligned in the time dimension, thus enhancing the detection of (e.g., robustly detecting) the plurality of aligned sets for detecting the IBIs from the cardiac signal.
[0064] The effectiveness of the 2D geometrical structure (e.g., comprising the above-mentioned three lines, namely, the backward, middle and forward oriented lines corresponding to the above-mentioned backward, middle and forward beaming orientations configured as described above) in improving IBI detection (or estimation) accuracy will now be explained. In particular, the 2D geometrical structure comprising the three lines are used for cepstral peak tracking in the segmented ROI of the 2D cepstrogram along the time dimension according to various example embodiments of the present invention.
[0065] Various example embodiments note that an IBI formed by two consecutive heart beats is typically manifested by a strip of cepstral peaks in the cepstrogram. In this regard, the cepstral peaks are the local maximum in the quefrency (or lag) dimension. By way of anexample only and without limitation, FIG. 10 illustrates three IBIs, each associated with a quefrency (or lag) value ‘a’, ‘b’, and ‘c’ . respectively. Various example embodiments note that each strip of cepstral peaks usually has a highest cepstral value (energy) at the center of the strip, as denoted by a small circle in FIG. 10. Typically, the lag values along the strip are of a same value particularly near the center, and the energy values (cepstral values) tapers down at both ends of the strip. According to various example embodiments, the length of the strip (duration in the time dimension) has the time length of the IBI, which is identical to the quefrency (or lag) value (in the quefrency (or lag) domain). For instance, the cepstral peak strip at lag=c has a duration of ‘c’ seconds as well.
[0066] Now for two consecutive IBIs, various example embodiments note that they are manifested by two cepstral peak strips next to each other in time, for instance, the two IBIs with lag value of ‘b’ and ‘c’ . At the time point of the transition between the two IBIs, various example embodiments further note that the cepstral (energy) value is usually lower than that at the centers of either one of the strips, and usually the lag value is between the two lag values (between ‘b’ and ‘c’ in this case).
[0067] As the angle between the middle beaming and forward beaming is arctan(0.5) according to various example embodiments of the present invention, which effectively make the middle beaming point to the transition of the two IBIs at the same time as the forward beaming points to the center point of the IBI with lag=c, as well as the same time the backward beaming points to the center point of IBI with lag=b. A technical advantage of the forward and backward beaming orientation is that the actual lag value is independent of the time of beaming origin. For example, if the actual IBI is with lag=a instead of lag=b in FIG. 10, the backward beaming will still point to the center of the correct peak strip.
[0068] As illustrated in FIG. 10, the point of intersection of the three lines at the bottom corresponds to lag = 0. The lag = 0 point serves as the geometric anchor of the predefined 2D geometrical structure (or the three-line structure) comprising the middle oriented straight line (or the central (vertical) line) and the backward and forward oriented straight lines (or the two slanted lines) slanted at ±arctan(0.5) all of which originate from this geometric anchor and extend upward across the lag axis to cover the target lag range. Although lag = 0 is used solely as the geometric anchor, the cepstral value at lag = 0 itself is not required to be utilized — only the cepstral values along the three lines within the target lag range (e.g., 0.5 to 1.5 seconds in FIG. 5) are utilized. The geometrical structure is scanned across the time dimension by sliding this anchor point (i.e., lag = 0) sample by sample along the time axis of the cepstrogram. Togenerate the plurality of cepstral peak tracks, at each anchor position in time across the full time dimension, the predefined 2D geometrical structure respectively intersect with the series cepstral peaks to generate the plurality of cepstral peak tracks respectively corresponding to where the plurality of lines respectively intersects with the series of cepstral peaks along the time dimension.
[0069] Accordingly, when performing cepstral peak tracking, at any time point in the time dimension, each of the three lines takes a slice of the segmented ROI of the 2D cepstrogram and the cepstral peak with local maximum is detected by the line (based on the above-mentioned intersection of the cepstral peak by the line) in the ID slice. This results in three cepstral peak tracks: middle beaming track (corresponding to the above-mentioned second cepstral peak track), backward beaming track (corresponding to the above-mentioned first cepstral peak track) and forward beaming track (corresponding to the above-mentioned third cepstral peak track). For each cepstral peak track, at any time point or index along the time dimension, there is a cepstral peak having associated therewith a quefrency (or lag) value and a cepstral energy value. The cepstral value at a given quefrency (time lag) represents the corresponding coefficient of the cepstrum, i.e., the output of the inverse Fourier transform of the log-magnitude spectrum: c[n] = IFFT(log|FFT(x[n])|), evaluated at the quefrency (or time lag) of interest. It may thus be referred to as an "energy" value in the sense that its magnitude reflects the strength of the periodic structure in the signal at that time lag: a larger cepstral magnitude at time lag M indicates a stronger repetition of the signal with period M, while a smaller magnitude indicates weaker or absent periodicity at that time lag. Accordingly, the term "energy" may be used to refer to the strength or prominence of the cepstral peak, rather than in the strict signalprocessing sense of squared amplitude. In various example embodiments, each cepstral peak track may have associated therewith two value curves, namely, a quefrency (or lag) value curve and an energy value (or cepstral value) curve.
[0070] FIG. 11A shows the segmented ROI of the 2D cepstrogram, FIG. 11B shows extracted cepstral value curves (or cepstral energy value curves) from using the above-mentioned 2D geometrical structure comprising the three lines for cepstral peak tracking in the segmented ROI, along with the detected peaks and troughs, and FIG. 11C shows extracted quefrency (or lag) value curves, along with detected cepstral peaks corresponding to IBIs. It can be seen from the energy curves in the middle subplot, when the middle beaming has a low value, the backward beaming and forward beaming also have a high value at about the same time. Accordingly, in various example embodiments, detecting the IBIs from the cardiac signalfurther comprises: generating a first cepstral value curve, a second cepstral value curve and a third cepstral value curve based on the first cepstral peak track (backward beaming track), the second cepstral peak track (middle beaming track) and the third cepstral peak track (forward beaming track), respectively.
[0071] In a subsequent step 326, the local maximum time points (or local maximum cepstral peaks) are detected in the energy curves of forward beaming and backward beaming tracks, and the local minimum time points (or local minimum cepstral peaks) are detected in the energy curve of middle beam. In various example embodiments, the segmented ROI may be subjected to a low-pass filter (e.g., a cutoff frequency of 1 Hz) prior to detecting the local maximum time points in the energy curves of forward beaming and backward beaming and detecting the local minimum time points in the energy curve of middle beaming.
[0072] In a subsequent step 330, a plurality of aligned triplets (or aligned sets) are determined or detected, each aligned triplet comprising a local maximum cepstral peak of the first cepstral peak track, a local minimum (or local trough) cepstral peak of the second cepstral peak track and a local maximum cepstral peak of the third cepstral peak track that are substantially aligned in the time dimension (based on their proximity in time). Accordingly, for each of the plurality of aligned triplets, the quefrency value associated with the local maximum cepstral peak of the first cepstral peak track of the aligned triplet corresponds to an IBI detected based on the backward oriented straight line, and the quefrency value associated with the local maximum cepstral peak of the third cepstral peak track of the aligned triplet corresponds to an IBI detected based on the forward oriented straight line. In various example embodiments, for each of the plurality of aligned triplets, the local maximum cepstral peak of the first cepstral peak track of the aligned triplet based on a corresponding local maximum cepstral value of the first cepstral value curve, the local minimum cepstral peak of the second cepstral peak track of the aligned triplet based on a corresponding local minimum cepstral value of the second cepstral value curve and the local maximum cepstral peak of the third cepstral peak track of the aligned triplet based on a corresponding local maximum of the third cepstral value curve. In this regard, the corresponding local maximum cepstral value of the first cepstral value curve, the corresponding local minimum cepstral value of the second cepstral value curve and the corresponding local maximum of the third cepstral value curve are substantially aligned in the time dimension (close proximity in the time dimension, e.g., as illustrated in FIG. 1 IB). In this regard, each aligned triplet corresponds to one successful detection of two consecutive IBIs.l ’ l
[0073] In a subsequent step 334, detecting of the IBIs from the cardiac signal further comprises: obtaining a first set of IBIs (e.g., see FIG. 12A) across the time dimension comprising a plurality of IBIs detected based on the backward oriented straight line; obtaining a second set of IBIs (e.g., see FIG. 12B) across the time dimension comprising a plurality of IBIs detected based on the forward oriented straight line; and fusing the first and second sets of IBIs to obtain an output set of IBIs across the time dimension. Accordingly, lag values of the two detected consecutive IBIs are indicated by the lag values of the corresponding backward beaming and forward beaming lag curves. FIGs. 12A and 12B illustrate the identified / detected IBIs as highlighted using horizonal white bars. In particular, FIG. 12A shows the detected IBIs from the backward beaming and FIG. 12B shows the detected IBIs from the forward beaming. The final IBI output may then be obtained based on a fusion of the two sets of detected IBIs.
[0074] In various example embodiments, the fusion of the two sets of detected IBIs may be performed as follows. In a clean scenario, the two sets of IBI outputs (from the forward and backward beaming, as shown in FIGs. 12A and 12B) are expected to be very similar — if not identical — except at the start and end of the sequence, where the two sets are offset from each other by one IBI due to the opposite scanning directions. For example, the very first IBI in FIG.12A is not present in FIG. 12B, the second IBI in FIG. 12A corresponds to the first IBI in FIG.12B, and similarly for the very last IBIs of the two sets of IBI outputs. The two sets may first be aligned by matching each IBI in one set with its temporally corresponding IBI in the other set, taking the start / end offset into account. After alignment, an "overlapping region" of mutually corresponding IBIs is identified, with one IBI at each end of the sequence belonging to only one of the two sets ("non-overlapping" IBIs). The final IBI output may then be produced as follows. For each pair of overlapping IBIs from the two sets, the final IBI may be an average of the two values. The non-overlapping IBIs at the start (from FIG. 12 A) and at the end (from FIG. 12B) are included directly in the final output, so that no IBI in the original sequence is dropped. The fusion also improves robustness against missed detections. For example, if an IBI is missed in one set due to noise or signal artifacts, it may still be present in the other set and can be recovered for the final output. Conversely, spurious detections that appear in only one set can be flagged or suppressed by the absence of a corresponding IBI in the other set.
[0075] Experimental results of the method of detecting (or estimating) IBIs from a cardiac signal according to various example embodiments of the present invention will now be discussed to demonstrate its improved effectiveness and accuracy.
[0076] The method is evaluated using experiment data collected in lab environment. The sensor mat based optical fiber sensors are used for collecting BCG, together with ECG recorded with time synchronized. The method is evaluated using data from 8 subjects with 5 minutes recording for each subject. For ECG, the R peaks are detected using the Pan-Tompkins algorithm and manually verified. The detected inter-beat intervals from BCG are compared with those computed from ECG based on synchronized time. IBI differences are computed, and mean absolute error is used as the accuracy measurement (in milliseconds). A mean beat-to-beat interval error is also computed as a percentage of the interval error over the true interval. The coverage of the BCG inter-beat interval detection is computed as the ratio of the total number of detected inter-beat intervals from BCG divided by the total number from ECG.
[0077] FIG. 13 shows the result of the IBI estimation from 1 minute BCG data. The mean absolute error is 2.86 ms, the mean beat-to-beat interval error is 0.34% and coverage of 92.5%.
[0078] The results of the method together with reported results of two existing methods (Briiser et al., “Robust inter-beat interval estimation in cardiac vibration signals”, Physiol Meas.2013 Feb;34(2): 123-38 and Antink et al., "Ballistocardiography Can Estimate Beat-to-Beat Heart Rate Accurately at Night in Patients After Vascular Intervention," in IEEE Journal of Biomedical and Health Informatics, vol. 24, no. 8, pp. 2230-2237, Aug. 2020,) are presented in Table 1 below.Table 1 - The performance of the proposed method benchmarked with the most established works Method IBI mean absolute Mean beat-to-beat Coverage (%) error (ms) interval error (%)Proposed 3.85 0.43 92.5 Briiser 5.82 0.61 84.53 Antink (healthy 12.67 1.22 73.24subjects)
[0079] Accordingly, various example embodiments provide a method for detecting / estimating inter-beat intervals from a cardiac signal, particularly a ballistocardiogram (BCG) signal, using 2D cepstrogram. In particular, the method advantageously utilises a predefined 2D geometrical structure for tracking cepstral peaks. In this regard, the geometrical features of two consecutive IBIs in the cepstrogram is defined by the predefined 2D geometrical structure and utilized to identify the correct cepstral peaks in the 2D time-quefrency domain. As discussed above, the method has been demonstrated to be robust and accurate to detect / estimate IBIs from a cardiac signal (e.g., a BCG signal).
[0080] While embodiments of the invention have been particularly shown and described with reference to specific embodiments, it should be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the scope of the invention as defined by the appended claims. The scope of the invention is thus indicated by the appended claims and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced.
Claims
CLAIMS1. A method of detecting inter-beat intervals from a cardiac signal using at least one processor, the method comprising:obtaining a cardiac signal of a subject, wherein the cardiac signal is in a time domain; transforming the cardiac signal into a two-dimensional (2D) cepstrogram comprising a 2D array of cepstral values in a 2D time-quefrency domain having a quefrency dimension and a time dimension;tracking a series of cepstral peaks in the 2D cepstrogram along the time dimension using a predefined 2D geometrical structure comprising a plurality of lines to respectively generate a plurality of cepstral peak tracks along the time dimension, wherein the predefined 2D geometrical structure defines a predefined geometrical relationship between consecutive interbeat intervals in the 2D cepstrogram; anddetecting inter-beat intervals from the cardiac signal based on the plurality of cepstral peak tracks.
2. The method according to claim 1, wherein said tracking the series of cepstral peaks in the 2D cepstrogram along the time dimension comprises scanning the predefined 2D geometrical structure across the time dimension for the plurality of lines to respectively intersect with the series of cepstral peaks along the time dimension to generate the plurality of cepstral peak tracks respectively corresponding to where the plurality of lines respectively intersects with the series of cepstral peaks along the time dimension.
3. The method according to claim 1 or 2, wherein said transforming the cardiac signal into the 2D cepstrogram comprises, for each sliding time window of the cardiac signal along the time dimension of the 2D cepstrogram, determining a column of cepstral values associated with the sliding time window of the cardiac signal based on a plurality of quefrency values along the quefrency dimension, wherein the column of cepstral values comprises a plurality of cepstral values associated with the plurality of quefrency values, respectively, and the plurality of cepstral values is determined based on the plurality of quefrency values, respectively.
4. The method according to claim 3, wherein for each sliding time window of the cardiac signal, each of the plurality of cepstral values of the column of cepstral values associated withthe sliding time window is determined based on a time window of the sliding time window, the time window having a length determined based on the quefrency value associated with the cepstral value being determined.
5. The method according to claim 4, wherein the length of the time window of the sliding time window determined based on the quefrency value associated with the cepstral value being determined is double the quefrency value associated with the cepstral value being determined.
6. The method according to any one of claims 3 to 5, further comprising segmenting a region of interest (RO I) of the 2D cepstrogram in relation to the series of cepstral peaks along the time dimension to obtain a segmented ROI of the 2D cepstrogram, wherein said tracking the series of cepstral peaks in the 2D cepstrogram is performed with respect to the segmented ROI of the 2D cepstrogram.
7. The method according to claim 6, wherein said segmenting the ROI of the 2D cepstrogram comprises:applying, for each of the plurality of quefrency values along the quefrency dimension of the 2D cepstrogram, a sliding smoothing time window to the 2D cepstrogram at the quefrency value and along the time dimension to obtain a smoothed 2D cepstrogram;extracting a quefrency value curve corresponding to the series of cepstral peaks along the time dimension from the smoothed 2D cepstrogram; andsegmenting the ROI of the 2D cepstrogram based on the extracted quefrency value curve.
8. The method according to claim 7, wherein for each of the plurality of quefrency values along the quefrency dimension of the 2D cepstrogram, the sliding smoothing time window applied to the 2D cepstrogram at the quefrency value and along the time dimension has a length equal to the quefrency value.
9. The method according to any one of claims 1 to 8, whereinthe plurality of lines of the predefined 2D geometrical structure comprises a backward oriented straight line, a middle oriented straight line and a forward oriented straight line arranged such that the middle oriented straight line is along the quefrency dimension, theforward oriented straight line is slanted forward in the time dimension and makes a relative acute angle with the middle oriented straight line and the backward straight light is slanted backward in the time dimension and makes a relative acute angle with the middle oriented straight line, andthe plurality of cepstral peak tracks comprising a first cepstral peak track, a second cepstral peak track and a third cepstral peak track generated based on the backward oriented straight line, the middle oriented straight line and the forward oriented straight line, respectively.
10. The method according to claim 9, wherein the relative acute angle which the forward oriented straight line makes with the middle oriented straight line and the relative acute angle which the backward oriented straight line makes with the middle oriented straight line are equal.
11. The method according to claim 10, wherein the relative acute angle which the forward oriented straight line makes with the middle oriented straight line and the relative acute angle which the backward oriented straight line makes with the middle oriented straight line are each arctan(0.5).
12. The method according to any one of claims 9 to 11, whereinsaid detecting the inter-beat intervals from the cardiac signal comprises:determining a plurality of aligned sets, each aligned set comprising a local maximum cepstral peak of the first cepstral peak track, a local minimum cepstral peak of the second cepstral peak track and a local maximum cepstral peak of the third cepstral peak track that are substantially aligned in the time dimension; andfor each of the plurality of aligned sets,the quefrency value associated with the local maximum cepstral peak of the first cepstral peak track of the aligned set corresponds to an inter-beat interval detected based on the backward oriented straight line, andthe quefrency value associated with the local maximum cepstral peak of the third cepstral peak track of the aligned set corresponds to an inter-beat interval detected based on the forward oriented straight line.
13. The method according to claim 12, wherein said detecting the inter-beat intervals from the cardiac signal further comprises:generating a first cepstral value curve, a second cepstral value curve and a third cepstral value curve based on the first cepstral peak track, the second cepstral peak track and the third cepstral peak track, respectively; anddetermining, for each of the plurality of aligned sets, the local maximum cepstral peak of the first cepstral peak track of the aligned set based on a corresponding local maximum cepstral value of the first cepstral value curve, the local minimum cepstral peak of the second cepstral peak track of the aligned set based on a corresponding local minimum cepstral value of the second cepstral value curve and the local maximum cepstral peak of the third cepstral peak track of the aligned set based on a corresponding local maximum of the third cepstral value curve, wherein the corresponding local maximum cepstral value of the first cepstral value curve, the corresponding local minimum cepstral value of the second cepstral value curve and the corresponding local maximum of the third cepstral value curve are substantially aligned in the time dimension.
14. The method according to claim 12 or 13, wherein said detecting the inter-beat intervals from the cardiac signal further comprises:obtaining a first set of inter-beat intervals across the time dimension comprising a plurality of inter-beat intervals detected based on the backward oriented straight line;obtaining a second set of inter-beat intervals across the time dimension comprising a plurality of inter-beat intervals detected based on the forward oriented straight line; and fusing the first and second sets of inter-beat intervals to obtain an output set of interbeat intervals across the time dimension.
15. A system for detecting inter-beat intervals from a cardiac signal, the system comprising:at least one memory; andat least one processor communicatively coupled to the at least one memory and configured to:obtain a cardiac signal of a subject, wherein the cardiac signal is in a time domain; transform the cardiac signal into a two-dimensional (2D) cepstrogram comprising a 2D array of cepstral values in a 2D time-quefrency domain having a quefrency dimension and a time dimension;track a series of cepstral peaks in the 2D cepstrogram along the time dimension using a predefined 2D geometrical structure comprising a plurality of lines to respectively generate a plurality of cepstral peak tracks along the time dimension, wherein the predefined 2D geometrical structure defines a predefined geometrical relationship between consecutive interbeat intervals in the 2D cepstrogram; anddetect inter-beat intervals from the cardiac signal based on the plurality of cepstral peak tracks.
16. The system according to claim 15, wherein said track the series of cepstral peaks in the 2D cepstrogram along the time dimension comprises scanning the predefined 2D geometrical structure across the time dimension for the plurality of lines to respectively intersect with the series of cepstral peaks along the time dimension to generate the plurality of cepstral peak tracks respectively corresponding to where the plurality of lines respectively intersects with the series of cepstral peaks along the time dimension.
17. The system according to claim 15 or 16, wherein said transform the cardiac signal into the 2D cepstrogram comprises, for each sliding time window of the cardiac signal along the time dimension of the 2D cepstrogram, determining a column of cepstral values associated with the sliding time window of the cardiac signal based on a plurality of quefrency values along the quefrency dimension, wherein the column of cepstral values comprises a plurality of cepstral values associated with the plurality of quefrency values, respectively, and the plurality of cepstral values is determined based on the plurality of quefrency values, respectively.
18. The system according to claim 17, wherein for each sliding time window of the cardiac signal, each of the plurality of cepstral values of the column of cepstral values associated with the sliding time window is determined based on a time window of the sliding time window, the time window having a length determined based on the quefrency value associated with the cepstral value being determined.
19. The system according to claim 18, wherein the length of the time window of the sliding time window determined based on the quefrency value associated with the cepstral value being determined is double the quefrency value associated with the cepstral value being determined.
20. The system according to any one of claims 17 to 19, wherein the at least one processor is further configured to segment a region of interest (ROI) of the 2D cepstrogram in relation to the series of cepstral peaks along the time dimension to obtain a segmented ROI of the 2D cepstrogram, wherein said track the series of cepstral peaks in the 2D cepstrogram is performed with respect to the segmented ROI of the 2D cepstrogram.
21. The system according to claim 20, wherein said segment the ROI of the 2D cepstrogram comprises:applying, for each of the plurality of quefrency values along the quefrency dimension of the 2D cepstrogram, a sliding smoothing time window to the 2D cepstrogram at the quefrency value and along the time dimension to obtain a smoothed 2D cepstrogram;extracting a quefrency value curve corresponding to the series of cepstral peaks along the time dimension from the smoothed 2D cepstrogram; andsegmenting the ROI of the 2D cepstrogram based on the extracted quefrency value curve.
22. The system according to claim 21 , wherein for each of the plurality of quefrency values along the quefrency dimension of the 2D cepstrogram, the sliding smoothing time window applied to the 2D cepstrogram at the quefrency value and along the time dimension has a length equal to the quefrency value.
23. The system according to any one of claims 15 to 22, whereinthe plurality of lines of the predefined 2D geometrical structure comprises a backward oriented straight line, a middle oriented straight line and a forward oriented straight line arranged such that the middle oriented straight line is along the quefrency dimension, the forward oriented straight line is slanted forward in the time dimension and makes a relative acute angle with the middle oriented straight line and the backward straight light is slanted backward in the time dimension and makes a relative acute angle with the middle oriented straight line, andthe plurality of cepstral peak tracks comprising a first cepstral peak track, a second cepstral peak track and a third cepstral peak track generated based on the backward oriented straight line, the middle oriented straight line and the forward oriented straight line, respectively.
24. The system according to claim 23, wherein the relative acute angle which the forward oriented straight line makes with the middle oriented straight line and the relative acute angle which the backward oriented straight line makes with the middle oriented straight line are equal.
25. The system according to claim 24, wherein the relative acute angle which the forward oriented straight line makes with the middle oriented straight line and the relative acute angle which the backward oriented straight line makes with the middle oriented straight line are each arctan(0.5).
26. The system according to any one of claims 23 to 25, whereinsaid detect the inter-beat intervals from the cardiac signal comprises:determining a plurality of aligned sets, each aligned set comprising a local maximum cepstral peak of the first cepstral peak track, a local minimum cepstral peak of the second cepstral peak track and a local maximum cepstral peak of the third cepstral peak track that are substantially aligned in the time dimension; andfor each of the plurality of aligned sets,the quefrency value associated with the local maximum cepstral peak of the first cepstral peak track of the aligned set corresponds to an inter-beat interval detected based on the backward oriented straight line, andthe quefrency value associated with the local maximum cepstral peak of the third cepstral peak track of the aligned set corresponds to an inter-beat interval detected based on the forward oriented straight line.
27. The system according to claim 26, wherein said detect the inter-beat intervals from the cardiac signal further comprises:generating a first cepstral value curve, a second cepstral value curve and a third cepstral value curve based on the first cepstral peak track, the second cepstral peak track and the third cepstral peak track, respectively; anddetermining, for each of the plurality of aligned sets, the local maximum cepstral peak of the first cepstral peak track of the aligned set based on a corresponding local maximum cepstral value of the first cepstral value curve, the local minimum cepstral peak of the second cepstral peak track of the aligned set based on a corresponding local minimum cepstral valueof the second cepstral value curve and the local maximum cepstral peak of the third cepstral peak track of the aligned set based on a corresponding local maximum of the third cepstral value curve, wherein the corresponding local maximum cepstral value of the first cepstral value curve, the corresponding local minimum cepstral value of the second cepstral value curve and the corresponding local maximum of the third cepstral value curve are substantially aligned in the time dimension.
28. The system according to claim 26 or 27, wherein said detect the inter-beat intervals from the cardiac signal further comprises:obtaining a first set of inter-beat intervals across the time dimension comprising a plurality of inter-beat intervals detected based on the backward oriented straight line;obtaining a second set of inter-beat intervals across the time dimension comprising a plurality of inter-beat intervals detected based on the forward oriented straight line; and fusing the first and second sets of inter-beat intervals to obtain an output set of interbeat intervals across the time dimension.
29. A computer program product, embodied in one or more non-transitory computer-readable storage mediums, comprising instructions executable by at least one processor to perform a method of detecting inter-beat intervals from a cardiac signal according to any one of claims 1 to 14.
