Description:F O R M 2
THE PATENTS ACT, 1970
(39 of 1970)
The patent Rule, 2003
COMPLETE SPECIFICATION
(See section 10 and rule 13)

AN AUTOMATED SIGNAL QUALITY ASSESSMENT SYSTEM AND METHOD FOR PPG SIGNALS

PARETO TREE PVT. LTD
C-9/120, Sector-8, Rohini, Delhi
110085, India

The following specification describes the invention and the manner in which is to be performed

FIELD OF THE INVENTION

[0001] The present disclosure relates to the field of biomedical signal processing, and more specifically, addresses an automated signal quality assessment system and method tailored for the Photoplethysmogram (PPG) signals. The invention is designed to assess the quality of PPG signals in real-time for biomedical applications. This encompasses the preprocessing, spectral analysis, morphological analysis, and classification of PPG signals, facilitating reliable and automated determination of signal quality. The system and method find application in various biomedical contexts, such as patient monitoring, healthcare devices, consumer wearable devices and medical diagnostics, where accurate assessment of PPG signal quality is crucial for obtaining reliable physiological information.

BACKGROUND OF THE INVENTION

[0002] Biomedical signal processing plays a pivotal role in modern healthcare, particularly in real-time monitoring and diagnostics. Photoplethysmogram (PPG) signals, obtained non-invasively, offer valuable insights into cardiovascular health and oxygen saturation. However, the reliability of these signals is contingent on various factors, including environmental noise, motion artifacts, and device-related issues. Consequently, there is a critical need for an automated signal quality assessment system tailored specifically for PPG signals.

[0003] Current methodologies for assessing PPG signal quality often involve manual inspection or rudimentary techniques, leading to subjective and time-consuming evaluations. Moreover, the growing ubiquity of wearable health devices and remote patient monitoring underscores the necessity for efficient ,automated and real-time quality assessment method- for PPG signals..

[0004] The task of determination of signal quality for PPG signals, obtained through sensors, is challenging yet pivotal for applications deriving physiological parameters. Current literature outlines methods largely categorized as statistical, frequency domain-based, and waveform-based measures. Statistical approaches, which mostly rely upon ICA or PCA methods to characterize the signal behavior, struggle with the factors highlighting similarity between good and bad quality signals, hindering effective discrimination. Frequency domain-based methods, which mostly rely upon -FFT or DST techniques, offer reliability but demand significant computational resources. Waveform-based methods employ template matching to achieve a reliable assessment , but involve managing multiple parameters concurrently. Overall, these methods collectively fall short in capturing subtle changes within PPG signals that influence signal quality determination. Moreover, being data-dependent, these methods necessitate parameter re-estimation when faced with new data, limiting adaptability and real-time applicability. A novel approach is warranted to overcome these deficiencies and effectively assess PPG signal quality by capturing nuanced variations.

[0005] In view of the foregoing discussion, it is portrayed that there is a need to have an automated signal quality assessment system and method for PPG signals in the biomedical signal processing domain. The present invention addresses these challenges by proposing an innovative Automated Signal Quality Assessment System and Method for PPG Signals. Leveraging advanced signal processing techniques, the system incorporates filtering, normalization, spectral analysis, and morphological analysis to autonomously evaluate the quality of real-time PPG signals. The method integrates a classification engine that systematically verifies signal characteristics, providing an objective and rapid assessment of signal quality. This invention is poised to significantly enhance the reliability and efficiency of PPG signal processing, with applications spanning wearable health devices, continuous monitoring systems, and various medical diagnostic tools. The automated nature of the proposed system alleviates the burden of manual inspection, paving the way for improved patient care, accurate health assessments, and streamlined biomedical signal processing in diverse healthcare settings.
SUMMARY OF THE INVENTION

[0006] The present disclosure seeks to provide an automated signal quality assessment system and method for PPG Signals in biomedical signal processing. The invention addresses the critical need for accurate extraction and assessment of signal quality in Photoplethysmogram (PPG) data, obtained through non-invasive sensors. Frequently, sensor-acquired data is plagued by noise, requiring filtering for subsequent technique development. The reliability of these techniques is contingent on the signal quality, defined by key parameters across different applications. These parameters include salient cardiac cycles, absence of low-frequency noise, and the unambiguous identification of peaks and valleys in the PPG signal. The innovation proposes a unique definition of a good Signal Quality Index (SQI) signal, encompassing criteria expressed across various publications but formalized in a distinctive manner. Building upon this definition, the invention introduces a novel method for deriving the quality of a PPG signal. The proposed technique, illustrated using 5-second segments of PPG data with both good and bad signal quality, systematically validates the uniqueness and efficacy of the defined SQI parameters. This invention sets forth a comprehensive and distinctive approach to objectively quantify and distinguish between good and bad quality PPG signals, thereby contributing to enhanced reliability and performance in biomedical signal processing applications.

[0007] In an embodiment, an automated signal quality assessment system for PPG Signals in biomedical signal processing is disclosed. The system includes a wearable sensor worn on a user's wrist configured to detect PPG signals.

[0008] The system further includes a signal processing unit connected to the wearable sensor to filter noise from the PPG signal, normalize filtered signal and highlight PPG signal characteristics using a notch filter and a bandpass filter.

[0009] The system further includes a central processing unit coupled to the signal processing unit having a 30–point Discrete Fourier Transform (DFT) template DFTtemp, ranging from 0 to 6 Hz, derived as an average of several DFT spectra from hand-annotated good SQI signal segments, used as an initial check for discriminating poor quality PPG segments.

[0010] The system further includes a DSP unit connected to the classification engine to calculate 30–point DFT spectra (0-6 Hz) and compute an overlap score between the DFT of a test segment and a spectral template DFTtemp, where the overlap score is determined by the number of points for which the DFT for the test segment is less or equal in magnitude to DFTtemp, while both spectra are aligned with the location of the respective dominant peaks, wherein segments with an overlap score exceeding 25 undergo further processing for extracting morphological parameters which implements identification of peaks and valleys in the filtered PPG signal with a distance parameter ensuring a minimum distance between successive peaks based on 50% of the peak periodicity derived from the dominant peak in the DFT. The DSP unit characterizes monotonicity in a systolic duration of a cardiac cycle by examining a slope and detecting local minima/maxima as zero crossings in the double differentiation of the signal. The DSP unit characterizes a diastolic duration, defined as the duration between a peak and a successive valley, examining the presence of local minima/maxima, with a maximum of 4 permissible points to deem the signal quality as GOOD.

[0011] The system further includes a calculation processing unit coupled to the DSP unit to examine heights of peaks for all cardiac cycles in the segment, computing the median of peak magnitudes, and ensuring each cardiac peak's magnitude lies within a 20% deviation of the median value.

[0012] The system further includes a label processing unit connected to the calculation processing unit to label a PPG segment as good quality if it satisfies criteria including unambiguous cardiac peaks and valleys, monotonous rise in systolic duration, and a limited number of local minima/maxima within the diastolic duration.

[0013] In an embodiment, an automated signal quality assessment method for PPG Signals in biomedical signal processing is disclosed. The method includes acquiring PPG signals using a wearable sensor on a user's wrist.

[0014] The method further includes filtering noise from the PPG signal using a notch filter and a bandpass filter thereby normalizing filtered signal and highlighting PPG signal characteristics using a signal processing unit.

[0015] The method further includes identifying poor-quality PPG segments by a 30–point Discrete Fourier Transform (DFT) template DFTtemp (0-6 Hz) derived as an average of several DFT spectra obtained from hand-annotated good signal segments using a central processing unit.

[0016] The method further includes calculating 30–point DFT spectra (0-6 Hz) and computing an overlap score between the DFT of a test segment and a spectral template DFTtemp, where the overlap score is determined by the number of points for which the DFT for the test segment is less or equal in magnitude to DFTtemp, aligned at dominant peaks.

[0017] The method further includes extracting morphological parameters and identification of peaks and valleys in the filtered PPG signal, with a distance parameter ensuring a minimum distance between successive peaks, based on 50% of the peak periodicity derived from the dominant peak in the DFT upon segmentation with an overlap score exceeding 25 undergo further processing.

[0018] The method further includes characterizing monotonicity in a systolic duration of a cardiac cycle by examining a slope and detecting local minima/maxima as zero crossings in the double differentiation of the signal.

[0019] The method further includes characterizing a diastolic duration, defined as the duration between a peak and a successive valley, examining the presence of local minima/maxima, with a maximum of 4 permissible points to deem the signal quality as GOOD.

[0020] The method further includes examining heights of peaks for all cardiac cycles in the segment, computing the median of peak magnitudes, and ensuring each cardiac peak's magnitude lies within a 20% deviation of the median value using a calculation processing unit.

[0021] The method further includes labeling a PPG segment as good quality if it satisfies criteria including unambiguous cardiac peaks and valleys, monotonous rise in systolic duration, and a limited number of local minima/maxima within the diastolic duration by employing a label processing unit.

OBJECTS OF THE INVENTION

[0022] An object of the present disclosure is to determine signal quality based on the morphology of the PPG signal using the parameters derived from their morphological structure.

[0023] Another object of the present disclosure is to assess the signal quality with small duration segments so that the technique can be implemented in real–time.

[0024] Another object of the present disclosure is to ensure good positivity and sensitivity of the technique while minimizing the false positives.

[0025] Another object of the present disclosure is to propose a 30-point DFT template, derived using the average spectral characteristics in the 0-6 Hz band, from a set of manually annotated good quality signals, to ensure a desired spectral behaviour.

[0026] Yet another object of the present invention is to deliver an expeditious and cost-effective method that computes morphological features from a low–pass PPG signal using signal processing methods.

To further clarify the advantages and features of the present disclosure, a more particular description of the invention will be rendered by reference to specific embodiments thereof, which are illustrated in the appended drawings. It is appreciated that these drawings depict only typical embodiments of the invention and are therefore not to be considered limiting of its scope. The invention will be described and explained with additional specificity and detail in the accompanying drawings.

BRIEF DESCRIPTION OF FIGURES

[0027] These and other features, aspects, and advantages of the present disclosure will be better understood when the following detailed description is read concerning the accompanying drawings in which like characters represent like parts throughout the drawings, wherein:

Figure 1 illustrates a block diagram of an automated signal quality assessment system for PPG signal in the biomedical signal processing domain in accordance with an embodiment of the present disclosure;
Figure 2 illustrates a flow chart of an automated signal quality assessment method for PPG signals in biomedical signal processing in accordance with an embodiment of the present disclosure;
Figure 3 illustrates graphs depicting a) raw PPG signal, b) filtered PPG signal, and c) the DFT template (DFT temp) in accordance with an embodiment of the present disclosure;
Figure 4 illustrates graphs depicting a) time vs signal intensity/magnitude, b) frequency vs spectral response magnitude, c) overlapping frequency vs spectral response magnitude in accordance with an embodiment of the present disclosure; and
Figure 5 illustrates graphs depicting a) identification of peaks and valleys, b) presence of local minima/maxima in this region for segments, c) the minima/maxima present in the diastolic region, and d) median value for good and bad quality signals in accordance with an embodiment of the present disclosure.

Further, skilled artisans will appreciate those elements in the drawings are illustrated for simplicity and may not have necessarily been drawn to scale. For example, the flow charts illustrate the method in terms of the most prominent steps involved to help to improve understanding of aspects of the present disclosure. Furthermore, in terms of the construction of the device, one or more components of the device may have been represented in the drawings by conventional symbols, and the drawings may show only those specific details that are pertinent to understanding the embodiments of the present disclosure so as not to obscure the drawings with details that will be readily apparent to those of ordinary skill in the art having the benefit of the description herein.

DETAILED DESCRIPTION:

[0028] To promote an understanding of the principles of the invention, reference will now be made to the embodiment illustrated in the drawings and specific language will be used to describe the same. It will nevertheless be understood that no limitation of the scope of the invention is thereby intended, such alterations and further modifications in the illustrated system, and such further applications of the principles of the invention as illustrated therein being contemplated as would normally occur to one skilled in the art to which the invention relates.

[0029] It will be understood by those skilled in the art that the foregoing general description and the following detailed description are exemplary and explanatory of the invention and are not intended to be restrictive thereof.

[0030] Reference throughout this specification to “an aspect”, “another aspect” or similar language means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present disclosure. Thus, appearances of the phrase “in an embodiment”, “in another embodiment” and similar language throughout this specification may, but do not necessarily, all refer to the same embodiment.

[0031] The terms ``comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process or method that comprises a list of steps does not include only those steps but may include other steps not expressly listed or inherent to such process or method. Similarly, one or more devices or subsystems or elements or structures or components proceeded by "comprises...a" does not, without more constraints, preclude the existence of other devices or other subsystems or other elements or other structures or other components or additional devices or additional sub-systems or additional elements or additional structures or additional components.

[0032] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skills in the art to which this invention belongs. The system, methods, and examples provided herein are illustrative only and not intended to be limiting.

[0033] Embodiments of the present disclosure will be described below in detail concerning the accompanying drawings.

[0034] Referring to Figure 1, a block diagram of an automated signal quality assessment system for PPG Signals in biomedical signal processing is illustrated in accordance with an embodiment of the present disclosure. The system 100 includes a wearable sensor (102) worn on a user's wrist configured to acquire PPG signals.

[0035] In an embodiment, a signal processing unit (104) is connected to the wearable sensor (102) to filter noise from the PPG signal, normalize filtered signal and highlight PPG signal characteristics using a notch filter and a bandpass filter.

[0036] In an embodiment, a central processing unit (106) is coupled to the signal processing unit (104) having a 30–point Discrete Fourier Transform (DFT) template (0-6 Hz) derived as an average of several DFT spectra from hand-annotated good signal segments, used as an initial check for identifying poor quality PPG segments.

[0037] In an embodiment, a DSP unit (110) is connected to the classification engine (108) to calculate 30–point DFT spectra (0-6 Hz) and compute an overlap score between the DFT of a test segment and a spectral template DFTtemp, where the overlap score is determined by the number of points for which the DFT for the test segment is less or equal in magnitude to DFTtemp, aligned at dominant peaks, wherein segments with an overlap score exceeding 25 undergo further processing for extracting morphological parameters and identification of peaks and valleys in the filtered PPG signal with a distance parameter ensuring a minimum distance between successive peaks based on 50% of the peak periodicity derived from the dominant peak in the DFT, the DSP unit (110) characterizes monotonicity in a systolic duration of a cardiac cycle by examining a slope and detecting local minima/maxima as zero crossings in the double differentiation of the signal, the DSP unit (110) characterizes a diastolic duration, defined as the duration between a peak and a successive valley, examining the presence of local minima/maxima, with a maximum of 4 permissible points to deem the signal quality as GOOD.

[0038] In an embodiment, a calculation processing unit (112) is coupled to the DSP unit (110) to examine heights of peaks for all cardiac cycles in the segment, computing the median of peak magnitudes, and ensuring each cardiac peak's magnitude lies within a 20% deviation of the median value.

[0039] In an embodiment, a label processing unit (114) is connected to the calculation processing unit (112) to label a PPG segment as good quality if it satisfies criteria including unambiguous cardiac peaks and valleys, monotonous rise in systolic duration, and a limited number of local minima/maxima within the diastolic duration.

[0040] In another embodiment, the notch filter removes power-line interference in the 50 Hz range, and the bandpass filter operates in the range of 2-5 Hz, and the resulting filtered PPG signal is normalized within the range of -1 to 1.

[0041] In another embodiment, a server unit (116) is used to provide reliable signal quality assessment for outputting signal quality assessment results for any duration of a PPG signal and outputting signal quality assessment with a resolution of 1 cardiac cycle or pulse.

[0042] In another embodiment, utilizes a threshold over the number of peaks in the slope of the signal to verify characteristic behaviour, including a monotonicity factor in the systolic region, the number of peaks in the diastolic region, and uniformity in the energy of the signal at peak locations.

[0043] In another embodiment, a verification unit containing a set of parameters defining a good signal quality across different applications, wherein the parameters are selected from a salient representation of cardiac cycles in the signal, facilitating easy and straightforward derivation of the cardiac frequency, absence of low-frequency noise components in the signal segment to ensure a uniform energy profile for all cardiac cycles, no high-frequency fluctuations in the signal at any point, preserving the intact form factor of the Photoplethysmography (PPG) signal across the cardiac cycle, unambiguous identification of peaks corresponding to the systolic event in the cardiac cycle, unambiguous identification of valleys corresponding to the diastolic event in the cardiac cycle, monotonic rise in the segment of the signal between a valley and the following peak, without exhibiting local maxima/minima within its duration, the segment of the signal from a peak till the duration of the cardiac cycle not exhibiting more than 2 peaks, representing the cardiac activity during the diastolic event, and optional presence of a dicrotic notch within this segment, with its presence or absence not affecting the signal quality.

[0044] In another embodiment, a morphological analysis is performed by identifying distinct and salient peaks, identifying distinct and salient valleys, identifying monotonous behaviour between troughs and the following peaks, and identifying a maximum of 2 local peaks within each peak and the following trough.

[0045] In another embodiment, the classification engine (108) classifies signal quality upon verifying the post-morphological analysis when the signal between a trough and the following peak rises monotonically, the signal between a peak and the following trough exhibits a maximum of 2 local peaks, and the energy of the signal at each peak location is within 25% of the median values of the signal magnitude at peak locations.

[0046] Figure 2 illustrates a flow chart of an automated signal quality assessment method for PPG Signals in biomedical signal processing in accordance with an embodiment of the present disclosure. At step 202, method 200 includes detecting PPG signals by wearing a wearable sensor on a user's wrist.

[0047] At step 204, method 200 includes filtering noise from the PPG signal using a notch filter and a bandpass filter thereby normalizing the filtered signal to highlight the PPG signal characteristics using a signal processing unit.

[0048] At step 206, method 200 includes calculating 30–point (0-6 Hz) DFT spectra and computing an overlap score between the DFT of a test segment and a spectral template DFTtemp, derived as an average of several DFT spectra from hand-annotated good signal segments using a central processing unit, where the overlap score is determined by the number of points for which the DFT for the test segment is less or equal in magnitude to DFTtemp, aligned at dominant peaks.

[0049] At step 208, method 200 includes extracting morphological parameters and identification of peaks and valleys in the filtered PPG signal with a distance parameter ensuring a minimum distance between successive peaks based on 50% of the peak periodicity derived from the dominant peak in the DFT upon segmentation with an overlap score exceeding 25 undergo further processing.

[0050] At step 210, method 200 includes characterizing monotonicity in a systolic duration of a cardiac cycle by examining a slope and detecting local minima/maxima as zero crossings in the double differentiation of the signal.

[0051] At step 212, method 200 includes characterizing a diastolic duration, defined as the duration between a peak and a successive valley, examining the presence of local minima/maxima, with a maximum of 4 permissible points to deem the signal quality as GOOD.

[0052] At step 214, method 200 includes examining heights of peaks for all cardiac cycles in the segment, computing the median of peak magnitudes, and ensuring each cardiac peak's magnitude lies within a 20% deviation of the median value using a calculation processing unit.

[0053] At step 216, method 200 includes labelling a PPG segment as good quality if it satisfies criteria including unambiguous cardiac peaks and valleys, monotonous rise in systolic duration, and a limited number of local minima/maxima within the diastolic duration by employing a label processing unit.

[0054] In another embodiment, the method further comprises calculating a signal spectral DFT matching score. Then, dividing signal spectral DFT distinction based on a DFT matching threshold. Then, detecting key fiducial points, including peaks, troughs, and max slope. Thereafter, performing morphological analysis by identifying distinct and salient peaks, identifying distinct and salient valleys, identifying monotonous behaviour between troughs and the following peaks, and identifying a maximum of 2 local peaks within each trough and the following peak.

[0055] In another embodiment, the DFT matching threshold is dynamically adjustable based on signal characteristics, wherein the classification engine further considers the temporal relationships between peaks and troughs for enhanced signal quality assessment, wherein the morphological analysis includes waveform characterization and differentiation to enhance the accuracy of peak and trough identification.

[0056] Figure 3 illustrates graphs depicting a) raw PPG signal, b) filtered PPG signal, and c) DFT template (DFTtemp) in accordance with an embodiment of the present disclosure. The invention aims at assessment of signal quality in PPG, acquired using non-invasive sensors. Quite often, the data collected using sensors is noisy, and is filtered or otherwise processed for further use. These filtered PPG signals are then further utilised for technique development. For a reliable performance of these techniques, a good signal quality is vital. A good signal quality ensures that the PPG follows a desirable structure and hence yields good results with different techniques/studies.

[0057] Across different applications, following parameters deem necessary to define a good signal quality. The cardiac cycles must be salient in the signal, it should be easy and straightforward to derive the cardiac frequency. There must not be low frequency noise components in the segment. This ensures that all the cardiac cycles in the signal maintain a uniform energy profile. The signal must not exhibit high frequency fluctuations at any point of time. This ensures that the form factor of PPG across the cardiac cycle is intact. The peaks corresponding to the systolic event in the cardiac cycle can be unambiguously identified. The valleys corresponding to the diastolic event in the cardiac cycle can also be unambiguously identified. The segment of a signal between a valley and the following peak must rise monotonically and must not exhibit a local maxima/minimum within its duration. The segment of signal from a peak till the duration of cardiac cycle must not exhibit more than 2 peaks. These peaks represent cardiac activity during the diastolic event. The presence of a dicrotic notch within this segment is optional, and hence its presence or absence does not affect the signal quality.

[0058] This definition of a good SQI signal is expressed across several publications but has not yet been invented in this manner. This proposed definition is unique as depicted in the invention. Based on this understanding of a good quality signal, a method is proposed to derive the quality of a PPG signal. Following are the illustrated steps of the proposed technique, for small segments of duration 5 seconds of the PPG signal. Two 5 second segments of PPG, with signal quality as good and bad, are used to illustrate successive steps of the technique.

[0059] STEP 1 : The raw PPG signal is acquired using a wearable sensor, tied to the wrist. Depicted in Figure 3a.

[0060] STEP 2 : The raw PPG exhibits significant interference with noise, and hence needs to be filtered appropriately to highlight the PPG signal characteristics. The filtering operation is performed using a notch filter and a bandpass filter. The notch filter removes the power–line interference in the range of 50 Hz. The bandpass filter then filters the signal in the range ( 2-5 Hz ) to result in a filtered PPG signal. The filtered PPG is then normalised between -1 and 1. Depicted in Figure 3b.

[0061] A 30–point DFT template (DFTtemp) (0-6 Hz), obtained as an average of several DFT spectra over hand–annotated good segments, is used to assist for signal quality assessment. The DFT template is used as an initial check for identification of bad quality PPG segments. Depicted in Figure 3c.

[0062] Figure 4 illustrates graphs depicting a) time vs signal intensity/magnitude, b) frequency vs spectral response magnitude, c) overlapping frequency vs DFTtemp and spectral response magnitude in accordance with an embodiment of the present disclosure. For further illustration, two different segments of filtered PPG are used, one deemed as ‘GOOD’ quality, and another deemed as ‘BAD’ quality. The signal quality is categorised based on the visual inspection while following the guidelines given above as depicted in Figure 4a.

[0063] A 30–point DFT spectra (0-6 Hz) is obtained for the signals for processing. Depicted in Figure 4b.

[0064] An overlap score of the DFT of the test segment is obtained with respect to the spectral template DFTtemp. An overlap score is computed as the number of points for which the DFT for the test segment is less or equal in magnitude of DFTtemp, when the dominant peaks of the test segment DFT and DFTtemp are aligned. If the overlap score exceeds 25, the segments are further processed for extracting morphological parameters. Depicted in Figure 4c.

[0065] Figure 5 illustrates graphs depicting a) identification of peaks and valleys, b) presence of local minima/maxima in this region for segments, c) the minima/maxima present in the diastolic region, and d) median value in accordance with an embodiment of the present disclosure.

[0066] The filtered PPG signal is identified for the peaks and valleys in the signal. The peaks are derived with a distance parameter, ensuring a minimum distance between successive peaks in the segment. A distance criterion of 50% of the peak periodicity of the segment (derived from the dominant peak in the DFT) is used.

[0067] The valleys in the segments are obtained as points of global minima within successive peak locations. Figure 5a illustrates the identification of peaks and valleys in the segments.

[0068] The next step is to characterise the monotonicity in the systolic duration of the cardiac cycle. A systolic duration is understood as the duration within a valley and the successive peak in PPG. This monotonicity is characterised by examining the slope within this duration. If there are local minima/maxima present within this duration, it is captured as a zero crossing in the double differentiation of the signal in this duration. Figure 5b illustrates the presence of local minima/maxima in this region for segments. It can be seen in the figure that the systolic duration for the first segment (x1) exhibits local systolic peaks, and hence the signal is characterized as a BAD quality signal.

[0069] The next step in the technique characterises the diastolic duration in the PPG segment. A diastolic duration is defined as the duration between a peak and a successive valley. The diastolic duration is examined for the presence of local minima/maxima. A maximum of 4 such points are permissible to deem the signal quality as GOOD. Figure 5c illustrates the minima/maxima present in the diastolic region.

[0070] In the next step, the technique examines the heights of peaks for all cardiac cycles in the segment. To ensure that all the peaks across successive cycles in PPG follow a uniform magnitude, a median of the peak magnitudes is computed, and each cardiac peak is weighed against this value. For a good quality signal, the magnitude of each cardiac peak is expected to lie within 20% deviation of the median value as depicted in Figure 5d.

[0071] For a segment to be labelled as a GOOD quality, it needs to exhibit unambiguous cardiac peaks and valleys. It must also exhibit a monotonous rise in the systolic duration, and the number of local minima/maxima within the diastolic duration must not exceed 4. Furthermore, the magnitude for all peaks in the signal must stay within a 20% margin of the median value of magnitudes. If a PPG segment satisfies all these criteria, it can be labelled as a GOOD quality signal.

[0072] The invention aims at identification of good signal quality in segments of PPG signal. The signal quality is defined on the morphology of the PPG. All applications pertaining to the detection of heart rate, respiratory rate, SpO2, Blood Pressure, Cardiac Output, Arterial Stiffness, Hydration, Stress etc. from the PPG signals require a good quality signal for processing. A compromise on signal quality results in poor accuracy and hence these applications need a continuous assessment of the signal quality. Furthermore, ensuring a good signal quality is also a requirement for PPG acquisition systems, and their performance is directly associated with the quality of signal they capture. The invention gives a new definition of the signal quality in PPG. The proposed technique is based on the morphological parameters derived from the signal, and can ensure a good form factor of PPG, as required by an application.

[0073] The novel features of our signal quality extraction method encompass several advancements in the assessment of PPG signals. First and foremost, the method uniquely determines signal quality by closely examining the morphology of the PPG signal. It introduces a novel approach by proposing a 30–point Discrete Fourier Transform (DFT) template, derived from the average spectral characteristics in the 0-6 Hz band, obtained from manually annotated good-quality signals. This serves as an initial gateway to filter the signal behavior according to a desired form factor. Remarkably, the method operates without the need for prior data to learn or tune parameters, eliminating dependencies on historical information. Similarly, it does not require prior data for deriving statistical parameters. The method employs signal processing-based techniques to compute morphological features from a low-pass filtered PPG signal, ensuring simplicity and low computational complexity. Leveraging a minimal set of parameters that reflect the morphological structure, the method guarantees a good form factor of the signal. Importantly, it can be implemented for short data streams and operates in real-time, marking a significant advancement in the efficiency and applicability of signal quality assessment for PPG signals.

[0074] The proposed technique in this invention presents several advantages over existing technologies for PPG signal quality assessment. Unlike many current methods, it operates independently of prior data or knowledge, eliminating the need for historical information to make statistical calculations. Distinctively, the technique doesn't rely on traditional features typically extracted from PPG signals; instead, it introduces form factor-based features that directly highlight the morphological behavior of the signal segment. Another notable advantage is its ability to efficiently process short-duration PPG data streams, around 3-5 seconds, making it conducive for real-time signal quality assessment when integrated with data-capturing sensors. Furthermore, the technique boasts low computational complexity and flexibility, avoiding reliance on specific packages. Its adaptable nature allows for seamless portability across multiple platforms, enhancing its versatility and usability in various applications.

[0075] The inventive features of the proposed technique distinguish it significantly from its predecessors in the field of PPG signal processing. Unlike previous approaches, this technique places the prime emphasis on utilizing PPG morphology as the key factor for deriving signal quality, a distinctive and innovative approach. The definition of PPG morphology involves parameters that are integral to the signal characteristics and can also easily be derived from the signal, enhancing the method's efficiency. A noteworthy feature is the extraction of these parameters, which doesn't mandate a specific technique and can be accomplished using any generic technique serving the intended purpose, contributing to its versatility. The technique exhibits low computational complexity and portability, making it adaptable to various platforms. Its real-time capabilities are a standout advantage, providing immediate signal quality assessment for potential integration with data-capturing sensors. Furthermore, the technique promises improved accuracy in measuring various physiological parameters from the signal, encompassing heart rate, blood pressure, respiratory rate, and cardiac output, thereby enhancing its utility in diverse healthcare applications.

[0076] Following are the features of the invention,

Morphological parameters derived from the PPG signal which have been proven adequate to determine a good signal quality, given by

Monotonicity factor in the systolic region, identified using the slope of the signal. The systolic region xsys, derived between a valley sV and the subsequent peak sP as xsys = x[sV : sP], is checked for monotonicity. The absolute slope dxs is derived for xsys, given by,
dxs = abs( xsys [t+1] - xsys [t] ) ; ? t ?? [sV : sP]
The peaks are derived in the differential of the slope (dxs). For a good signal quality, the segment xsys must not experience a change in monotonicity. This implies that the dxs must not observe more than one peak. This test ensures that the monotonous rise of PPG segment in the systolic region
Behaviour of the signal in the diastolic region, given by the peaks in this region. The diastolic region xdia, derived between a peak sP and the following valley sV as xdia = x[sP : sV], is checked for the number of peaks. The slope absolute dxd is hence derived for xdia, given by,
dxd = abs( xdia [t+1] - xdia [t] ) ; ? t ?? [sP : sv]
For each peak and valley pair, 2 peaks in dxd are observed, and hence for characterising a good quality signal, the number of peaks in the dxd must not exceed 5. This test ensures a good form factor of PPG segment in the diastolic region
The uniformity of cardiac cycle energy, given by the peak magnitude in each cardiac cycle. The peaks are identified in a PPG segment of duration 5 seconds, and the median value of their magnitude values (mXP) is computed as,
mXP = med{ x[sP] } ;
where sP ? peak locations in the PPG segment of 5 sec
To characterise a good signal quality, all of the x[sP] values must stay within ( mXP + 0.2 * mXP ). This test ensures that there are no sudden surge of energy within the segment
A spectral template (X?T), which is derived from averaging the DFT spectra of good quality segments. The good quality segments are identified by manual examination, and their 30–point DFT (X?) (0-6 Hz) is computed. The X?T is then derived by,

X?T = (1/N) * ? (X?);
where N is the total good quality segments

To characterise a good quality signal, the total points (?) for which X? > X?T must not exceed 5. This test ensures a desired spectral behaviour and form factor of the PPG segment.
The idea, which is implemented in the form of an technique is presently developed in the Python language, and derives the signal quality over small duration ( 5 s) of PPG
The technique accepts the raw PPG as input, in the form of a .csv file
The technique pre–processes the raw signal, using a band–pass filter, and then yields the filtered PPG output
The bandpass filter is implemented as a third order Butterworth filter with low and high cutoff frequencies as 1 Hz and 5 Hz
The filtered output is then processed through the proposed technique
As a first step, a 30–point DFT (X?) (0-6 Hz) is computed for the filtered signal. The DFT is then matched against the spectral template X?T which is stored in a file.
If the minimum number of points for which X? < X?T is 25, the PPG segment is then processed further, else it is labelled as a bad quality segment
If the segment passes through the previous step, a distance parameter is computed based on the dominant peak in X?. The dominant peak location PD is obtained as PD = argmax(|X?|). The distance coefficient (dmin) is then computed as
dmin = 5 * (fs / (2 * ( 1 + PD) ) )
The technique further derives parameters owing to the morphology from the 5 seconds PPG segment. The parameters are,
Peaks in the signal (sP): with dmin as a minimum distance criterion
Valleys in the signal (sV): the valleys are obtained as the point of minimum magnitude between each pair of peaks
The slope dxs for the segment in the systolic duration for each cardiac cycle. The systolic duration is identified between each valley location and the following peak location
The slope dxd for the segment in the diastolic duration for each cardiac cycle. The diastolic duration is identified between each peak location and the following valley location
The median value of peak magnitudes, mXP, is obtained from the PPG segment magnitudes at sP
To determine the signal quality measure as GOOD, the following criteria are ensured
? X? < X?T for a minimum of 25 points out of 30
? peaks during xsys < 2
? peaks during xdia < 6
? x[sP] < 1.2*mXP
If any of the parameters fail to meet the above criteria, the signal quality is labelled as BAD

[0077] The idea can be used by the people who are building applications over the PPG signal. The idea can be used by applications which assess body vitals using PPG. The idea can also benefit people working towards designing sensors to capture the PPG signal as it can integrate as a feedback mechanism.

[0078] Following are the points that highlight the differences between the current invention and other products,
¦ The idea uses morphological features to define the form factor of the PPG signal, these features define a good quality signal
¦ The idea extracts signal quality over small segments of PPG and can be used for assessment in real–time
¦ The idea exhibits low time and computational complexity
¦ The signal quality is assessed over data collected using a wearable PPG sensor, against manually annotated labels. The proposed technique performs with a 98% accuracy which is unheard of.

[0079] The method / technique uses low pass and band pass filtering which can be implemented using any suitable package available with a programming language. The method / technique implements peak identification in the segment of the signal using inbuilt Python package, this can also be derived using any other suitable package available with a programming language. The method / technique implements routine to derive the DFT coefficients for a segment of PPG, this can also be derived alternatively using any other suitable package. This can be implemented for a wider variety of signals including ECG, EEG, BCG, ICG, and Phonocardiography.

[0080] An information processing chip to automatically classify the quality of real-time streaming biomedical signal, comprising of:
a. Filtering of the input signal
b. Normalisation of the filtered signal
c. Signal spectral DFT calculation
d. Signal spectral DFT matching with template DFT
e. Signal spectral DFT matching score
f. Signal spectral DFT distinction based on DFT matching threshold
g. Post threshold DFT signals are used for further classification as below
h. Key Fiducial points are detected including peak, trough and max slope
i. Morphological analysis to analyse and verify
i. Distinct and salient peaks are identified
ii. Distinct and salient valleys are identified
iii. A monotonous behaviour between troughs and the following peaks are identified
iv. A maximum of 2 local peaks within each peak and the following trough are identified
j. A classification engine which classifies signal quality if the below are verified post morphological analysis
i. The signal between a trough and the following peak rises monotonically
ii. The signal between a peak and the following trough exhibits a maximum of 2 local peaks
iii. The energy of signal at each peak location is within 20% of the median values of the signal magnitude at peak locations
k. Outputting a signal quality assessment or either “Good” or “Bad / “Usable” or “Unusable”
l. The system that can be inbuilt in any biomedical signal acquisition device.
The system can be embedded in a wearable device such as a wrist watch or chest patch.
The system can be connected to the MCU of any acquisition device to provide reliable assessment of signal quality.
The system can directly be connected to a server as a standalone unit to provide reliable signal quality assessment to the server.
The system can produce signal quality assessment results for any duration of a signal.
The system can output signal quality assessment with a resolution of 1 cardiac cycle or pulse.
The system processes real-time streaming signals from any biomedical signal acquisition device.
The system improves assessment accuracy with every new signal inputted.
The system uses a 1Hz to 5Hz bandpass filter to filter the device.
The system uses a threshold over the number of peaks in the slope of the signal to verify characteristic behaviour.
m. monotonicity factor in the systolic region, identified using the slope of the signal. The systolic region xsys, derived between a valley sV and the next peak sP as xsys = x[sV : sP], is checked for monotonicity. The slope dxS is derived for xsys, given by,
dxs = xsys [t+1] - xsys [t] ; ? t
dxs must observe not more than 1 peak in the duration of xsys
n. number of peaks in the diastolic region, identified as peaks in the slope of the diastolic region. The diastolic region xdia, derived between a peak sP and the next valley sV as xdia = x[sP : sV], is checked for the number of peaks. The slope dxD is derived for xdia, given by,
dxD = xdia [t+1] - xdia [t] ; ? t
dxD must observe not more than 5 peaks in the duration of xdia
o. a uniformity in the energy of the signal at peak locations. The magnitude of the normalised signal at the peak locations is obtained as |x[sP]| ? sP . The median (mXP) of these values is obtained , and each |x[sP]| is then expected to follow |x[sP]| < 1.25 * mXP
The system uses a discrete fourier transform ( DFT ) computation.
The system uses a 30–point (0-6 Hz) DFT matching threshold of 25 matching points.

The low computational method to classify biomedical signals’ quality on-device, comprising of:
p. Filtering of the input signal
q. Normalisation of the filtered signal
r. Signal spectral DFT calculation
s. Signal spectral DFT matching with template DFT
t. Signal spectral DFT matching score
u. Signal spectral DFT distinction based on DFT matching threshold
v. Post threshold DFT signals are used for further classification as below
w. Key Fiducial points are detected including peak, trough and max slope
x. Morphological analysis to analyse and verify
i. There is a set of distinct peaks in the segment
ii. There is a set of distinct troughs in the segment
iii. There is monotonous behaviour between troughs and peaks
iv. There is a limited number of peaks between peaks and troughs
v. There is a uniformity in the magnitude of signal at cardiac peak locations.
[0081] The drawings and the forgoing description give examples of embodiments. Those skilled in the art will appreciate that one or more of the described elements may well be combined into a single functional element. Alternatively, certain elements may be split into multiple functional elements. Elements from one embodiment may be added to another embodiment. For example, orders of processes described herein may be changed and are not limited to the manner described herein. Moreover, the actions of any flow diagram need not be implemented in the order shown; nor do all of the acts necessarily need to be performed. Also, those acts that are not dependent on other acts may be performed in parallel with the other acts. The scope of embodiments is by no means limited by these specific examples. Numerous variations, whether explicitly given in the specification or not, such as differences in structure, dimension, and use of material, are possible. The scope of embodiments is at least as broad as given by the following claims.

[0082] Benefits, other advantages, and solutions to problems have been described above about specific embodiments. However, the benefits, advantages, solutions to problems, and any component(s) that may cause any benefit, advantage, or solution to occur or become more pronounced are not to be construed as a critical, required, or essential feature or component of any or all the claims.

SHWETA SEN
IN/PA No-3010
Dated- 15th April, 2024 PATENT AGENT FOR APPLICANT

, Claims:WE CLAIM:

1. An automated signal quality assessment system for PPG Signals, the system comprises:

a wearable sensor worn by a user configured to detect PPG signal;
a signal processing unit connected to the wearable sensor to filter noise from the PPG signal, normalize filtered signal and highlight PPG signal characteristics using a notch filter and a bandpass filter;
a central processing unit coupled to the signal processing unit having a 30–point Discrete Fourier Transform (DFT) template (0-6 Hz) derived as an average of several DFT spectra from hand-annotated good signal segments, used as an initial check for identifying poor quality PPG segments;

a DSP unit connected to the classification engine to calculate 30–point DFT spectra (0-6 Hz) and compute an overlap score between the DFT of a test segment and a spectral template DFTtemp, where the overlap score is determined by the number of points for which the DFT for the test segment is less or equal in magnitude to DFTtemp, aligned at dominant peaks;
wherein segments with an overlap score exceeding 25 undergo further processing for extracting morphological parameters and identification of peaks and valleys in the filtered PPG signal with a distance parameter ensuring a minimum distance between successive peaks based on 50% of the peak periodicity derived from the dominant peak in the DFT;
the DSP unit characterizes monotonicity in a systolic duration of a cardiac cycle by examining a slope and detecting local minima/maxima as zero crossings in the double differentiation of the signal;
the DSP unit characterizes a diastolic duration, defined as the duration between a peak and a successive valley, examining the presence of local minima/maxima, with a maximum of 4 permissible points to deem the signal quality as GOOD;
a calculation processing unit coupled to the DSP unit to examine heights of peaks for all cardiac cycles in the segment, computing the median of peak magnitudes, and ensuring each cardiac peak's magnitude lies within a 20% deviation of the median value; and
a label processing unit connected to the calculation processing unit to label a PPG segment as good quality if it satisfies criteria including unambiguous cardiac peaks and valleys, monotonous rise in systolic duration, and a limited number of local minima/maxima within the diastolic duration.

2. The system as claimed in claim 1, wherein the notch filter removes power-line interference in the 50 Hz range, and the bandpass filter operates in the range of 2-5 Hz, resulting in normalized filtered PPG signals within the range of -1 to 1.

3. The system as claimed in claim 1, further comprises a server unit to provide reliable signal quality assessment for outputting signal quality assessment results for any duration of a PPG signal and outputting signal quality assessment with a resolution of 1 cardiac cycle or pulse.

4. The system as claimed in claim 1, utilizes a threshold over the number of peaks in the slope of the signal to verify characteristic behavior, including a monotonicity factor in the systolic region, the number of peaks in the diastolic region, and uniformity in the energy of the signal at peak locations.

5. The system as claimed in claim 1, further comprises a verification unit containing a set of parameters defining a good signal quality across different applications, wherein the parameters are selected from a salient representation of cardiac cycles in the signal, facilitating easy and straightforward derivation of the cardiac frequency, absence of low-frequency noise components in the signal segment to ensure a uniform energy profile for all cardiac cycles, no high-frequency fluctuations in the signal at any point, preserving the intact form factor of the Photoplethysmography (PPG) signal across the cardiac cycle, unambiguous identification of peaks corresponding to the systolic event in the cardiac cycle, unambiguous identification of valleys corresponding to the diastolic event in the cardiac cycle, monotonic rise in the segment of the signal between a valley and the following peak, without exhibiting local maxima/minima within its duration, the segment of the signal from a peak till the duration of the cardiac cycle not exhibiting more than 2 peaks, representing the cardiac activity during the diastolic event, and optional presence of a dicrotic notch within this segment, with its presence or absence not affecting the signal quality.

6. The system as claimed in claim 1, wherein a morphological analysis is performed by identifying distinct and salient peaks, identifying distinct and salient valleys, identifying monotonous behaviour between troughs and the following peaks, and identifying a maximum of 2 local peaks within each peak and the following trough.

7. The system as claimed in claim 6, wherein the classification engine classifies signal quality upon verifying the post-morphological analysis when the signal between a trough and the following peak rises monotonically, the signal between a peak and the following trough exhibits a maximum of 2 local peaks, and the energy of the signal at each peak location is within 25% of the median values of the signal magnitude at peak locations.

8. An automated signal quality assessment method for PPG Signals in biomedical signal processing, the method comprises:

detecting PPG signal by wearing a wearable sensor on a user's wrist;
filtering noise from the PPG signal using a notch filter and a bandpass filter thereby normalizing filtered signal and highlighting PPG signal characteristics using a signal processing unit;
identifying poor-quality PPG segments by a 30–point Discrete Fourier Transform (DFT) template (0-6 Hz) derived as an average of several DFT spectra from hand-annotated good signal segments using a central processing unit;
calculating 30–point DFT spectra (0-6 Hz) and computing an overlap score between the DFT of a test segment and a spectral template DFTtemp, where the overlap score is determined by the number of points for which the DFT for the test segment is less or equal in magnitude to DFTtemp, aligned at dominant peaks;
extracting morphological parameters and identification of peaks and valleys in the filtered PPG signal with a distance parameter ensuring a minimum distance between successive peaks is based on 50% of the peak periodicity derived from the dominant peak in the DFT upon segmentation with an overlap score exceeding 25 undergo further processing;
characterizing monotonicity in a systolic duration of a cardiac cycle by examining a slope and detecting local minima/maxima as zero crossings in the double differentiation of the signal;
characterizing a diastolic duration, defined as the duration between a peak and a successive valley, examining the presence of local minima/maxima, with a maximum of 4 permissible points;
examining heights of peaks for all cardiac cycles in the segment, computing the median of peak magnitudes, and ensuring each cardiac peak's magnitude lies within a 20% deviation of the median value using a calculation processing unit; and
labeling a PPG segment as GOOD quality if it satisfies criteria including unambiguous cardiac peaks and valleys, monotonous rise in systolic duration, and a limited number of local minima/maxima within the diastolic duration by employing a label processing unit.

9. The method as claimed in claim 8, further comprises:
calculating a signal spectral DFT matching score;
dividing signal spectral DFT distinction based on a DFT matching threshold;
detecting key fiducial points, including peaks, troughs, and max slope; and
performing morphological analysis by identifying distinct and salient peaks, identifying distinct and salient valleys, identifying monotonous behavior between troughs and the following peaks, and identifying a maximum of 2 local peaks within each trough and the following peak.

10. The method as claimed in claim 9, wherein the template matching is performed for the signal DFT, wherein the classification engine further considers the temporal relationships between peaks and troughs for enhanced signal quality assessment, wherein the morphological analysis includes waveform characterization and differentiation to enhance the accuracy of peak and trough identification.

SHWETA SEN
IN/PA No-3010
Dated- 15th April, 2024 PATENT AGENT FOR APPLICANT