Abnormal heart rhythm detection method and system based on PPG waveform segmentation and rhythm discrimination

Abstract

The application discloses an abnormal heart rhythm detection method and system based on PPG waveform segmentation and rhythm discrimination, and relates to the technical field of heart rhythm abnormality discrimination. The method comprises the following steps: in the process of scatter plot image processing, the original PPG light signal is subjected to scatter plot image analysis, light intensity sequence data is obtained, and waveform segment splitting is performed; in the split waveform segment, interval peak detection is performed, each main wave peak point and the corresponding main wave timestamp are identified, a wave peak interval sequence is obtained, and it is judged whether to perform sequence interval processing, so as to obtain a rhythm interval sequence; a panoramic Lorenz scatter plot is generated based on the rhythm interval sequence, global pixel distribution feature analysis is performed, a pixel distribution feature vector is obtained, local classification matching or global classification matching is simultaneously performed, and based on the results of the local classification matching or the global classification matching, it is judged whether to output the final abnormal heart rhythm detection result, thereby solving the problem of high heart rhythm detection misjudgment rate.

Description

Technical Field

[0001] This invention relates to the field of cardiac arrhythmia detection technology, and in particular to an abnormal cardiac arrhythmia detection method and system based on PPG waveform segmentation and rhythm discrimination. Background Technology

[0002] Abnormal heart rhythm detection based on photoplethysmography (PPG) signals has become an important research direction in the fields of mobile health and wearable devices in recent years. Its core advantage lies in the fact that PPG signals can be conveniently acquired through devices such as smartphones and smart bracelets, making large-scale cardiac rhythm screening possible. Current mainstream technologies mostly focus on baseline heart rate calculation or attempt to indirectly infer arrhythmias by measuring and analyzing the morphological characteristics of individual pulse wave waveforms (such as peaks, troughs, and notches). However, the analytical capabilities and reliability of these methods are often limited when faced with noise interference, individual vascular differences, and more complex non-periodic arrhythmias.

[0003] In the field of clinical electrocardiogram (ECG) analysis, the Lorenz scatter plot transforms the continuous RR interval sequence extracted from long-term recorded electrocardiograms (ECGs) into a point set distribution graphic on a two-dimensional plane; the coordinates of each scatter point are determined by the two adjacent RR intervals. This transformation maps one-dimensional time series information to a two-dimensional geometric distribution pattern, thereby intuitively and holistically revealing the intrinsic laws and nonlinear characteristics of cardiac rhythm.

[0004] For example, the invention patent with announcement number CN111374657B discloses a method and apparatus for detecting arrhythmia based on photoplethysmography (PPG) signals, including: receiving a PPG image of a user's movement and its motion signal; segmenting the PPG image and its motion signal, as well as the PPG image and motion signal for the corresponding time period, at regular intervals, filtering motion artifact noise, and converting them into corresponding spectrograms; obtaining an estimated heart rate based on the spectrograms of the PPG image and its motion signal, as well as the shortest distance between two peaks and the peak-to-peak interval in the PPG signal segment, inputting them into a prediction model, and determining whether arrhythmia exists.

[0005] First, PPG signals are susceptible to motion artifacts, individual vascular characteristics, and acquisition environment interference, resulting in unstable original waveform morphology. Second, the pulse peak interval sequence extracted from PPG signals is essentially a one-dimensional time series. When characterizing complex heart rhythms (such as the inherent randomness of atrial fibrillation or the cyclical pattern of premature bigeminy), this sequence cannot fully preserve the dynamic correlation and overall structural information between heartbeats, resulting in limitations in the expression of the global and nonlinear characteristics of rhythm.

[0006] Furthermore, when using Lorenz scatter plots for automated heart rhythm classification, if two or more different types of arrhythmias exist simultaneously in the electrocardiogram rhythm, the generated scatter plot will have feature superposition and morphological mixing, resulting in a blurred graphic structure. This makes it difficult to accurately identify the individual components in the mixed rhythm using pixel comparison processing methods, leading to a high misjudgment rate of abnormal heart rhythms when performing fine identification of complex arrhythmias. Summary of the Invention

[0007] To address the high false positive rate in existing technologies when performing precise identification of complex arrhythmias, this invention provides an abnormal heart rhythm detection method and system based on PPG waveform segmentation and rhythm discrimination. The technical solution is as follows: On the one hand, an abnormal heart rhythm detection method based on PPG waveform segmentation and rhythm discrimination is provided. The method includes: S1, in the process of scatter plot image processing, scatter plot image analysis is performed on the original PPG light signal to obtain light intensity sequence data, and waveform segments are split; S2, within the split waveform segments, interval peak detection is performed to identify each main wave peak point and the corresponding main wave timestamp, obtain the peak time interval sequence, and determine whether to perform sequence interval processing to obtain the rhythm interval sequence; S3, based on the panoramic Lorenz scatter plot generated by the rhythm interval sequence, global pixel distribution feature analysis is performed to obtain pixel distribution feature vectors, and local classification matching or global classification matching is performed simultaneously; S4, based on the results of local classification matching or global classification matching, it is determined whether to output the final abnormal heart rhythm detection result.

[0008] On the other hand, an abnormal heart rhythm detection system based on PPG waveform segmentation and rhythm discrimination is provided. This system includes: a PPG waveform segmentation module, used to perform scatter image analysis on the original PPG optical signal during scatter plot image processing to obtain light intensity sequence data and segment the waveform; a rhythm interval sequence processing module, used to perform interval peak detection within the segmented waveform segments, identify each main wave peak point and its corresponding main wave timestamp, obtain the peak time interval sequence, and determine whether to perform sequence interval processing to obtain the rhythm interval sequence; an abnormal heart rhythm detection module, based on the panoramic Lorenz scatter plot generated by the rhythm interval sequence, performs global pixel distribution feature analysis to obtain pixel distribution feature vectors, and simultaneously performs local or global classification matching; and a classification matching output module, based on the results of local or global classification matching, determines whether to output the final abnormal heart rhythm detection result.

[0009] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: 1. This invention performs scatter plot analysis and waveform segmentation on the raw PPG optical signal, dividing the continuous PPG signal into multiple independent waveform analysis units. Within each segment, interval peak detection and timestamp extraction are performed to obtain a peak time interval sequence. Based on this, the peak change rate is introduced as a quantitative indicator to judge the stability of peak detection. When the peak change rate exceeds a preset range, the system automatically prompts for signal re-acquisition, thereby avoiding false detections caused by poor signal quality or motion artifacts. Compared to traditional methods that directly perform global analysis on the entire PPG signal, the segmented processing strategy adopted in this invention can more flexibly adapt to individual differences and signal fluctuations. Through independent processing after segmentation, abnormal interference can be identified and eliminated within a local range, ensuring the accuracy of subsequent rhythm analysis, thus laying a reliable data foundation for the detection of complex arrhythmias.

[0010] 2. After obtaining the peak time interval sequence, further sequence interval processing is performed. Abnormal intervals exceeding the qualified range are removed to construct the rhythm interval sequence. If data is missing in the sequence, interpolation is automatically performed according to the preset data missing-interpolation repair mapping relationship to ensure the integrity and continuity of the time series. This mechanism avoids the analysis discontinuity problem caused by signal interruption or unstable acquisition, making the subsequently generated Lorenz scatter plot more representative. Compared with the traditional method of directly using the original interval sequence and ignoring the impact of data missing, the sequence repair strategy of this invention can maintain the temporal coherence of rhythm analysis, improve the ability to capture arrhythmias, and provide high-quality input for subsequent global and local classification matching.

[0011] 3. This invention converts rhythm interval sequences into panoramic Lorenz scatter plots. By pairing adjacent intervals as two-dimensional coordinate points and visualizing them within a discretized pixel grid, one-dimensional time series information is mapped to a two-dimensional spatial distribution pattern. Based on this, global pixel distribution feature analysis is performed on the scatter plot, extracting dispersion and clustering metrics, and further calculating scatter accuracy indices to evaluate image quality and classification reliability. This analysis method, based on image representation and feature quantification, can reveal the nonlinear characteristics and inherent patterns of rhythms, overcoming the limitations of traditional methods that rely solely on numerical sequence analysis. By transforming rhythm information into image features, this invention provides richer and more discriminative input features for subsequent intelligent classification and pattern matching.

[0012] 4. In local classification matching, this invention divides the scatter plot into multiple feature analysis regions, calculates the similarity index of each region, and constructs a multi-region similarity parameter matrix. By introducing region weight coefficients, the similarity of each region is weighted and fused to select the most representative region as a classification candidate. If the matching results are inconsistent or the boundaries are blurred, the geometric boundaries of the regions are dynamically adjusted according to the pixel density, and the matching is performed again. This mechanism is suitable for scenarios with mixed or complex arrhythmias, and can effectively distinguish multiple overlapping rhythm patterns through independent analysis and comprehensive judgment of local features. Compared with the traditional approach of performing single matching on the entire image, the multi-region analysis strategy of this invention enhances the ability to analyze complex scatter plots, reduces the misjudgment rate caused by feature mixing, and improves practicality and reliability. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A flowchart of an abnormal heart rhythm detection method based on PPG waveform segmentation and rhythm discrimination provided in an embodiment of the present invention; Figure 2 This is a flowchart corresponding to the scatter image parsing provided in an embodiment of the present invention; Figure 3 A flowchart corresponding to the global classification matching provided in this embodiment of the invention; Figure 4 A flowchart corresponding to local classification matching provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the abnormal heart rhythm detection system based on PPG waveform segmentation and rhythm discrimination provided in an embodiment of the present invention; Figure 6 A table showing the results of abnormal heart rhythm simulation detection provided in an embodiment of the present invention. Detailed Implementation

[0015] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0016] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0017] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Embodiment 1 of the present invention provides an abnormal heart rhythm detection method based on PPG waveform segmentation and rhythm discrimination, such as... Figure 1 The flowchart shown is for an abnormal heart rhythm detection method based on PPG waveform segmentation and rhythm discrimination. The processing flow of this method can include the following steps: S1, in the scatter plot image processing, the original PPG optical signal is parsed to obtain light intensity sequence data, and waveform segments are split; S2, within the split waveform segments, interval peak detection is performed to identify each main wave peak point and its corresponding main wave timestamp, the peak time interval sequence is obtained, and it is determined whether to perform sequence interval processing to obtain the rhythm interval sequence; S3, based on the panoramic Lorenz scatter plot generated by the rhythm interval sequence, global pixel distribution feature analysis is performed to obtain pixel distribution feature vectors, and local classification matching or global classification matching is performed simultaneously; S4, based on the results of local classification matching or global classification matching, it is determined whether to output the final abnormal heart rhythm detection result.

[0020] When establishing the standard image library, firstly, representative scatter plot samples were collected and screened based on the Lorenz scatter plot characteristics of different heart rhythm types. After image normalization, binarization, and rotation alignment, the pixel distribution feature parameters of each image were extracted to construct a standard image library containing multiple heart rhythm types. Based on this, according to the global fan-shaped distribution characteristics of atrial fibrillation rhythms, the pixel density and coordinate features of their scatter plots were extracted and integrated separately to form the first standard vector library, specifically used for the identification of atrial fibrillation types.

[0021] Secondly, for sinus rhythm and other non-atrial fibrillation abnormal rhythms, based on the different distribution patterns of their scatter plots in the central and non-central regions, feature regions are further divided, and the pixel density and position parameters of the corresponding regions are extracted to construct a second standard vector library. This library centrally represents the regionalized features of various abnormal rhythms other than atrial fibrillation, and is used for further subdivision and discrimination of rhythm types after excluding atrial fibrillation.

[0022] like Figure 2 The diagram shows a flowchart of the scatter image analysis provided in this embodiment of the invention. The scatter image analysis process is as follows: A photoplethysmography (PPG) acquisition device (usually a wearable / portable terminal such as a smart bracelet, watch, or dedicated finger clip) acquires the raw PPG light signal. The raw PPG light signal output by the acquisition device (usually a binary data stream or file) is decoded according to the data format specification and communication protocol, restoring it to a discrete light intensity sampling value sequence with high-precision timestamps. This yields the raw light intensity sampling values ​​arranged in chronological order, and the acquisition time corresponding to each sampling point is recorded. These are then summarized to form a light intensity sequence with chronological order, used to quantify the heart rhythm state of the heartbeat.

[0023] The light intensity sequence is divided into a preset number of waveform analysis segments according to equal time lengths. After differential filtering, PPG waveform sequence segments are obtained. The segment is cut by setting a fixed segment duration (e.g., 5 seconds or 10 seconds to meet the main wave of the heart rhythm characteristics). The preset number of segments = total acquisition time (seconds) / preset segment duration (seconds). The PPG waveform sequence segment represents an independent PPG waveform unit that has been preprocessed and can be used for feature extraction. The specific process of differential filtering is as follows: first, the original waveform analysis segment is bandpass filtered to remove baseline drift and some high-frequency noise. Then, the filtered signal is differentially processed to further sharpen the waveform. The final signal is the PPG waveform sequence segment.

[0024] The peak detection algorithm searches within PPG waveform sequence segments to obtain the main peak point and corresponding main peak timestamp of each segment. For example, the Automatic Multiscale-based Peak Detection (AMPD) algorithm compares the peak points on both sides using a multi-scale sliding window to find the local maximum value and determine the current peak point and its timestamp. It further calculates the time difference between adjacent main peak timestamps; the difference between the time corresponding to the next main peak timestamp and the time corresponding to the current main peak timestamp is used as the peak time interval of the waveform segment. All peak time intervals are arranged in ascending order to obtain a peak time interval sequence, used to quantify the time interval changes between heartbeats. The standard deviation of the peak time interval sequence is calculated to obtain the peak change rate, used to quantify the discrete fluctuation of the PPG waveform sequence segment. When the peak change rate is greater than a preset peak change rate, it indicates that the current peak detection is unstable, and a prompt is made to re-acquire the original PPG optical signal; otherwise, sequence interval processing is performed. The preset peak change rate is represented by the summation and averaging of historical peak change rates.

[0025] The specific process of sequence interval processing is as follows: peak time intervals that exceed the preset qualified interval in the peak time interval sequence are removed, and the peak time interval sequence after peak time interval removal is used as the rhythm interval sequence and sequence integrity is checked; the preset qualified interval is the intersection of the qualified interval corresponding to the physiological limit of heartbeat and the statistical characteristic range of the current peak time interval sequence. The first quartile and the third quartile of the current peak time interval sequence are obtained, and the interquartile range (i.e., the difference between the third quartile and the first quartile) is used as the statistical characteristic range.

[0026] The algorithm checks whether adjacent valid data points (i.e., peak time intervals) in the interrhythm interval sequence are continuous in the time dimension. If a missing segment is detected in the interrhythm interval sequence, the corresponding missing data length is input into the preset missing data-interpolation repair mapping relationship to obtain the missing data filling value to fill the missing segment. Based on the filled interrhythm interval sequence, global pixel distribution feature analysis is performed. If no missing segment is detected in the interrhythm interval sequence, global pixel distribution feature analysis is performed directly.

[0027] The pre-defined data missing-interpolation repair mapping relationship is based on the classification of the key parameter of data missing length and the pre-defined corresponding interpolation strategy logic. Using a verified accurate rhythm interval sequence as the baseline sequence, test sequences with known missing segments are generated on this baseline sequence using a pre-defined missing length gradient (simulating cases with 1, 2, and more than 3 missing segments respectively). For each missing length, a pre-defined candidate repair strategy (including linear interpolation and arithmetic mean based on the effective intervals before and after the missing position) is applied for repair. The repaired test sequences are evaluated against the baseline sequence to select the optimal specific strategy for each missing length. For cases with only a single missing segment (data missing length = 1), the mapping is to use the arithmetic mean of the verified intervals before and after the missing segment or a linear interpolation strategy. For short segments with two consecutive missing segments (data missing length = 2), the mapping is to apply a linear interpolation strategy. For long segments with three or more consecutive missing segments (data missing length ≥ 3), the sequence is marked to ensure a reliable balance between repair continuity and preventing erroneous interpolation.

[0028] like Figure 3 The diagram shows a flowchart of the global classification matching process provided in this embodiment of the invention. The global pixel distribution feature analysis process is as follows: within the interrhythm sequence, the current peak time interval R is... n The time interval R between the next peak n-1 Pairing is performed to form (R) n R n+1 The interval coordinates are then filled into a two-dimensional discretized pixel grid corresponding to the preset coordinate range to form a standard coordinate system. The specific process for determining the preset coordinate range is as follows: First, the statistical percentiles of the current rhythm interval sequence are calculated (e.g., the 2.5 percentile (P2.5) and the 97.5 percentile (P97.5)). The final preset coordinate range is usually dynamically set to [P2.5-M, P97.5+M], where M is a safety margin calculated based on the data dispersion (e.g., interquartile range). The physical meaning of this coordinate range is the continuous RR interval value (unit: milliseconds). Its lower limit can be set to 250-300 milliseconds (corresponding to extreme heart rate), and its upper limit can be set to 1500-2000 milliseconds (corresponding to extremely slow heart rate or missed beats) to ensure that each generated image is comparable and focuses on valid data.

[0029] The pixel value at the grid position where the inter-coordinate pair is located is set to 1 and used as the foreground pixel. The pixel values ​​at other unoccupied grid positions are set to 0 and used as the background pixels. In the standard coordinate system, a two-dimensional view composed of 0s and 1s, i.e., a panoramic Lorenz scatter plot presented in a visual form, is used as the scatter plot to be detected. Simultaneously, feature extraction calculations are performed on the scatter plot to be detected using image recognition algorithms to obtain the pixel distribution feature vector, including pixel density, spatial distribution moment, and axial projection. The pixel density is the ratio of the total number of foreground pixels with a value of 1 to the total number of pixels in the image, used to quantify the overall density of the scatter points; the spatial distribution moment includes... The first moment (centroid coordinates) and second central moment (variance and covariance) of all foreground pixel coordinates are included. The centroid describes the distribution coordinates of the scatter point set, while the variance and covariance describe the dispersion of the scatter points around the centroid and the correlation between the two. The axial projection is the projection of the two-dimensional view on the X-axis and the Y-axis, respectively, to obtain the X-axis projection distribution histogram and the Y-axis distribution projection histogram. The number of peaks, the width of the main peak (half-width at half maximum), and the distance between peaks are extracted from the histogram after projection for statistical analysis. This is used to numerically describe the distribution clustering pattern and dispersion of the scatter point plot in the horizontal (X-axis) and vertical (Y-axis) directions.

[0030] The dispersion measure and clustering measure are obtained after statistical analysis of the scatter plot to be detected. The dispersion measure is used as the numerator and the clustering measure is used as the denominator to obtain the scatter accuracy index, which reflects the graphic quality and classification reliability of the scatter plot to be detected. If the scatter accuracy index is greater than the preset scatter accuracy index, global classification matching is performed; otherwise, local classification matching is performed.

[0031] Because persistent atrial fibrillation has a cloud-like fan-shaped distribution (without a clear center of aggregation) and paroxysmal atrial fibrillation has a local fan-shaped distribution, this feature is more global than the scatter plots of other types of heart rhythms. Therefore, the pixel density and coordinate parameters of the image to be detected will first be input into the first standard vector library in the image library for comparison.

[0032] Furthermore, the global classification matching process is as follows: the dispersion measure and the clustering measure are compared with the first standard feature vector library one by one to obtain similarity groups, and the similarity of the similarity groups is summed and averaged to obtain the standard similarity; if the standard similarity exceeds the preset standard similarity, fine identification matching is performed, otherwise the data changes of the scatter plot to be detected are monitored and a new analysis cycle is awaited.

[0033] The dispersion measure quantifies the degree of dispersion of foreground pixels in a two-dimensional plane. Principal Component Analysis (PCA) is used to calculate the largest eigenvalue of the covariance matrix of all foreground pixel coordinates. The larger the value, the wider the data distribution in the main dispersion direction. The clustering measure quantifies whether foreground pixels are tightly clustered. Using a connected component labeling algorithm in image processing, all independent connected regions (i.e., point clusters) composed of adjacent foreground pixels are identified. For each connected region, its pixel density is calculated as the number of foreground pixels in the region divided by the area of ​​the smallest bounding rectangle of the region. The pixel densities of all connected regions are weighted and averaged (the weight is the number of pixels in each region). The resulting value is the overall clustering measure. The higher the value, the more the scattered points tend to form a tight and clear cluster structure.

[0034] Specifically, the fine-grained identification and matching process is as follows: taking the diagonal line corresponding to the origin of the scatter plot to be detected and its largest horizontal and vertical coordinate points as the central axis, a sliding analysis window with a fixed step size is set; the step size should not be less than the size of one pixel, otherwise a large number of overlapping windows will be generated, increasing unnecessary calculations; it should also not be too large, so as not to miss key local pattern changes; a typical setting is to take 1% to 5% of the total pixel length of the image in the sliding direction (i.e., the direction of the central axis) as the step size.

[0035] The variance of scattered pixels within the sliding analysis window is used as the divergence index of the sliding analysis window. For a given sliding analysis window, the two-dimensional coordinate set of all foreground pixels (i.e., pixels with a value of 1) within it is extracted in the image matrix. The covariance matrix of this coordinate set on the two-dimensional plane is calculated. The rows of the matrix are the variances of the X coordinates of the foreground pixels, and the columns are the variances of the Y coordinates of the foreground pixels, describing the degree of dispersion of all foreground pixels on the two-dimensional plane. The divergence index is defined as the total variance of this matrix, i.e., the trace of the covariance matrix, which quantifies the overall dispersion on the two-dimensional plane. The larger the index value, the more dispersed the scattered points are within the window; the smaller the value, the more concentrated the scattered points are.

[0036] If the divergence index exceeds the preset divergence index in each sliding analysis window, the corresponding scatter plot to be detected is determined to show the expected abnormal divergence pattern, and the abnormal heart rhythm detection result is output as follows: the scatter plot to be detected matches the scatter plot corresponding to persistent atrial fibrillation (without a clear aggregation center); if the divergence index does not exceed the preset divergence index in each sliding analysis window, the corresponding scatter plot to be detected is determined to not have a heart rhythm abnormality and does not trigger a heart rhythm abnormality alarm. In addition to the above two cases, the scatter plot to be detected is determined to show a paroxysmal pattern, and the abnormal heart rhythm detection result is output as follows: it matches the local fan-shaped distribution characteristics of paroxysmal atrial fibrillation. The preset divergence index is the result of summing and averaging the historical divergence indices in the historical fine identification and matching process.

[0037] like Figure 4 The diagram shows a flowchart of local classification matching provided in an embodiment of the present invention. The specific process is as follows: Based on preset feature region division rules, the scatter plot to be detected is divided into a preset number of feature analysis regions; the preset feature region division rules, based on the graphical and physiological mechanisms of the ECG scatter plot, divide the scatter plot... Figure 2 The 3D plane is transformed into a series of functional subspaces with specific diagnostic orientations.

[0038] The dispersion measure and clustering measure obtained from each feature analysis region are compared with the second standard feature vector library to calculate the region similarity. The region similarity index is calculated using the Structure Similarity Index Measure (SSIM). For a certain feature analysis region of the scatter plot to be detected and the corresponding region of a certain type of heart rhythm template in the standard feature vector library, the same multidimensional feature vector (dispersion measure and clustering measure) are extracted respectively.

[0039] Each feature analysis region of the image to be tested is compared with the corresponding region of a certain type of heart rhythm template in the standard library for similarity calculation. All results are organized into a matrix, which is the multi-region similarity parameter matrix. The rows represent each feature analysis region, and the columns represent the region similarity index. Region category matching is performed based on the multi-region similarity parameter matrix, specifically as follows: Set regional weight coefficients to characterize the degree of abnormality in each feature analysis area; arrhythmias caused by different physiological mechanisms will produce characteristic manifestations in specific areas of the scatter plot, so the pre-set personnel will set the weight of each area in different graphic areas.

[0040] Multiply all region similarity indices in the multi-region similarity parameter matrix by the region weight coefficient of their respective feature analysis regions, and then sum them to obtain the overall regional similarity of each feature analysis region in the multi-region similarity parameter matrix. Then, perform category matching verification. If the overall similarity of each feature analysis region exceeds the corresponding reference value, it is determined that the corresponding scatter plot presents the expected regional abnormal pattern, and the abnormal heart rhythm detection result is output. The corresponding reference value is specifically obtained as follows: for each region of each type of heart rhythm, calculate the similarity distribution of all samples in its standard feature vector library on the features of that region, and take the lower confidence boundary of the distribution as the reference value.

[0041] If the overall similarity of the feature analysis regions exceeds the corresponding reference value, the pixel density of the current scatter plot to be detected is input into the preset density distribution mapping relationship to obtain the boundary ambiguity parameters of each feature analysis region. This parameter guides how to adjust the geometric boundaries of the feature analysis regions. Specifically, it includes adjusting the direction and magnitude of each boundary of the region. Based on the adjustment direction and magnitude, the feature analysis regions are re-divided, and local classification matching is re-performed.

[0042] The specific construction process of the preset density distribution mapping relationship is as follows: collect PPG scatter plots containing various heart rhythms and signal qualities as a sample set. For each plot, select the optimal boundary of each feature analysis region after verification. Use pixel density as input feature and target boundary ambiguity parameter as output label. Use Support Vector Regression (SVR) to capture the nonlinear relationship between density and boundary adjustment amount. This nonlinear relationship is the preset density distribution mapping relationship.

[0043] In typical applications of wearable device monitoring, users are in non-laboratory environments where their PPG signals are often subject to complex interference due to micro-movements, instantaneous changes in vascular tension, or fluctuations in device contact pressure. In such cases, the point cloud distribution of the Lorenz scatter plot reconstructed from the signal often does not ideally fall within the feature analysis area with rigid geometric boundaries pre-defined according to standard physiological templates.

[0044] Due to slight local motion interference in the initial acquisition phase, the preliminary peak interval sequence generated after waveform segmentation and peak detection exhibits temporary instability. At this point, by calculating the degree of discrete fluctuation in the sequence, the system automatically identifies a risk to the reliability of peak detection in this signal segment. Based on this, an internal quality indicator is generated, guiding subsequent processing to apply a more cautious sequence optimization strategy to the data for this segment, including dynamic filtering based on statistical distribution. The optimization effect is that it avoids completely discarding the data for this segment due to local interference and avoids directly using unreliable raw peak sequences for subsequent analysis. Instead, through automated data cleaning and repair, it extracts the most effective rhythmic information, laying a reliable data foundation for generating high-quality Lorenz scatter plots, thereby reducing the possibility of overall analysis bias caused by signal quality issues at the source.

[0045] A Lorenz scatter plot was generated for an optimized interval sequence, and the process entered the classification matching stage. The scatter plot failed to match the atrial fibrillation template with high confidence in terms of global features, thus transitioning to a local classification matching process. After pre-defined region division and feature extraction, the multi-region similarity parameter matrix showed that the matching indicators for some feature regions were ambiguous, with no single rhythm type's overall region similarity clearly exceeding the dynamically calculated reference threshold. At this point, an intelligent processing mechanism for matching uncertainty was triggered. Based on the pixel density distribution of the scatter plot, the geometric boundaries of the feature analysis regions were dynamically fine-tuned, and region feature comparisons were re-performed. This process can identify and handle this matching ambiguity, adaptively adjusting the analysis framework to explore feature expressions that better fit the data's distribution. Ultimately, the output indicates the coexistence of multiple rhythm features, or maintains the judgment of no clear match. This objectively reflects the complexity of the current signal rhythm pattern and avoids misjudgments that may result from simple classification.

[0046] After the final abnormal heart rhythm judgment result is compared, if the judgment conclusion belongs to the first type of feature vector library (persistent atrial fibrillation or paroxysmal atrial fibrillation), the corresponding detection conclusion and its similarity are output; if the judgment conclusion does not belong to the first type of feature vector library, the overall similarity is calculated based on the similarity result and the set regional weights, and the heart rhythm type corresponding to the high similarity is used as the detection conclusion of the scatter plot to be detected; if the similarity of each heart rhythm type is lower than the set threshold standard, the detection conclusion is judged as other types of abnormality.

[0047] like Figure 5 The diagram shows the structure of an abnormal heart rhythm detection system based on PPG waveform segmentation and rhythm discrimination provided in an embodiment of the present invention. This system includes the following modules: a PPG waveform segmentation module, used to perform scatter image analysis on the original PPG optical signal during scatter plot image processing to obtain light intensity sequence data and segment the waveform; a rhythm interval sequence processing module, used to perform interval peak detection within the segmented waveform segments, identify each main wave peak point and its corresponding main wave timestamp, obtain the peak time interval sequence, and determine whether to perform sequence interval processing to obtain the rhythm interval sequence; an abnormal heart rhythm detection module, based on the panoramic Lorenz scatter plot generated by the rhythm interval sequence, performing global pixel distribution feature analysis to obtain pixel distribution feature vectors, and simultaneously performing local classification matching or global classification matching; and a classification matching output module, based on the results of local or global classification matching, determining whether to output the final abnormal heart rhythm detection result.

[0048] During the analysis of the segmented PPG waveform in Example 1, the acquired Lorenz scatter plot may contain multiple arrhythmic phenomena coupled together. When the PPG signal to be tested originates from a complex rhythm with multiple arrhythmic couplings, the generated Lorenz scatter plot will mix multiple graphic features, resulting in a low overall matching degree with any single rhythm template in the standard library. In this case, the category matching verification also includes: In Example 2, the second feature vector library corresponding to the column with the highest regional similarity index in the multi-region similarity parameter matrix is ​​obtained, and the upper quartile of the corresponding regional similarity index in the second feature vector library is used as the regional feature consistency probability. If there is at least one regional feature consistency probability greater than the preset regional feature consistency probability, the scatter plot to be detected is determined to be a composite feature pattern and is used as the output abnormal heart rhythm detection result. The preset regional feature consistency probability is represented by the summation and averaging of the historical regional feature consistency probabilities.

[0049] If the regional feature consistency probability is not greater than the preset regional feature consistency probability, the scatter plot to be detected is determined not to match the standard image library and is output as an abnormal heart rhythm detection result, while prompting the preset personnel to intervene.

[0050] By quantifying the matching strength of each region with different templates, the phenomenon of feature mixing can be intelligently identified; this avoids the high false alarm rate caused by misclassifying complex situations as a single heart rhythm type, thus significantly improving the reliability of the analysis conclusions; at the same time, it provides pre-set professionals with clear low-match or composite pattern modes, rather than meaningless fuzzy information, enabling them to quickly focus on key areas that need to be reviewed, achieving effective synergy between automated analysis and expert experience.

[0051] like Figure 6 The table shown is a table of abnormal heart rhythm simulation detection results provided in the embodiment of the present invention. It evaluates the recognition performance of the simulation detection process on nine types of heart rhythms and shows its simulation detection effect from multiple dimensions such as accuracy, recall, precision, and specificity. Overall, it shows high reliability in most heart rhythm types, especially high accuracy in sinus tachycardia and persistent atrial fibrillation, reflecting high recognition consistency and stability.

[0052] Table 1 Explanation of Quantitative Indicators The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0053] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0054] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0055] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0056] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0057] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0058] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0059] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0060] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0061] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An abnormal cardiac rhythm detection method based on PPG waveform segmentation and rhythm discrimination, characterized in that, Includes the following steps: S1, In the process of scatter plot image processing, the original PPG optical signal is analyzed by scatter plot image to obtain light intensity sequence data and perform waveform segment splitting; S2, within the split waveform segments, perform interval peak detection, identify each main wave peak point and its corresponding main wave timestamp, obtain the wave peak time interval sequence, and determine whether to perform sequence interval processing, thereby obtaining the rhythm interval sequence; S3. Based on the panoramic Lorenz scatter plot generated by the rhythm interphase sequence, perform global pixel distribution feature analysis to obtain pixel distribution feature vector, and simultaneously perform local classification matching or global classification matching. S4. Based on the results of the local classification matching or global classification matching, determine whether to output the final abnormal heart rhythm detection result. The global pixel distribution feature analysis process is as follows: Within the rhythm interphase sequence, the current peak time interval is paired with the next peak time interval to form interphase coordinate pairs, which are then filled into the two-dimensional discretized pixel grid corresponding to the preset coordinate range. The pixel value of the grid position where the inter-coordinate pair is located is set to 1 and used as the foreground pixel, while the pixel position values ​​of other pixels in the grid that are not occupied by the coordinate pair are all set to 0 and used as the background pixel. A scatter plot of points to be detected is formed based on foreground and background pixels; The dispersion measure and clustering measure of the scatter plot to be detected are obtained after statistical analysis. After scaling, the scatter accuracy index is obtained to reflect the graphic quality and classification reliability of the scatter plot to be detected. If the scatter point accuracy index is greater than the preset scatter point accuracy index, then global classification matching is performed; otherwise, local classification matching is performed. Based on the global fan-shaped distribution characteristic of atrial fibrillation rhythms, the pixel density and coordinate features of their scatter plots are extracted and integrated to form a first standard vector library. The pixel density and coordinate parameters of the image to be detected are input into the first standard vector library in the image library for comparison. For sinus rhythm and other non-atrial fibrillation abnormal rhythms, based on the different distribution patterns of their scatter plots in the central and non-central regions, feature regions are further divided and the pixel density and position parameters of the corresponding regions are extracted to construct a second standard vector library. This library centrally represents the regionalized features of various abnormal rhythms other than atrial fibrillation and is used for further subdivision and discrimination of rhythm types after excluding atrial fibrillation. The local classification matching process is as follows: Based on preset feature region division rules, the scatter plot to be detected is divided into a preset number of feature analysis regions; the dispersion measure and clustering measure obtained from each feature analysis region are compared with the second standard feature vector library to calculate the region similarity index; all feature analysis regions are traversed, and the region similarity index is coupled with the corresponding feature analysis region to obtain a multi-region similarity parameter matrix, where the rows are each feature analysis region and the columns are the region similarity index; region category matching is performed based on the multi-region similarity parameter matrix. The specific process of the region category matching is as follows: Set a region weight coefficient representing the degree of abnormality in each feature analysis region; multiply all region similarity indices in the multi-region similarity parameter matrix by the region weight coefficient of their respective feature analysis regions and sum them to obtain the overall region similarity of each feature analysis region in the multi-region similarity parameter matrix, and perform category matching verification: if the overall region similarity of each feature analysis region exceeds the corresponding reference value, it is determined that the corresponding scatter plot presents the expected region abnormality pattern, and the abnormal heart rhythm detection result is output; if the overall region similarity of each feature analysis region exceeds the corresponding reference value, the pixel density of the current scatter plot to be detected is input into a preset density distribution mapping relationship to obtain the boundary ambiguity parameters of each feature analysis region, and the geometric boundaries of each feature analysis region are adjusted, specifically including the adjustment direction and adjustment magnitude of each boundary of the region, and the feature analysis regions are re-divided according to the adjustment direction and adjustment magnitude, and local classification matching is re-performed; The specific construction process of the preset density distribution mapping relationship is as follows: collect PPG scatter plots containing various heart rhythms and signal qualities as a sample set. For each plot, select the optimal boundary of each feature analysis region after verification. Use pixel density as input feature and target boundary ambiguity parameter as output label. Use support vector regression to capture the nonlinear relationship between density and boundary adjustment amount. This nonlinear relationship is the preset density distribution mapping relationship. When the overall similarity of the regions does not exceed the corresponding reference value, the category matching verification further includes: obtaining the second feature vector library corresponding to the column with the highest region similarity index in the multi-region similarity parameter matrix; The upper quartile of the corresponding region similarity index in the second feature vector library is used as the region feature consistency probability. If at least one region feature consistency probability is greater than the preset region feature consistency probability, the scatter plot to be detected is determined to be a composite feature pattern, and the abnormal heart rhythm detection result is output. If none of the region feature consistency probabilities are greater than the preset region feature consistency probability, the scatter plot to be detected is determined to be not matched with the standard image library, the abnormal heart rhythm detection result is output, and the preset personnel are prompted to intervene.

2. The abnormal heart rhythm detection method based on PPG waveform segmentation and rhythm discrimination as described in claim 1, characterized in that, The specific process of scatter image parsing is as follows: The acquired raw PPG optical signal is analyzed to obtain the raw optical intensity sample values ​​arranged in time order, and the acquisition time corresponding to each sampling point is recorded to form an optical intensity sequence with time label; The light intensity sequence is divided into a preset number of waveform analysis segments according to equal time lengths, and PPG waveform sequence segments are obtained after differential filtering. The PPG waveform sequence segments represent preprocessed, independent PPG waveform units that can be used for feature extraction; Obtain the peak point of the main wave and the corresponding main wave timestamp of each PPG waveform sequence segment, and calculate the time difference between adjacent main wave timestamps as the peak time interval of the waveform segment. Arrange all peak time intervals in chronological order to obtain the peak time interval sequence; The standard deviation of the peak time interval sequence is calculated to obtain the peak change rate used to quantify the discrete fluctuation of PPG waveform sequence segments; When the peak change rate is greater than the preset peak change rate, it indicates that the current peak detection is unstable and prompts to reacquire the original PPG optical signal; otherwise, inter-sequence processing is performed.

3. The abnormal heart rhythm detection method based on PPG waveform segmentation and rhythm discrimination as described in claim 2, characterized in that, The interphase processing of the sequence is specifically as follows: Peak time intervals that exceed the preset qualified range in the peak time interval sequence are removed, and the peak time interval sequence after peak time interval removal is used as the rhythm interval sequence and sequence integrity is checked. If a missing segment is detected in the interrhythm sequence, the corresponding missing data length is input into the preset missing data-interpolation repair mapping relationship to obtain the missing data filling value to fill the missing segment. Based on the filled interrhythm sequence, global pixel distribution feature analysis is performed. If no missing segments are detected in the interrhythm sequence, global pixel distribution feature analysis is performed directly.

4. The abnormal heart rhythm detection method based on PPG waveform segmentation and rhythm discrimination as described in claim 1, characterized in that, The global classification and matching process is as follows: The dispersion measure and the clustering measure are compared with the first standard feature vector library one by one to obtain similarity groups, and the summation and average of the similarity groups are used as the standard similarity. If the standard similarity exceeds the preset standard similarity, then fine-grained identification and matching will be performed; otherwise, the data changes of the scatter plot to be detected will continue to be monitored, and a new analysis cycle will be waited for.

5. The abnormal heart rhythm detection method based on PPG waveform segmentation and rhythm discrimination as described in claim 4, characterized in that, The detailed identification and matching process is as follows: Using the origin of the scatter plot to be detected and the diagonal line corresponding to its maximum horizontal and vertical coordinate points as the central axis, a sliding analysis window with a fixed step size is set. The discrete variance of scattered pixels within the sliding analysis window is used as the divergence index of the sliding analysis window. If the divergence index exceeds the preset divergence index in each sliding analysis window, it is determined that the corresponding scatter plot to be detected presents the expected abnormal divergence pattern, and the abnormal heart rhythm detection result is output. Otherwise, the corresponding scatter plot to be tested is determined to be within the normal fluctuation range or a local anomaly.

6. An abnormal heart rhythm detection system based on PPG waveform segmentation and rhythm discrimination, applying the abnormal heart rhythm detection method based on PPG waveform segmentation and rhythm discrimination as described in any one of claims 1-5, includes the following modules: The PPG waveform splitting module is used to perform scatter image analysis on the original PPG optical signal during the scatter plot image processing, obtain light intensity sequence data, and split the waveform segments. The rhythm interval sequence processing module is used to perform interval peak detection within the split waveform segments, identify each main wave peak point and its corresponding main wave timestamp, obtain the wave peak time interval sequence, and determine whether to perform sequence interval processing, thereby obtaining the rhythm interval sequence. The abnormal heart rhythm detection module performs global pixel distribution feature analysis based on the panoramic Lorenz scatter plot generated by the rhythm interval sequence, obtains pixel distribution feature vector, and performs local classification matching or global classification matching. The classification matching output module determines whether to output the final abnormal heart rhythm detection result based on the results of the local classification matching or global classification matching.

Images