Heart valve disease pre-diagnosis method based on heart sound sound propagation topological map

Abstract

The application discloses a heart valve disease pre-diagnosis method based on heart sound sound propagation topology atlas, and comprises the following steps: acquiring heart sound signals of multiple standard heart auscultation areas of a subject; preprocessing and extracting effective heart cycles; establishing a unified heart cycle phase coordinate based on the first and second heart sounds; extracting acoustic event descriptors of each auscultation area; constructing a multi-auscultation area heart sound sound propagation topology atlas containing node features and edge weights according to energy attenuation, time delay, spectral similarity, phase consistency and conduction direction; extracting an observed acoustic distribution from the node features, and estimating valve sound source contribution in combination with the edge weights and the valve sound source-auscultation area propagation relationship; matching the node features, the edge weights and the valve sound source contribution results with a valve disease acoustic template library generated by historical labeled samples to obtain an evidence score, and then outputting a patient-level pre-diagnosis result. The application can be used for interpretable pre-diagnosis by utilizing the multi-auscultation area heart sound sound propagation relationship, and is helpful for early screening and auxiliary judgment of heart valve diseases.

Description

Technical Field

[0001] This invention relates to the fields of medical signal processing, intelligent auscultation, and heart sound-assisted analysis, specifically to a method for the pre-diagnosis of valvular heart disease based on a heart sound propagation topology map. Background Technology

[0002] Valvular heart disease is one of the most common structural heart diseases. Clinical diagnosis typically relies on imaging examinations such as echocardiography and CT scans. However, these examinations have specific requirements for equipment, facilities, and professional personnel, making them insufficient to meet the needs of large-scale early screening in primary healthcare institutions, community screening centers, health checkup centers, and telemedicine settings. Traditional cardiac auscultation has advantages such as being non-invasive, convenient, and low-cost, but its results are highly dependent on the physician's experience and easily affected by factors such as auscultation location, background noise, operator experience, and the patient's position. With the development of electronic stethoscopes, mobile terminals, and cloud platforms, heart sound signals can now be digitally acquired, stored, transmitted, and analyzed, providing the foundation for intelligent pre-diagnosis of valvular heart disease.

[0003] Existing patents and research on heart sounds have disclosed various methods for heart sound acquisition, segmentation, classification, and acoustic atlas construction. For example, CN114159091B discloses a heart sound propagation relationship detection system based on a wearable sensor array, which uses multi-channel heart sound array signals and sensor positions to draw a heart sound energy relationship map to show the heart sound energy relationship between different locations on the body surface; CN105653880B discloses a method for tracing the source of the cardiac sound field based on heart sounds, which infers the energy distribution of the internal heart sound field from the energy distribution of the external heart sound field and reconstructs the cardiac sound field. The above schemes indicate that multi-channel heart sounds, heart sound propagation relationships, and heart sound field tracing have a certain technical foundation, but their main purpose is biased towards the detection of heart sound propagation patterns, sound field reconstruction, or heart sound source component analysis, and has not formed an integrated method for multi-auscultation zone acoustic propagation topology mapping, valve sound source contribution estimation, and valvular disease evidence scoring for the prediagnosis of valvular heart disease.

[0004] On the other hand, an existing patent (CN111759345B) also discloses a method for analyzing heart valve abnormalities based on convolutional neural networks. This method collects heart sounds from multiple auscultation areas, calculates the time spectrum, inputs it into a convolutional neural network to determine whether it is normal or abnormal, and then combines the envelope spectrum, power spectrum, hidden semi-Markov model and support vector machine to obtain the analysis results of aortic or mitral valve related abnormalities. Related papers (Guo et.al. J Pers Med. 12(12):2011.; Kawamura et.al. AnnuInt Conf IEEE Eng Med Biol Soc. 2007:2875-8) also propose methods such as multi-channel heart sound visualization system, chest wall heart sound propagation path estimation, cardiac acoustic mapping and multi-channel heart sound coupling feature classification, which are used to study the spatial distribution and propagation path of heart sounds in the chest wall or improve the classification performance of heart diseases. Although the above methods utilize multi-channel or multi-auscultation area information, they are mostly used for acoustic visualization, propagation path research, input of ordinary classification models, or multi-channel feature fusion. They have not yet uniformly mapped the heart sounds of multiple standard cardiac auscultation areas to the cardiac cycle phase coordinates, further constructed a topological map of heart sound propagation containing node features and edge weights, and performed valve sound source contribution estimation and valvular disease template matching based on the valve sound source-auscultation area propagation relationship.

[0005] Therefore, existing technologies still have the following shortcomings: First, they do not make sufficient use of the acoustic propagation relationship between heart sounds in multiple auscultation areas, often simply splicing, classifying, or merging heart sounds from different auscultation areas by voting; second, they lack a structured intermediate representation that can simultaneously express auscultation area, cardiac cycle phase, frequency band characteristics, and cross-auscultation area propagation relationship; third, existing classification results are mostly black-box outputs, making it difficult to indicate which auscultation area, cardiac cycle phase, frequency band, and valve sound source the abnormal heart sounds mainly occur in; and fourth, existing methods cannot distinguish between local noise enhancement and pathological acoustic abnormalities that conform to the propagation law of valve sound sources. Summary of the Invention

[0006] The purpose of this invention is to provide a method for the pre-diagnosis of valvular heart disease based on the topological map of heart sound propagation, in order to solve the problems of insufficient utilization of sound propagation relationships in multiple auscultation zones, insufficient expression of acoustic structures across auscultation zones, insufficient interpretation of valvular sound sources, and insufficient interpretability of patient-level pre-diagnosis results in existing intelligent analysis methods for valvular heart disease.

[0007] Existing methods for analyzing heart sounds often treat them as ordinary audio signals, typically segmenting, extracting features, or classifying heart sounds in a single or multiple auscultation areas to indicate normality, abnormality, or a specific disease. However, valvular heart disease-related murmurs are not merely localized abnormalities in a single audio segment; they are also closely related to the valve where the abnormal sound source is located, the phase of the cardiac cycle, frequency band distribution, the location of the auscultation area, and the propagation path through the chest wall. Simply classifying a single heart sound segment or simply voting on the results from multiple auscultation areas fails to adequately represent the sound propagation relationships between different auscultation areas, such as energy attenuation, phase consistency, spectral similarity, and conduction direction. Therefore, this invention aims to convert heart sound signals collected from multiple standard cardiac auscultation areas into a structured multi-auscultation area heart sound propagation topology map. Furthermore, by combining the valve sound source-auscultation area propagation relationship, a valvular heart disease acoustic template library, and an evidence scoring mechanism, it achieves interpretable patient-level prediagnosis of valvular heart disease.

[0008] To achieve the above objectives, this invention establishes a method for the pre-diagnosis of valvular heart disease based on the topological map of cardiac sound propagation, comprising the following steps: S1. Acquire heart sound signals from the subject in multiple standard cardiac auscultation areas. These multiple standard cardiac auscultation areas include at least two of the aortic valve area, pulmonary valve area, Erb area, tricuspid valve area, and mitral valve area. Each auscultation area corresponds to a segment of heart sound signal, and the corresponding auscultation area label is recorded.

[0009] S2. Preprocess the heart sound signals from each auscultation area and extract the effective cardiac cycle from the preprocessed heart sound signals. The preprocessing includes at least one of bandpass filtering, environmental noise suppression, amplitude normalization, abnormal segment removal, clipping distortion repair, and probe friction interference repair. The effective cardiac cycle is determined based on the first and second heart sounds.

[0010] S3. Based on the first and second heart sounds in the effective cardiac cycle, map the heart sound signals from different auscultation areas to a unified cardiac cycle phase coordinate system. The unified cardiac cycle phase coordinate system includes a systolic phase interval and a diastolic phase interval; wherein the systolic phase interval is determined by the interval from the first heart sound to the second heart sound, and the diastolic phase interval is determined by the interval from the second heart sound to the next first heart sound. Through this step, heart sound signals from different auscultation areas and different cardiac cycle lengths can be compared under a unified phase scale.

[0011] S4. Under the unified cardiac cycle phase coordinates, extract acoustic event descriptors for each auscultation zone. The acoustic event descriptors include at least one of the following: first heart sound intensity, second heart sound intensity, systolic murmur energy, diastolic murmur energy, murmur onset phase, murmur termination phase, murmur duration ratio, peak phase, spectral centroid, spectral bandwidth, spectral entropy, band energy ratio, period consistency, and cross-cycle energy variation coefficient.

[0012] S5. Based on the acoustic event descriptors of multiple auscultation areas, construct a multi-auscultation area heart sound propagation topology map, which is represented as follows: in, This represents the topological map of cardiac sound propagation in multiple auscultation areas. Represents a set of nodes. The set of edges representing the nodes in the auscultation area can be represented by a fully connected approach or a connection method defined based on the anatomical positional relationships of the auscultation areas. Representing node characteristics, The edge weights are represented. The nodes include at least auscultation area nodes; the node features... The edges are used to represent the acoustic event characteristics of each auscultation zone within different cardiac cycle phase intervals and different frequency bands; the edges are used to represent the sound propagation relationships between different auscultation zones; the edge weights are... Used to represent at least one of the following: energy attenuation relationship, time delay relationship, spectral similarity, phase consistency, conduction direction, and periodic stability between different auscultation zones.

[0013] S6. Based on the node features in the multi-auscultation area cardiac sound propagation topology map Extracting acoustic distribution from multiple auscultation areas Based on the edge weights in the multi-auscultation area cardiac sound propagation topology map Constraints on sound propagation relationships across auscultation zones are extracted, and combined with the propagation relationship between the valve sound source and the auscultation zone, the contribution of the valve sound source is estimated to obtain the valve sound source contribution vector. Specifically, the propagation relationship between the valve sound source and the auscultation area is represented by a propagation matrix. : in, This represents the valve sound source-stethoscope area propagation matrix. Indicates the first Typical propagation vectors of a valvular sound source across multiple standard cardiac auscultation areas. This indicates the number of sound sources in the valve.

[0014] Furthermore, the valve sound source contribution estimation includes: 1) based on the node characteristics of the multi-auscultation area cardiac sound propagation topology map. Extract the observed acoustic distribution vectors on multiple auscultation areas of the subject. ;2) Based on the edge weights of the multi-auscultation area cardiac sound propagation topology map 3) Extract the constraints on sound propagation relationships across auscultation zones; 4) Based on the propagation matrix The observed acoustic distribution vector and the edge weights Estimate the valve sound source contribution vector ;in, This indicates the observed acoustic distribution across multiple auscultation areas. This indicates the sound propagation relationship between different auscultation areas. This represents the degree of contribution of different valvular sound sources to the observed acoustic distribution and sound propagation relationship; the estimation of the valvular sound source contribution is implemented by any one or more of the following: non-negativity constraint, sparse constraint, template similarity calculation, Bayesian inversion, graph matching, or weighted least squares.

[0015] S7. Analyze the node features in the multi-area cardiac sound propagation topology map. Edge weight and valve sound source contribution vector Matching with a valvular disease acoustic template library yields evidence scores for at least one type of valvular heart disease.

[0016] Furthermore, the valvular disease acoustic template library is generated from historical heart sound samples with valvular disease annotation information. The valvular disease annotation information originates from echocardiographic examination results, clinical expert confirmation results, medical record results, or a combination of the above. Specifically, 1) historical heart sound samples with valvular disease annotation information are obtained; 2) S1 to S6 are performed on each historical heart sound sample to obtain the corresponding historical multi-auscultation area heart sound propagation topology map and historical valve sound source contribution estimation results; 3) historical heart sound samples are grouped according to the valvular disease annotation information; 4) based on the node features, edge weights, and valve sound source contribution estimation results of historical heart sound samples within the same group, corresponding valvular disease acoustic templates are generated.

[0017] Each type of valvular disease acoustic template includes at least one of the following: typical node features, typical edge weights, typical valvular sound source contribution patterns, and template quality parameters for the corresponding valvular disease type. Specifically, the typical node features represent the distribution of acoustic events in different auscultation zones, different cardiac cycle phase intervals, and different frequency bands for this type of valvular disease; the typical edge weights represent the sound propagation relationship between different auscultation zones for this type of valvular disease; and the typical valvular sound source contribution patterns represent the valvular sound source contribution characteristics corresponding to this type of valvular disease.

[0018] Furthermore, the evidence score We obtain it from the following formula: in, Indicates the subject and the first Evidence score between acoustic templates of valvular disease; Indicates the node feature matching score; This represents the edge weight matching score; This indicates the valve sound source contribution matching score; This indicates the heart sound quality matching score; This represents the weighting coefficient of the corresponding matching score.

[0019] S8. Based on the evidence score and the confidence level of heart sound quality, output a patient-level prediagnostic result for valvular heart disease. The patient-level prediagnostic result for valvular heart disease includes at least one of the following: normal indication, suspected valvular abnormality indication, suspected regurgitation-type valvular heart disease indication, suspected stenotic valvular heart disease indication, valvular heart disease risk level indication, and recommendation for further echocardiography.

[0020] Furthermore, in addition to outputting patient-level prediagnostic results for valvular heart disease, interpretable evidence is also output. This interpretable evidence includes at least one of the following: major abnormal auscultation areas, abnormal cardiac cycle phases, major abnormal frequency bands, suspected valvular sound sources, sound propagation topology matching scores, and heart sound quality reliability.

[0021] Beneficial effects Compared with the prior art, the present invention has the following beneficial effects.

[0022] First, this invention converts heart sound signals from multiple standard cardiac auscultation areas into a multi-auscultation area heart sound propagation topology map, which can explicitly express the energy attenuation, time delay relationship, spectral similarity, phase consistency, conduction direction and periodic stability between different auscultation areas. Compared with single-segment heart sound classification or simple multi-auscultation area result fusion, it can better reflect the spatial propagation law of acoustic events related to valvular heart disease.

[0023] Secondly, by unifying the phase coordinates of the cardiac cycle, this invention enables heart sound signals from different auscultation areas and with different cardiac cycle lengths to be compared under the same cardiac cycle phase scale, which is beneficial for identifying the distribution patterns of abnormal acoustic events during systole, diastole, or other specific phase intervals.

[0024] Third, this invention estimates the contribution of valvular sound sources based on the propagation relationship between valvular sound sources and auscultation areas. This can link the acoustic distribution observed in multiple auscultation areas with potential valvular sound sources, thereby providing interpretable evidence such as suspected valvular sound sources, major abnormal auscultation areas, and abnormal propagation relationships for the pre-diagnosis of patient-level valvular heart disease.

[0025] Fourth, this invention uses a valvular heart disease acoustic template library generated from historically labeled heart sound samples to perform template matching and evidence scoring. It can not only output patient-level pre-diagnosis results for valvular heart disease, but also output evidence of node feature matching, edge weight matching, sound source contribution matching, and heart sound quality matching corresponding to the results, which is beneficial for doctors to review and make further examination decisions.

[0026] Fifth, this invention can be used in electronic stethoscopes, mobile terminals, or cloud-based heart sound analysis platforms, and is applicable to scenarios such as primary healthcare, community screening, physical examination centers, telemedicine, and bedside heart sound auxiliary analysis. It helps to improve the accessibility, standardization, and interpretability of early screening and auxiliary diagnosis of valvular heart disease. Attached Figure Description

[0027] Figure 1 This is the overall flowchart of the present invention.

[0028] Figure 2 This is a schematic diagram illustrating the construction of the cardiac sound propagation topology map in multiple auscultation areas according to the present invention.

[0029] Figure 3 This is a schematic diagram illustrating the estimation of valvular sound source contribution, acoustic template matching for valvular diseases, and the output of pre-diagnosis results in this invention. Detailed Implementation

[0030] The present invention will be further described below with reference to the accompanying drawings and embodiments. It should be understood that the following embodiments are only for illustrating the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Where there is no conflict, the technical features in the following embodiments can be combined with each other.

[0031] Example 1: Prediction of regurgitant valvular heart disease based on the acoustic propagation topology of heart sounds in five standard auscultation areas This embodiment uses heart sounds from five standard cardiac auscultation areas as input to illustrate how the present invention completes the entire process from heart sound acquisition to outputting patient-level pre-diagnosis results for valvular heart disease.

[0032] S1 Acquisition of Heart Sound Signals in Multiple Auscultation Zones Heart sounds were collected using an electronic stethoscope in the aortic valve area (A), Erb area (E), mitral valve area (M), pulmonary valve area (P), and tricuspid valve area (T). Each auscultation area was recorded for 10 seconds at a sampling rate of 8000 Hz.

[0033] The heart sound signals collected from the five auscultation areas were recorded as follows: , , , , in, Indicates time, , , , , These represent the aortic valve area, Erb area, mitral valve area, pulmonary valve area, and tricuspid valve area, respectively.

[0034] S2 Heart Sound Preprocessing and Effective Cardiac Cycle Extraction Bandpass filtering, amplitude normalization, abnormal segment removal, and distortion region repair were performed on the heart sound signals from each auscultation area to obtain the preprocessed heart sound signals: , , , , In this embodiment, the filtering frequency band is set to 25 Hz to 800 Hz to retain the main heart sounds and noise-related frequency bands. Sudden abnormal segments with amplitudes exceeding 5 times the local mean standard deviation are removed or interpolated for repair.

[0035] Subsequently, the first heart sound S1 and the second heart sound S2 were detected in each auscultation area. Taking the mitral valve area M as an example, 11 valid cardiac cycles were detected in a 10-second heart sound interval, of which 10 cycles met the signal quality requirements. The number of valid cycles was recorded as follows. .

[0036] S3 Unified Cardiac Cycle Phase Coordinate Establishment For the For each effective cardiac cycle, the location of the first heart sound is defined as [number] effective cardiac cycles. The position of the second heart sound is The next first heart sound position is Map this cycle to a unified cardiac cycle phase coordinate system. ,in .

[0037] In this embodiment, the contraction period is mapped to Mapping diastole to That is, the boundary parameter The phase mapping formula is as follows: in, Indicates the auscultation area number. Indicates the cardiac cycle number. Indicates the original time point, This represents the normalized phase of the cardiac cycle. This represents the boundary parameter between systole and diastole in a unified phase coordinate system.

[0038] Through this step, the heart sounds in all five auscultation areas are mapped to a unified cardiac cycle phase scale.

[0039] S4 Acoustic Event Descriptor Extraction Under a unified cardiac cycle phase coordinate system, a cardiac cycle is divided into several phase intervals. This embodiment selects the mid-to-late systolic phase interval. And select the mid-to-high frequency band Hz is used as the target frequency band to characterize noise-related acoustic events.

[0040] For the The first auscultation area, the first Phase intervals and the Each frequency band Acoustic energy Calculate using the following formula: in, Indicates the first Each auscultation zone in the phase interval and frequency band The average acoustic energy within; Indicates the first The number of effective cardiac cycles in each auscultation area; Indicates the first The first auscultation area A valid cardiac cycle in uniform phase and frequency The time-frequency representation below; Indicates frequency.

[0041] This embodiment obtains five auscultation zones in , The normalized acoustic energy within Hz is as follows: This results in an acoustic distribution vector observed in multiple auscultation zones: in, This indicates the observed acoustic distribution of the five auscultation zones within the target phase interval and target frequency band interval.

[0042] In addition to the energy characteristics mentioned above, this embodiment also extracts the noise onset phase, noise termination phase, noise duration ratio, spectral centroid, spectral entropy, and cross-cycle energy variation coefficient. For example, the noise onset phase in the mitral lobe region M is 0.22, the termination phase is 0.44, the noise duration ratio is 0.22, the spectral centroid is 248 Hz, and the cross-cycle energy variation coefficient is 0.18.

[0043] Construction of the cardiac sound propagation topology map in the S5 multi-auscultation area Based on the acoustic event descriptors of the five auscultation areas, a multi-auscultation area cardiac sound propagation topology map is constructed: in, This represents the topological map of cardiac sound propagation in multiple auscultation areas. Represents the set of nodes in the auscultation area; Represents the set of edges between nodes in the auscultation area; Represent node characteristics; This represents the edge weight.

[0044] In this embodiment, node features This includes noise energy, peak phase, spectral centroid, and periodic stability in each auscultatory zone within the target phase and frequency band intervals. The node feature matrix can be represented as: The first column represents the target phase and normalized noise energy within the frequency band; the second column represents the peak phase; the third column represents the spectral centroid in Hz; and the fourth column represents the periodic stability, with higher values ​​indicating stronger cross-cycle repeatability.

[0045] edge weight This is used to represent the sound propagation relationship between different auscultation zones. In this embodiment, energy attenuation is selected as one of the edge weights. For the... The auscultation area and the first The energy attenuation relationship of each auscultation area The calculation is as follows: in, Indicates the first The auscultation area and the first The energy attenuation relationship between each auscultation zone; Indicates the first Acoustic energy of the auscultation area; Indicates the first Acoustic energy of the auscultation area; To prevent positive numbers with a denominator of zero, this embodiment takes... .

[0046] For example, the energy attenuation relationship between the mitral valve region M and the Erb region E is as follows: The energy attenuation relationship between the mitral valve region M and the aortic valve region A is as follows: The above results indicate that, within the target phase and frequency band of this subject, abnormal acoustic events are strongest in the mitral valve region M, and attenuate to varying degrees towards the Erb region E, the aortic valve region A, and the tricuspid valve region T.

[0047] S6 valve acoustic source contribution estimation This embodiment sets up four potential valve sound sources, including aortic valve sound source AV, pulmonary valve sound source PV, mitral valve sound source MV, and tricuspid valve sound source TV.

[0048] Valve sound source-auscultation area propagation matrix The settings are as follows: in, This represents the valve sound source-auscultation area propagation matrix; each column represents the typical propagation vector of a valve sound source across the five auscultation areas; the four columns correspond to the aortic valve sound source (AV), pulmonary valve sound source (PV), mitral valve sound source (MV), and tricuspid valve sound source (TV) respectively; the five rows correspond to areas A, E, M, P, and T respectively. These values ​​can be predefined by experts or obtained from statistical analysis of historical labeled heart sound samples.

[0049] Based on observation vector Edge weight relationship and propagation matrix Non-negative least squares estimation is used to estimate the valve sound source contribution vector. : in, This represents the estimated valve sound source contribution vector; Indicates by node features The acoustic distribution of the extracted multi-auscultation zone was observed. This represents the valve sound source-auscultation area propagation matrix; Represents the edge weights in the acoustic propagation topology graph; Represented by the propagation matrix Harmony source contribution vector The derived theoretical sound propagation relationship; Indicates the weight of edge relationship constraints; Indicates the sparse constraint weights; This represents the Frobenius norm.

[0050] The calculations obtained in this embodiment are as follows: After normalization, we get: The four elements represent the contribution proportions of the aortic valve sound source (AV), pulmonary valve sound source (PV), mitral valve sound source (MV), and tricuspid valve sound source (TV), respectively. This result suggests that the target anomalous acoustic event primarily conforms to the mitral valve sound source propagation pattern.

[0051] S7 Valvular Disease Acoustic Template Matching and Evidence Scoring In this embodiment, the acoustic template library for valvular heart disease includes normal templates, regurgitation-type valvular heart disease templates, stenosis-type valvular heart disease templates, and mixed-type valvular abnormality templates. Each type of template is generated from historically labeled heart sound samples, and the valvular heart disease labeling of the historical samples comes from echocardiographic examination results and clinical expert confirmation results.

[0052] For each historical heart sound sample, S1 to S6 are performed to obtain its historical acoustic propagation topology map and historical valvular sound source contribution vector. After grouping according to valvular disease labeling, the typical node features, typical edge weights, typical valvular sound source contribution patterns and template quality parameters of each group are calculated to form valvular disease acoustic templates.

[0053] For the Valvular disease template, evidence score between the subject and the template The calculation is as follows: in, Indicates the subject and the first Evidence score between acoustic templates of valvular disease; The node feature matching score is represented by the node features in the subject topology graph. Comparison with template node features yields the following results; The edge weight matching score is represented by the edge weights in the subject topology graph. Obtained by comparison with template edge weights; The valve sound source contribution matching score is represented by the subject's valve sound source contribution vector. This was obtained by comparing with the template sound source contribution mode; This indicates the heart sound quality matching score; This represents the weighting coefficient of the corresponding matching score.

[0054] In this embodiment, the following is taken: , , , Based on the regurgitation-type valvular disease template, the following calculations were performed: , , , therefore: The evidence scores for the same subject compared to other templates are as follows: S8 Patient-Level Prediagnosis Results Output Let the evidence score threshold be... Because the evidence score for the regurgitation-type valvular disease template is the highest, and it meets the following criteria: Therefore, the patient-level prediagnosis result for valvular heart disease output in this embodiment is: suspected regurgitation type valvular heart disease, and further echocardiography is recommended.

[0055] Simultaneously, the output provides interpretable evidence: abnormal acoustic events are mainly located in the mid-to-late systolic phase; the main abnormal frequency band is 150 Hz to 400 Hz; the main abnormal auscultation area is the mitral valve area M; abnormal acoustic events propagate attenuatedly from area M to areas E, A, and T; valve sound source contribution estimation suggests that the mitral valve has the highest sound source contribution; and the acoustic propagation topology map has a high matching score with the template of regurgitation-type valvular heart disease.

[0056] Example 2: Pre-diagnosis of valvular stenosis based on acoustic propagation topology and template library This embodiment illustrates the complete execution process of the present invention under another type of valvular disease acoustic mode. Similar to Embodiment 1, this embodiment still includes all steps, but the observed acoustic distribution and template matching results are different.

[0057] S1 Multi-Auscultation Zone Heart Sound Signal Acquisition Heart sounds were acquired using an electronic stethoscope in the aortic valve area (A), Erb area (E), mitral valve area (M), pulmonary valve area (P), and tricuspid valve area (T). Each auscultation area was recorded for 30 seconds at a sampling rate of 16000 Hz.

[0058] Heart sounds were detected in five auscultation areas: , , , , in, Indicates time.

[0059] S2 Heart Sound Preprocessing and Effective Cardiac Cycle Extraction The raw heart sound signals were bandpass filtered, amplitude normalized, background noise suppressed, and abnormal segment removed. After preprocessing, the number of effective cardiac cycles obtained in each of the five auscultation areas is as follows: Periods with poor quality, unstable envelope, or predominantly burst noise are removed so that subsequent analysis is based solely on valid cardiac cycles.

[0060] S3 Unified Cardiac Cycle Phase Coordinate Establishment For each valid cardiac cycle, the first heart sound (S1) and the second heart sound (S2) are located, and S1 to S2 are mapped to the systolic phase interval, while S2 to the next S1 is mapped to the diastolic phase interval. This embodiment also uses... .

[0061] For any auscultation area and any effective period , original time Mapping to normalized cardiac cycle phase After mapping is completed, heart sound events in different auscultation areas can all be mapped to a unified phase coordinate system. Below is a comparison.

[0062] S4 Acoustic Event Descriptor Extraction This embodiment selects the early to mid-contraction phase interval. and mid-to-high frequency bands Hz is used to extract noise-related acoustic events.

[0063] Within this phase and frequency band range, the normalized acoustic energy of the five auscultation zones was calculated, and the results are as follows: Forming the observed acoustic distribution vector: in, This represents the observed acoustic distribution of the subjects across five standard auscultation areas. The results show that abnormal acoustic events were strongest in the aortic valve area (A), followed by the Erb area (E), and then decreased towards other auscultation areas.

[0064] This embodiment also extracts node features, including target phase and acoustic energy within the frequency band, peak phase, spectral centroid, and periodic stability, to obtain a node feature matrix: The columns represent, in order, normalized noise energy, peak phase, spectral centroid, and periodic stability.

[0065] Construction of the cardiac sound propagation topology map in the S5 multi-auscultation area Constructing a topological map of cardiac sound propagation across multiple auscultation areas: in, , The above node feature matrix, This is the edge weight matrix.

[0066] In this embodiment, the edge weights consider both energy decay and phase consistency. The energy decay relationship is as follows: Phase consistency is: in, and The first The and the first Acoustic energy of the auscultation area; and The first The and the first Peak phase of abnormal acoustic events in a single auscultation area; To prevent positive numbers with a denominator of zero.

[0067] Taking areas A and E as examples: The energy attenuation relationship between region A and region M is as follows: Since the peak phase of region A is 0.26 and the peak phase of region E is 0.27, therefore: The above results indicate that the target anomalous acoustic event of the subject was strongest in region A, had high phase consistency with region E, and showed significant energy attenuation compared to regions M and T.

[0068] S6 valve acoustic source contribution estimation The same valve sound source-auscultation zone propagation matrix as in Example 1 was used. : Among them, the four columns correspond to the aortic valve sound source AV, the pulmonary valve sound source PV, the mitral valve sound source MV, and the tricuspid valve sound source TV, respectively.

[0069] The valve sound source contribution vector is estimated using non-negative least squares: The calculations obtained in this embodiment are as follows: After normalization, we get: The four elements represent the contribution proportions of the aortic valve sound source (AV), pulmonary valve sound source (PV), mitral valve sound source (MV), and tricuspid valve sound source (TV), respectively. This result suggests that abnormal acoustic events primarily conform to the aortic valve sound source propagation pattern.

[0070] S7 Valvular Disease Acoustic Template Matching and Evidence Scoring This embodiment utilizes a valvular heart sound acoustic template library. The library is generated from historical heart sound samples, all of which have echocardiographic examination results or clinical expert confirmation. The library includes normal templates, regurgitation-type valvular heart sound templates, stenotic-type valvular heart sound templates, and mixed-type valvular abnormality templates.

[0071] In this embodiment, for the stenotic valvular disease template, the following calculations were performed: , , , Take the weight: , , , The evidence score is calculated as follows: The evidence scores for the same subject compared to other templates are as follows: S8 Patient-Level Prediagnosis Results Output Let the evidence score threshold be... Because the evidence score for the stenotic valvular disease template is the highest, and it meets the following criteria: Therefore, the output of this embodiment is: suspected stenotic valvular heart disease, and further echocardiography is recommended.

[0072] Simultaneously, the output provides interpretable evidence: aberrant acoustic events are mainly located in the early to mid-systolic phase; the main aberrant frequency band is 200 Hz to 500 Hz; the main aberrant auscultation area is aortic valve area A; the aberrant events are strongest in area A and propagate towards area E (Erb); there is a high phase consistency between areas A and E; valve sound source contribution estimation suggests that the aortic valve has the highest sound source contribution; the acoustic propagation topology map has a high matching score with the stenosis-like valvular disease template.

[0073] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for the pre-diagnosis of valvular heart disease based on cardiac sound propagation topology, characterized in that, Includes the following steps: S1. Acquire heart sound signals from the subject in multiple standard cardiac auscultation areas; S2. Preprocess the heart sound signals in each auscultation area and extract the effective cardiac cycle from the preprocessed heart sound signals; S3. Based on the first and second heart sounds in the effective cardiac cycle, map the heart sound signals from different auscultation areas to a unified cardiac cycle phase coordinate; S4. Extract acoustic event descriptors for each auscultation zone under the unified cardiac cycle phase coordinates; S5. Based on the acoustic event descriptors of multiple auscultation areas, construct a multi-auscultation area heart sound propagation topology map, which is represented as follows: in, This represents the topological map of cardiac sound propagation in multiple auscultation areas. Represents a set of nodes. Denotes the set of edges. Representing node characteristics, The edge weights are represented; the nodes include at least auscultation zone nodes, and the edges are used to represent the sound propagation relationships between different auscultation zones; S6. Based on the node features in the multi-auscultation area cardiac sound propagation topology map Extracting acoustic distribution from multiple auscultation areas Based on the edge weights in the multi-auscultation area cardiac sound propagation topology map Constraints on sound propagation relationships across auscultation zones are extracted, and combined with the propagation relationship between the valve sound source and the auscultation zone, the contribution of the valve sound source is estimated to obtain the valve sound source contribution vector. ; S7. Analyze the node features in the multi-area cardiac sound propagation topology map. Edge weight and valve sound source contribution vector Matching with an acoustic template library of valvular heart disease yields evidence scores for at least one type of valvular heart disease; S8. Based on the evidence score and the confidence level of heart sound quality, output the patient-level prediagnosis results for valvular heart disease.

2. The method according to claim 1, characterized in that, In step S1, the plurality of standard cardiac auscultation areas include at least two of the aortic valve area, pulmonary valve area, Erb area, tricuspid valve area, and mitral valve area; in step S2, the preprocessing includes at least one of bandpass filtering, environmental noise suppression, amplitude normalization, abnormal segment removal, clipping distortion repair, and probe friction interference repair.

3. The method according to claim 1, characterized in that, In step S3, the unified cardiac cycle phase coordinates include a systolic phase interval and a diastolic phase interval, wherein the systolic phase interval is determined by the first heart sound to the second heart sound, and the diastolic phase interval is determined by the second heart sound to the next first heart sound.

4. The method according to claim 1, characterized in that, In step S4, the acoustic event descriptor includes at least one of the following: first heart sound intensity, second heart sound intensity, systolic murmur energy, diastolic murmur energy, murmur onset phase, murmur termination phase, murmur duration ratio, peak phase, spectral centroid, spectral bandwidth, spectral entropy, band energy ratio, period consistency, and cross-period energy variation coefficient; in step S5, the node features... Used to represent the acoustic event characteristics of each auscultation zone within different cardiac cycle phase intervals and different frequency bands; the edge weights Used to represent at least one of the following: energy attenuation relationship, time delay relationship, spectral similarity, phase consistency, conduction direction, and periodic stability between different auscultation zones; in step S6, the valve sound source-auscultation zone propagation relationship is represented as a propagation matrix. : in, This represents the valve sound source-stethoscope area propagation matrix. Indicates the first Typical propagation vectors of a valvular sound source across multiple standard cardiac auscultation areas. The number of valve sound sources is indicated; the valve sound sources include at least one of the aortic valve sound source, pulmonary valve sound source, mitral valve sound source, and tricuspid valve sound source.

5. The method according to claim 4, characterized in that, In step S6, the estimation of the valve sound source contribution includes: 1) Based on the node characteristics of the cardiac sound propagation topology map of the multi-auscultation area. Extract the observed acoustic distribution vectors on multiple auscultation areas of the subject. ; 2) Based on the edge weights of the multi-auscultation area cardiac sound propagation topology map Extract constraints on sound propagation relationships across auscultation areas; 3) Based on the propagation matrix The observed acoustic distribution vector and the edge weights Estimate the valve sound source contribution vector ; in, This indicates the observed acoustic distribution across multiple auscultation areas. This indicates the sound propagation relationship between different auscultation areas. This indicates the degree to which different valvular sound sources contribute to the observed acoustic distribution and sound propagation relationship.

6. The method according to claim 1, characterized in that, In step S6, the valve sound source contribution estimation is implemented using one or more of the following: non-negativity constraint, sparse constraint, template similarity calculation, Bayesian inversion, graph matching, or weighted least squares; in step S7, the valvular disease acoustic template library is generated from historical heart sound samples with valvular disease annotation information; the valvular disease annotation information comes from echocardiogram examination results, clinical expert confirmation results, medical record results, or a combination of the above results.

7. The method according to claim 1, characterized in that, The acoustic template library for valvular diseases is generated as follows: 1) Obtain historical heart sound samples with valvular heart disease annotation information; 2) Perform steps S1 to S6 for each historical heart sound sample to obtain the corresponding historical multi-auscultation area heart sound propagation topology map and historical valve sound source contribution estimation results; 3) Group the historical heart sound samples according to the valvular heart disease labeling information; 4) Generate acoustic templates for valvular heart disease of the corresponding category based on the node features, edge weights and valve sound source contribution estimation results of historical heart sound samples within the same group.

8. The method according to claim 7, characterized in that, Each type of valvular disease acoustic template includes at least one of the following: typical node features, typical edge weights, typical valvular sound source contribution patterns, and template quality parameters for the corresponding valvular disease type. Specifically, the typical node features represent the distribution of acoustic events in different auscultation zones, different cardiac cycle phase intervals, and different frequency bands for this type of valvular disease; the typical edge weights represent the sound propagation relationship between different auscultation zones for this type of valvular disease; and the typical valvular sound source contribution patterns represent the valvular sound source contribution characteristics corresponding to this type of valvular disease.

9. The method according to claim 1, characterized in that, In step S7, the evidence score We obtain it from the following formula: in, Indicates the subject and the first Evidence score between acoustic templates of valvular disease; Indicates the node feature matching score; This represents the edge weight matching score; This indicates the valve sound source contribution matching score; This indicates the heart sound quality matching score; This represents the weighting coefficient of the corresponding matching score.

10. The method according to claim 1, characterized in that, In step S8, the patient-level prediagnostic result for valvular heart disease includes at least one of the following: normal indication, suspected valvular abnormality indication, suspected regurgitation-type valvular heart disease indication, suspected stenosis-type valvular heart disease indication, valvular heart disease risk level indication, and suggestion for further echocardiography. In step S8, while outputting the patient-level prediagnostic result for valvular heart disease, interpretable evidence is also output. The interpretable evidence includes at least one of the following: major abnormal auscultation area, abnormal cardiac cycle phase, major abnormal frequency band, suspected valvular sound source, sound propagation topology matching score, and heart sound quality reliability.

Images