Heart failure disease detection auxiliary system

Abstract

The invention relates to the technical field of medical detection information, in particular to a heart failure disease detection auxiliary system which comprises an acquisition module, an analysis module, a preliminary detection module, a heart sound and electrocardio learning model, a special detection acquisition module, a heart failure staging module and a treatment verification module. The acquisition module acquires electrocardio and heart sound data in real time by using a portable heart sound and electrocardio detector; the analysis module compares the data with a heart failure feature library and extracts heart failure features; the preliminary detection module sets a threshold value, evaluates a primary risk value by using a heart sound and electrocardio learning model, and sends an early warning to a doctor when the primary risk value exceeds the standard; the special detection acquisition module collects patient information and hospital detection results; the heart failure staging module judges a heart failure stage in combination with multi-source data and assists a doctor in formulating a treatment scheme; the treatment verification module establishes a patient personal deep learning model and predicts the effect and prognosis of a treatment scheme for doctors to adjust the scheme. According to the invention, auxiliary information can be provided for doctors in a targeted manner according to individual specific conditions of patients.

Description

Technical Field

[0001] The present invention relates to the field of medical detection information technology, and in particular to a heart failure disease detection auxiliary system. Background Art

[0002] Heart failure, a manifestation of severely impaired cardiac function, is caused by a variety of structural or functional cardiac diseases, leading to decreased ventricular filling or ejection capacity, making it unable to meet the metabolic needs of the body's tissues. Clinically, this often manifests as pulmonary or systemic congestion, as well as inadequate blood perfusion to organs and tissues. In heart failure, the uncoordinated contraction and relaxation of the heart can affect the production of heart sounds. Therefore, detecting heart sound signals can be a convenient and noninvasive means of assessing changes in myocardial contractility and peripheral resistance. Altered heart sound characteristics can be considered a noninvasive indicator for detecting abnormalities in the cardiac mechanical dynamics caused by left ventricular systolic dysfunction and systemic circulation disorders. In other words, heart sound characteristics can serve as a noninvasive indicator for monitoring changes in cardiac hemodynamics during the development and progression of chronic heart failure. However, manually analyzing each patient's heart sound and electrocardiogram (ECG) data is tedious and time-consuming, reducing the efficiency of medical services. Therefore, heart failure disease detection assistance systems play an important role in assisting doctors in diagnosing heart failure.

[0003] Existing heart failure detection assistance systems analyze patients' ECG and heart sound test results, leveraging big data and machine learning to quickly identify characteristic changes or patterns associated with heart failure. These systems provide physicians with tools to process and analyze large amounts of ECG and heart sound data, improving diagnostic efficiency and accuracy. However, existing heart failure detection assistance systems may fail to fully consider individual patient differences and specific conditions when providing auxiliary information, resulting in overly general recommendations.

[0004] In summary, how to solve the problem that the existing heart failure disease detection assistance system may fail to fully consider the individual differences and specific conditions of patients when providing auxiliary information, which may lead to relatively general suggestions, has become a difficult problem that needs to be solved urgently in this field. Therefore, it is necessary to propose a heart failure disease detection assistance system. Summary of the Invention

[0005] To solve the above problems, the present invention provides a heart failure disease detection assistance system to assist doctors in quickly conducting in-depth analysis and interpretation of heart failure-related data, and to provide doctors with personalized and targeted auxiliary diagnosis suggestions.

[0006] In order to achieve the above-mentioned purpose, the technical solution of the present invention is as follows: a heart failure disease detection auxiliary system, including an acquisition module, an analysis module, a preliminary detection module, a heart sound and electrocardiogram learning model, a special detection acquisition module, a heart failure staging module, a doctor's personal client and a treatment verification module.

[0007] The acquisition module includes a portable heart sound and electrocardiogram detector, and is used to use the portable heart sound and electrocardiogram detector to acquire the patient's electrocardiogram data and heart sound data in real time.

[0008] The analysis module is used to establish a heart failure feature database by collecting doctors' clinical diagnosis and treatment data, compare the electrocardiogram data and heart sound data with the heart failure feature database, and extract the murmurs, extra heart sounds and electrocardiogram change characteristics caused by heart failure.

[0009] The preliminary detection module is used to set thresholds and input the murmurs, extra heart sounds and ECG change characteristics caused by heart failure into the heart sound and ECG learning model for learning to generate learning results; the learning results include primary risk values.

[0010] When the primary risk value exceeds the threshold, the preliminary detection module sends a suggestion message to the doctor's personal client.

[0011] The special test collection module is used to collect patients' personal information and the test results of special tests taken by patients at the hospital.

[0012] The heart failure staging module is used to use the heart failure staging algorithm to evaluate heart failure using the test results of special tests, electrocardiogram data, heart sound data and primary risk values, generate heart failure assessment results, and send the heart failure assessment results to the doctor's personal client. The doctor uses the heart failure assessment results as a reference to formulate a treatment plan for the patient.

[0013] The treatment verification module is used to establish a patient-specific deep learning model using the patient's personal information, primary risk value, and special test results, and input the patient's treatment plan into the patient's personal deep learning model for learning and prediction, generate prognostic information, and send the prognostic information to the doctor's personal client. The doctor adjusts the patient's treatment plan based on the prognostic information.

[0014] Furthermore, in the analysis module, the electrocardiogram change characteristics of heart failure include heart rate variability and QRS complex morphology changes.

[0015] Furthermore, in the preliminary detection module, the method for generating learning results by the heart sound and electrocardiogram learning model includes the following steps:

[0016] Step 1, feature extraction: extract heart sound features from murmurs and extra heart sounds caused by heart failure; heart sound features include the type, intensity and duration of murmurs and extra heart sounds.

[0017] Step 2: Model training and learning: Input the heart sound characteristics and ECG change characteristics of heart failure into the heart sound and ECG learning model, and use the deep learning algorithm to train the heart sound and ECG learning model to obtain a trained heart sound and ECG learning model.

[0018] Step 3: Generate a primary risk value: Input the newly collected murmurs and extra heart sounds caused by heart failure and the electrocardiogram change characteristics of heart failure into the trained heart sound and electrocardiogram learning model to obtain a primary risk value.

[0019] Step 4: Risk value assessment and warning: compare the primary risk value with the preset threshold; when the primary risk value exceeds the threshold, the preliminary detection module sends a recommendation message to the doctor's personal client.

[0020] Furthermore, in the preliminary testing module, the recommendation information includes a recommendation to notify the patient to go to the hospital for special testing.

[0021] Furthermore, in the special test acquisition module, the test results of the special test include blood biochemical indicators, echocardiography results and cardiac magnetic resonance imaging results.

[0022] Furthermore, in the preliminary detection module, the primary risk value includes a heart sound characteristic risk score and an electrocardiogram characteristic risk score.

[0023] Furthermore, in the heart failure staging module, the heart failure staging algorithm includes the following steps:

[0024] S101, data integration: Integrate blood biochemical indicators, echocardiography results, cardiac magnetic resonance imaging results, electrocardiogram data, heart sound data, heart sound characteristic risk scores and electrocardiogram characteristic risk scores to generate a patient heart failure assessment dataset.

[0025] S102, feature selection and processing: Extract BNP level, NT-proBNP level, LVEF and myocardial fibrosis degree features in cardiac magnetic resonance imaging from the patient heart failure assessment dataset, perform standardization processing, and obtain standardized BNP level, NT-proBNP level, LVEF and myocardial fibrosis degree features in cardiac magnetic resonance imaging.

[0026] S103, Feature selection and weight assignment: Based on medical research and clinical experience, weights are assigned to the features of BNP level, NT-proBNP level, LVEF, and myocardial fibrosis degree in cardiac magnetic resonance imaging.

[0027] S104, calculation of comprehensive score: The BNP level, NT-proBNP level, LVEF, and myocardial fibrosis characteristics on cardiac magnetic resonance imaging, all of which were assigned different weights, were used to calculate the comprehensive heart failure score using the following formula:

[0028]

[0029] Among them, ω i is the weight of the i-th feature; f iis the value of the i-th feature, which is the value of the i-th feature after normalization; n is the total number of features.

[0030] S105, Heart failure assessment: Compare the comprehensive heart failure score with the preset staging threshold, and classify heart failure into different levels according to the New York Heart Association (NYHA) classification method to obtain the heart failure assessment results; the heart failure assessment results include NYHA class I, NYHA class II, NYHA class III and NYHA class IV.

[0031] S106, result output and feedback: Send the heart failure assessment results to the doctor's personal client.

[0032] Furthermore, in the treatment verification module, the patient's personal information includes age, gender, height, weight, blood pressure and medical history.

[0033] Furthermore, in the treatment verification module, the patient's personal deep learning model performs learning and prediction including the following steps:

[0034] S201, data integration: integrating the patient's personal information, primary risk value, special test results, and treatment plan into a patient's personal information data set.

[0035] S202, model construction: Based on the integrated patient personal information dataset, input it into a blank deep learning model to build a patient personal deep learning model.

[0036] S203, model training: using a deep learning algorithm to train the patient's personal deep learning model to obtain a trained patient's personal deep learning model.

[0037] S204, prediction generation: input the patient's current treatment plan into the trained patient-specific deep learning model to generate prognostic information.

[0038] Furthermore, prognostic information includes sequelae, side effects, treatment efficacy, and recurrence risk.

[0039] The above scheme has the following beneficial effects:

[0040] 1. The present invention realizes the comprehensive collection and analysis of the patient's electrocardiogram data and heart sound data by integrating multiple functional modules such as the acquisition module, analysis module, and preliminary detection module. It can quickly and accurately extract the characteristic changes related to heart failure, providing doctors with a more detailed and accurate basis for the diagnosis of heart failure, thereby improving the efficiency and accuracy of heart failure diagnosis.

[0041] 2. This invention utilizes a heart sound and electrocardiogram learning model and deep learning algorithms to conduct deep learning and intelligent analysis of a patient's heart failure data. It automatically generates a primary risk score and sends timely advisory information to the doctor's personal client when the risk score exceeds a preset threshold. This intelligent early warning mechanism not only helps doctors identify potential heart failure risks in a timely manner, but also buys patients valuable treatment time, reducing the risk and harm of heart failure attacks.

[0042] 3. The present invention also achieves personalized analysis of the patient's heart failure condition and optimization of the treatment plan through the heart failure staging module and the treatment verification module. The heart failure staging module can comprehensively consider the patient's multiple test results and risk factors to generate detailed heart failure assessment results, providing strong support for doctors to formulate personalized treatment plans. The treatment verification module can use the patient's personal deep learning model to predict and evaluate the treatment plan, and generate prognostic information including sequelae, side effects, treatment effects, and recurrence risks, providing a scientific basis for doctors to adjust and optimize the treatment plan, thereby improving the pertinence and effectiveness of heart failure treatment.

[0043] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a structural diagram of the heart failure disease detection auxiliary system of the present invention.

[0045] Figure 2 Schematic diagram of the steps for learning and predicting the patient's personal deep learning model in the heart failure disease detection assistance system of the present invention. DETAILED DESCRIPTION

[0046] The following is further described in detail through specific implementation methods:

[0047] As attached Figure 1-Figure 2 Shown: A heart failure disease detection auxiliary system, including an acquisition module, an analysis module, a preliminary detection module, a heart sound and electrocardiogram learning model, a special detection acquisition module, a heart failure staging module, a doctor's personal client and a treatment verification module.

[0048] Among them, the acquisition module is mainly used to collect ECG data and heart sound data in real time; the analysis module is mainly used to extract heart failure-related features from ECG data and heart sound data; the preliminary detection module is mainly used to evaluate the primary risk value, and when the primary risk value exceeds the threshold, it sends an early warning signal to the doctor's personal client; the special detection acquisition module is mainly used to collect patient information and hospital special test results; the heart failure staging module is mainly used to determine the heart failure stage and assist doctors in formulating treatment plans; the treatment verification module is mainly used to establish a patient's personal deep learning model, predict the prognosis of the treatment plan, and the doctor adjusts the treatment plan based on the prognosis information.

[0049] The following is a detailed explanation of the functions of each module:

[0050] The acquisition module includes a portable heart sound and electrocardiogram detector, and is used to use the portable heart sound and electrocardiogram detector to collect the patient's electrocardiogram data and heart sound data in real time.

[0051] Specifically, first, the doctor fixes the heart sound sensor and electrode sheets on the portable heart sound electrocardiogram detector on the chest wall near the patient's heart and on the patient's limbs and chest; in this embodiment, the heart sound sensor used is the Thinklabs ONE heart sound sensor; the patient's electrocardiogram data and heart sound data are collected in real time, and uploaded to the acquisition module of the heart failure disease detection auxiliary system of the present invention through 5G communication technology. Compared with the traditional hospital large-scale equipment detection, the use of the portable heart sound electrocardiogram detector realizes non-invasive detection and avoids the pain and discomfort that traditional detection methods may bring to patients. Secondly, the use of the portable heart sound electrocardiogram detector makes the patient detection process very convenient and can be tested at home without waiting for a long time or performing complicated examinations, which greatly saves time and energy.

[0052] The analysis module is used to establish a heart failure feature database by collecting doctors' clinical diagnosis and treatment data, compare the electrocardiogram data and heart sound data with the heart failure feature database, and extract the murmurs, extra heart sounds and electrocardiogram change characteristics of heart failure caused by heart failure; the electrocardiogram change characteristics of heart failure include heart rate variability and changes in QRS wave group morphology.

[0053] The preliminary detection module sets thresholds and inputs the characteristics of heart failure-related murmurs, extra heart sounds, and ECG changes into the heart sound and ECG learning model for learning, generating learning results. The learning results include a preliminary risk value. The preliminary risk value includes a risk score for heart sound characteristics and an ECG characteristic risk score. When the preliminary risk value exceeds the threshold, the preliminary detection module sends a recommendation message to the doctor's personal client.

[0054] In the preliminary detection module, the method for generating learning results by the heart sound and electrocardiogram learning model includes the following steps:

[0055] Step 1, feature extraction: extract heart sound features from murmurs and extra heart sounds caused by heart failure; heart sound features include the type, intensity and duration of murmurs and extra heart sounds.

[0056] Step 2: Model training and learning: Input the heart sound characteristics and ECG change characteristics of heart failure into the heart sound and ECG learning model, and use the deep learning algorithm to train the heart sound and ECG learning model to obtain a trained heart sound and ECG learning model.

[0057] Step 3: Generate a primary risk value: Input the newly collected murmurs and extra heart sounds caused by heart failure and the electrocardiogram change characteristics of heart failure into the trained heart sound and electrocardiogram learning model to obtain a primary risk value.

[0058] Step 4: Risk value assessment and early warning: Compare the primary risk value with the preset threshold; when the primary risk value exceeds the threshold, the preliminary detection module sends a recommendation message to the doctor's personal client, including a suggestion to notify the patient to go to the hospital for special testing.

[0059] Specifically, for example, when a patient feels palpitations, shortness of breath, and occasionally has symptoms of difficulty breathing, the patient uses a portable phonocardiogram (ECG) detector to perform self-examination at home to collect ECG data and heart sound data; the ECG data and heart sound data are sent by the portable phonocardiogram (ECG) detector to the acquisition module of the heart failure disease detection auxiliary system via 5G technology.

[0060] The analysis module compares the collected ECG data and heart sound data with the heart failure feature database, and extracts murmurs caused by heart failure (such as the third and fourth heart sounds), extra heart sounds (such as gallop rhythm), and ECG change characteristics of heart failure (such as increased heart rate variability and changes in QRS wave morphology).

[0061] The preliminary detection module first sets a threshold, which is trained based on the test data of a large number of heart failure patients and is used to distinguish between patients at high risk and low risk of heart failure. Then, the preliminary detection module inputs the extracted heart failure features into the trained heart sound and electrocardiogram learning model, and the model calculates the primary risk value based on these features.

[0062] When the primary risk value exceeds the preset threshold, it means that the patient has a higher risk of heart failure. The preliminary detection module will then send the recommendation information to the doctor's personal client and display "It is recommended to notify the patient to go to the hospital for special testing" on the doctor's personal client.

[0063] After receiving the recommendation information, the doctor can also immediately review the patient's test data to determine whether further special tests are needed to confirm the diagnosis. If necessary, the doctor can contact the patient and arrange for him to go to the hospital for a detailed examination.

[0064] This will help patients achieve convenient and non-invasive self-testing, provide timely warnings to high-risk patients, and provide doctors with powerful auxiliary diagnostic tools to more effectively manage heart failure.

[0065] The special test collection module is used to collect patients' personal information and the test results of special tests taken by patients at the hospital.

[0066] In the special test acquisition module, the test results of special tests include blood biochemical indicators, echocardiography results and cardiac magnetic resonance imaging results.

[0067] The heart failure staging module is used to use the heart failure staging algorithm to evaluate heart failure using the test results of special tests, electrocardiogram data, heart sound data and primary risk values, generate heart failure assessment results, and send the heart failure assessment results to the doctor's personal client. The doctor uses the heart failure assessment results as a reference to formulate a treatment plan for the patient.

[0068] In the heart failure staging module, the heart failure staging algorithm includes the following steps:

[0069] S101, data integration: Integrate blood biochemical indicators, echocardiography results, cardiac magnetic resonance imaging results, electrocardiogram data, heart sound data, heart sound characteristic risk scores and electrocardiogram characteristic risk scores to generate a patient heart failure assessment dataset.

[0070] S102, feature selection and processing: Extract BNP level, NT-proBNP level, LVEF and myocardial fibrosis degree features in cardiac magnetic resonance imaging from the patient heart failure assessment dataset, perform standardization processing, and obtain standardized BNP level, NT-proBNP level, LVEF and myocardial fibrosis degree features in cardiac magnetic resonance imaging.

[0071] S103, Feature selection and weight assignment: Based on medical research and clinical experience, weights are assigned to the features of BNP level, NT-proBNP level, LVEF, and myocardial fibrosis degree in cardiac magnetic resonance imaging.

[0072] S104, calculation of comprehensive score: The BNP level, NT-proBNP level, LVEF, and myocardial fibrosis characteristics on cardiac magnetic resonance imaging, all of which were assigned different weights, were used to calculate the comprehensive heart failure score using the following formula:

[0073]

[0074] Among them, ω i is the weight of the i-th feature; f i is the value of the i-th feature, which is the value of the i-th feature after normalization; n is the total number of features.

[0075] S105, heart failure assessment: pre-set the staging threshold, compare the heart failure comprehensive score with the preset staging threshold, divide heart failure into different levels according to the New York Heart Association NYHA classification method, and obtain the heart failure assessment results; the heart failure assessment results include NYHA class I, NYHA class II, NYHA class III and NYHA class IV.

[0076] S106, result output and feedback: Send the heart failure assessment results to the doctor's personal client.

[0077] Specifically, for example, before heart failure staging, the heart failure staging module standardizes the patient's blood biochemical indicators, echocardiography results, cardiac magnetic resonance imaging results, electrocardiogram data, heart sound data, heart sound characteristic risk score and electrocardiogram characteristic risk score to ensure the consistency and comparability of the data.

[0078] The heart failure staging module then selects features that have a significant impact on heart failure staging based on medical research and clinical experience, and assigns corresponding weights to each feature. For example, the level of BNP (brain natriuretic peptide) in the blood biochemical index may be closely related to the severity of heart failure and therefore can be given a higher weight. The weight assignment may be based on expert scoring, statistical analysis, or automatic determination by machine learning algorithms.

[0079] After the weight distribution of each feature is completed, the staging threshold is pre-set manually; the heart failure staging module inputs each feature into the heart failure comprehensive score calculation formula to obtain the score of a single feature, and then calculates the score of a single feature through weighted average calculation to obtain the comprehensive score of heart failure. The comprehensive score of heart failure is compared with the preset staging threshold to divide heart failure into different stages or levels.

[0080] For example, several thresholds can be set to classify heart failure into NYHA class I, NYHA class II, NYHA class III, and NYHA class IV. The staging thresholds are derived based on statistical analysis of large-scale clinical data.

[0081] The results of the heart failure assessment are presented to physicians in an easy-to-understand manner, along with corresponding explanations and recommendations. For example, the specific heart failure stage, possible pathophysiological mechanisms, and treatment recommendations based on the stage can be provided.

[0082] Regularly collect new clinical data to validate and optimize the heart failure staging algorithm. By comparing the algorithm's predictions with actual clinical data, we adjust parameters such as feature selection, weight allocation, and staging threshold setting to improve the algorithm's accuracy and reliability.

[0083] The treatment verification module uses the patient's personal information, primary risk score, and special test results to build a personalized deep learning model. The patient's treatment plan is then fed into the model for learning and prediction, generating a prognosis. This prognosis is then sent to the doctor's personal client, who then adjusts the patient's treatment plan based on the prognosis. The patient's personal information includes age, gender, height, weight, blood pressure, and medical history.

[0084] The patient's personal deep learning model learns and predicts by following these steps:

[0085] S201, data integration: integrating the patient's personal information, primary risk value, special test results, and treatment plan into a patient's personal information data set.

[0086] S202, model construction: Based on the integrated patient personal information dataset, input it into a blank deep learning model to build a patient personal deep learning model.

[0087] S203, model training: using a deep learning algorithm to train the patient's personal deep learning model to obtain a trained patient's personal deep learning model.

[0088] S204, prediction generation: input the patient's current treatment plan into the trained patient-specific deep learning model to generate prognostic information.

[0089] Prognostic information includes sequelae, side effects, treatment efficacy, and risk of recurrence.

[0090] Specifically, for example, in the treatment verification stage, the doctor formulates a preliminary treatment plan based on the heart failure assessment results and inputs the preliminary treatment plan into the treatment verification module.

[0091] The treatment verification module first integrates the patient's personal information (such as age, gender, height, weight, blood pressure and medical history), primary risk value, special test results (such as blood biochemical indicators, echocardiography results and cardiac magnetic resonance imaging results) and the preliminary treatment plan to form a comprehensive patient personal information data set.

[0092] Next, the treatment verification module uses this patient's personal information dataset to transform the blank deep learning model into a personalized deep learning model. This model is then personalized based on the patient's specific circumstances. This model is then trained using a deep learning algorithm to accurately predict the patient's prognosis after receiving the current treatment plan.

[0093] After model training is complete, doctors can input the patient's current treatment plan into the trained deep learning model. The model will then generate a detailed prognostic report based on the patient's personal information, historical test data, and treatment plan. This report includes possible sequelae, side effects, treatment effectiveness, and recurrence risk.

[0094] After receiving the prognostic information report, doctors can gain a more comprehensive understanding of the patient's likely outcome after accepting the current treatment plan. Based on the patient's actual situation and the recommendations in the prognostic information report, they can make necessary adjustments and optimizations to the treatment plan. This not only improves the effectiveness of treatment but also reduces the risks and discomfort that patients may experience during treatment.

[0095] Furthermore, the treatment verification module regularly collects new clinical data to validate and optimize the patient's personalized deep learning model. By comparing the patient's personalized deep learning model's predictions with the actual clinical situation, doctors can further adjust the patient's personalized deep learning model's parameters and structure to improve its accuracy and reliability. As clinical data continues to accumulate and optimize, the patient's personalized deep learning model will be able to more accurately predict patient prognoses, providing doctors with a more powerful decision-making tool.

[0096] The heart failure disease detection auxiliary system of the present invention realizes personalized analysis of the patient's heart failure condition and optimization of the treatment plan through the heart failure staging module and the treatment verification module. The heart failure staging module can comprehensively consider the patient's multiple test results and risk factors to generate detailed heart failure assessment results, providing strong support for doctors to formulate personalized treatment plans. The treatment verification module can use the patient's personal deep learning model to predict and evaluate the treatment plan, generate prognostic information including sequelae, side effects, treatment effects and recurrence risks, and provide scientific advice to doctors for them to adjust and optimize the treatment plan, thereby improving the pertinence and effectiveness of heart failure treatment.

[0097] In another embodiment, the specific implementation of the data acquisition module and data verification:

[0098] This embodiment uses a portable heart sound and electrocardiogram detector (model: THinklabs ONE) with a sampling frequency of 4 kHz (heart sound) and 500 Hz (electrocardiogram), supporting 12-lead electrocardiogram to ensure data accuracy.

[0099] This embodiment was tested on 100 heart failure patients and 50 healthy controls. The device's detection sensitivity was 92% (heart murmur identification) and 89% (QRS complex abnormality detection), and the consistency of the results with traditional hospital equipment (such as the GE Vivid E95 ultrasound) was 93%.

[0100] Construction of feature database for analysis module:

[0101] The heart failure feature database used in this example contains annotated data of 2,000 clinically diagnosed heart failure patients, covering:

[0102] Heart sound characteristics: The detection rate of the third heart sound is 82%, the detection rate of the fourth heart sound is 68%, and the systolic murmur accounts for 75%.

[0103] ECG characteristics: QRS complex width >120ms (accounting for 63%), heart rate variability (SDNN <50ms, accounting for 78%).

[0104] Deep learning model training for the preliminary detection module:

[0105] Model architecture: A phonocardiogram (ECG) learning model is used, with the input being the phonocardiogram signal (4-second segment) and the ECG waveform (10-second segment).

[0106] Training data: 1,500 patients (including 800 heart failure patients) from a multicenter clinical trial, with a validation set AUC of 0.94, a sensitivity of 90%, and a specificity of 89%.

[0107] Threshold setting: The primary risk value threshold is set to 0.75. When the heart sound and electrocardiogram learning model outputs a risk value greater than 0.75, an early warning is triggered.

[0108] Key indicators of special detection and acquisition modules:

[0109] Blood biochemical indicators: BNP>400pg / mL (sensitivity 85%), NT-proBNP>1500pg / mL (sensitivity 90%).

[0110] Echocardiography: LVEF < 40% was used as the criterion for systolic dysfunction (specificity 95%).

[0111] Cardiac MRI: The degree of myocardial fibrosis (late gadolinium enhancement accounting for >15% of the myocardial area) was significantly associated with heart failure mortality (HR=2.3, p<0.01).

[0112] In this embodiment, the weight distribution and calculation example of the heart failure staging algorithm is as follows:

[0113] Feature weights: BPN level (0.4), NT-proBNP (0.3), LVEF (0.2), and myocardial fibrosis (0.1).

[0114] Step 1: Feature extraction

[0115] Heart sound feature extraction (based on short-time Fourier transform STFT):

[0116]

[0117] Among them, x(τ) is the heart sound signal, w(τ) is the Hamming window function, and the frequency domain energy ratio of S3 / S4 is extracted.

[0118] Calculation of ECG QRS complex width: QRS width = t end -t start (Unit: ms)

[0119] The QRS onset and endpoint were detected by the differential threshold method.

[0120] Step 2: Model training and risk value generation

[0121] ECG and heart sound learning model output:

[0122]

[0123] Where σ is the sigmoid function, which outputs the probability of abnormal heart sound (0-1).

[0124] Heart sound and ECG learning model output:

[0125] h t =LSTM(x ECG,t ,h t-1 )

[0126] y ECG =σ(W·h T +b)

[0127] Final time step h T The hidden state of is output through the fully connected layer to output the probability of ECG abnormality.

[0128] Primary VaR fusion:

[0129] Primary risk value = 0.6 × yaudio + 0.4 × yECG

[0130] The weights are set based on SHAP values.

[0131] If the patient's heart sound and ECG learning model outputs yaudio = 0.85 and yECG = 0.72, then:

[0132] Primary risk value = 0.6 × 0.85 + 0.4 × 0.72 = 0.798 (exceeding the threshold of 0.75, triggering an early warning) In this embodiment, the heart failure comprehensive scoring formula in the heart failure staging module is explained in detail:

[0133] Step 1: Data standardization

[0134] Z-score standardization (taking BNP as an example):

[0135]

[0136] Among them, μ BNP =300 pg / mL, σBNP = 150 pg / mL (based on training set statistics)

[0137] LVEF normalization (converted to a 0-1 score):

[0138]

[0139] Step 2: Weight allocation and comprehensive score calculation. Table 1 is the basis for weight allocation (based on multivariate regression analysis).

[0140] feature Regression coefficient β <![CDATA[Normalized weight w i > BNP levels 0.45 0.4 NT-proBNP level 0.35 0.3 LVEF 0.25 0.2 Degree of myocardial fibrosis 0.15 0.1

[0141] Table 1

[0142]

[0143] Specific expansion:

[0144] Score = 0.4 × BNP_std + 0.3 × NT-proBNP_std + 0.2 × LVEF_std + 0.1

[0145] Patient data:

[0146] BNP=600pg / mL→BNP_std=(600-300) / 150=2.0

[0147] NT-proBNP=2000pg / mL→NT-proBNP_std=(2000-1000) / 500=2.0

[0148] LVEF=35%→LVEF_std=1-35 / 100=0.65

[0149] Myocardial fibrosis degree = 20% → Fibrosis_std = 20 / 15 = 1.33 (assuming σ = 15%)

[0150] Substitute into the heart failure comprehensive scoring formula:

[0151] Rating = 0.4 × 2.0 + 0.3 × 2.0 + 0.2 × 0.65 + 0.1 × 1.33 = 0.8 + 0.6 + 0.13 + 0.133 = 1.663

[0152] Comparison period threshold (preset):

[0153] NYHA class IV (score>0.9) is considered severe heart failure.

[0154] Patient-specific deep learning model:

[0155] Input encoding: Numerical features (age, BNP, etc.) are embedded as vector E num Treatment options (such as drug dosage) are coded as E treatment .

[0156] Self-attention calculation:

[0157]

[0158] Among them, Q, K, V are query, key, and value matrices, d k For dimension.

[0159] Prognostic probability output:

[0160] p(side effect)=σ(W p ·h [CLS] +b p )

[0161] Among them, h [CLS] is the hidden state of the classification flag, W p is the weight matrix.

[0162] Example calculation:

[0163] Patient input: age = 70 years, BNP = 800 pg / mL, LVEF = 30%, medication used: sacubitril-valsartan 200 mg bid.

[0164] Model predictions:

[0165] After Transformer encoding, h [CLS] =[0.2,-0.5,1.3,...].

[0166] Side effect probability calculation:

[0167]

[0168] The doctor will adjust the dose or change the medication accordingly.

[0169] Interpretability of the comprehensive heart failure scoring formula: The high weighting of BNP and NT-proBNP in the formula reflects their central role as heart failure biomarkers. LVEF and myocardial fibrosis are given lower weightings but provide additional structural information. A score >0.9 requires urgent intervention (such as mechanical circulatory support), consistent with clinical guidelines.

[0170] Practicality of the prognostic prediction model: A side effect probability > 0.5 suggests a need for regimen adjustment, which reduced the hospitalization rate by 30% in clinical trials (p < 0.01).

[0171] Necessity of standardization: Different test indicators have different dimensions (e.g., BNP is measured in pg / mL and LVEF is measured in percentage). Z-score and normalization ensure fairness in scoring.

[0172] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will readily appreciate that other variations or modifications based on the above descriptions are possible. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A heart failure disease detection auxiliary system, characterized in that: It includes acquisition module, analysis module, preliminary detection module, heart sound and electrocardiogram learning model, special detection acquisition module, heart failure staging module, doctor's personal client and treatment verification module; The acquisition module includes a portable heart sound and electrocardiogram detector; the acquisition module is used to use the portable heart sound and electrocardiogram detector to collect the patient's electrocardiogram data and heart sound data in real time; The analysis module is used to collect clinical diagnosis and treatment data from doctors to establish a heart failure feature database, compare the ECG data and heart sound data with the heart failure feature database, and extract murmurs, extra heart sounds, and ECG change characteristics caused by heart failure; The preliminary detection module is used to set the threshold and input the murmurs, extra heart sounds and ECG change features caused by heart failure into the heart sound and ECG learning model for learning and generating learning results; Learning outcomes include primary risk values; When the primary risk value exceeds the threshold, the preliminary detection module sends a suggestion message to the doctor's personal client; Special test collection module, used to collect patients' personal information and test results of special tests taken by patients at the hospital; The heart failure staging module is used to evaluate heart failure using a heart failure staging algorithm, combining special test results, electrocardiogram data, heart sound data, and primary risk values. This module generates a heart failure assessment result and sends it to the doctor's personal client. The doctor uses the heart failure assessment result as a reference to formulate a treatment plan for the patient. The treatment verification module is used to establish a patient-specific deep learning model using the patient's personal information, primary risk value, and special test results, and input the patient's treatment plan into the patient's personal deep learning model for learning and prediction, generate prognostic information, and send the prognostic information to the doctor's personal client. The doctor adjusts the patient's treatment plan based on the prognostic information.

2. The heart failure disease detection auxiliary system according to claim 1, characterized in that: In the analysis module, the electrocardiogram change characteristics of heart failure include heart rate variability and QRS complex morphology changes.

3. The heart failure disease detection auxiliary system according to claim 2, characterized in that: In the preliminary detection module, the method for generating learning results by the heart sound and electrocardiogram learning model includes the following steps: Step 1: Feature extraction: Extract heart sound features from murmurs and extra heart sounds caused by heart failure; Heart sound characteristics include the type, intensity, and duration of murmurs and extra heart sounds; Step 2: Model training and learning: Input the heart sound features and ECG change features of heart failure into the heart sound and ECG learning model, and use the deep learning algorithm to train the heart sound and ECG learning model to obtain a trained heart sound and ECG learning model; Step 3: Generate a primary risk value: Input the newly collected murmurs and extra heart sounds caused by heart failure and the electrocardiogram change characteristics of heart failure into the trained heart sound and electrocardiogram learning model to obtain a primary risk value; Step 4: Risk value assessment and warning: compare the primary risk value with the preset threshold; when the primary risk value exceeds the threshold, the preliminary detection module sends a recommendation message to the doctor's personal client.

4. The heart failure disease detection auxiliary system according to claim 3, characterized in that: In the preliminary testing module, the recommendation information includes a recommendation to notify the patient to go to the hospital for special testing.

5. The heart failure disease detection auxiliary system according to claim 4, characterized in that: In the special test acquisition module, the test results of special tests include blood biochemical indicators, echocardiography results and cardiac magnetic resonance imaging results.

6. The heart failure disease detection auxiliary system according to claim 5, characterized in that: In the preliminary detection module, the primary risk value includes the heart sound characteristic risk score and the electrocardiogram characteristic risk score.

7. The heart failure disease detection auxiliary system according to claim 6, characterized in that: In the heart failure staging module, the heart failure staging algorithm includes the following steps: S101, Data Integration: Integrate blood biochemical indicators, echocardiography results, cardiac magnetic resonance imaging results, electrocardiogram data, heart sound data, heart sound characteristic risk scores, and electrocardiogram characteristic risk scores to generate a patient heart failure assessment dataset; S102, feature selection and processing: extracting BNP level, NT-proBNP level, LVEF, and myocardial fibrosis degree features in cardiac magnetic resonance imaging from the patient heart failure assessment dataset, performing standardization processing, and obtaining standardized BNP level, NT-proBNP level, LVEF, and myocardial fibrosis degree features in cardiac magnetic resonance imaging; S103, Feature Selection and Weight Assignment: Based on medical research and clinical experience, weights were assigned to the features of BNP level, NT-proBNP level, LVEF, and myocardial fibrosis degree in cardiac MRI; S104, calculation of comprehensive score: The BNP level, NT-proBNP level, LVEF, and myocardial fibrosis characteristics on cardiac magnetic resonance imaging, all of which were assigned different weights, were used to calculate the comprehensive heart failure score using the following formula: Among them, ω i is the weight of the i-th feature; f i is the value of the i-th feature, which is the value of the i-th feature after normalization; n is the total number of features; S105, heart failure assessment: pre-set staging thresholds, compare the heart failure comprehensive score with the pre-set staging thresholds, classify heart failure into different levels according to the New York Heart Association (NYHA) classification method, and obtain the heart failure assessment results; Heart failure assessment results include NYHA class I, NYHA class II, NYHA class III, and NYHA class IV; S106, result output and feedback: Send the heart failure assessment results to the doctor's personal client.

8. The heart failure disease detection auxiliary system according to claim 7, characterized in that: In the treatment verification module, the patient's personal information includes age, gender, height, weight, blood pressure and medical history.

9. The heart failure disease detection auxiliary system according to claim 8, characterized in that: In the treatment verification module, the patient's personal deep learning model learns and predicts the following steps: S201, data integration: integrating the patient's personal information, primary risk value, special test results, and treatment plan into a patient's personal information dataset; S202, model construction: Based on the integrated patient personal information dataset, input it into a blank deep learning model to build a patient personal deep learning model; S203, model training: using a deep learning algorithm to train the patient's personal deep learning model to obtain a trained patient's personal deep learning model; S204, prediction generation: input the patient's current treatment plan into the trained patient-specific deep learning model to generate prognostic information.

10. The heart failure disease detection auxiliary system according to claim 9, characterized in that: Prognostic information includes sequelae, side effects, treatment efficacy, and risk of recurrence.

Images