Heart failure nuclear magnetic image recognition analysis method based on multi-modal deep learning

Abstract

The invention discloses a heart failure nuclear magnetic image recognition and analysis method based on multi-modal deep learning, and relates to the technical field of medical treatment. In the first stage, a first discrimination coefficient and a second discrimination coefficient are calculated through preprocessing and multi-dimensional data extraction of a nuclear magnetic image, the heart failure type is accurately judged, and the compensation mechanism activation state of a patient is evaluated according to the first discrimination coefficient, so that the heart failure type is further clarified. And integrating the multi-modal data to fit and output a comprehensive discrimination index Zx, and realizing quantitative evaluation of the mixed heart failure degree. In the S4 stage, the positive likelihood ratio LR is calculated through data verification of a large number of patients and normal control individuals, and the efficiency of the diagnosis method is checked.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and specifically to a method for identifying and analyzing heart failure nuclear magnetic resonance images based on multimodal deep learning. Background Art

[0002] In the field of medical image analysis, multimodal deep learning is gradually becoming the core technical force driving the accurate diagnosis of diseases. By integrating various types of data, such as images, clinical indicators, laboratory test results, etc., multimodal deep learning provides a comprehensive and in-depth perspective for the analysis of complex diseases. In the category of cardiovascular diseases, heart failure (HF) is a clinical syndrome caused by the dysfunction of the heart's pumping function, which cannot meet the body's needs for oxygen and nutrients. Its diagnosis and treatment have always faced many challenges. The method for identifying and analyzing heart failure nuclear magnetic resonance images based on multimodal deep learning focuses on using nuclear magnetic resonance images of heart failure obtained by magnetic resonance imaging (MRI), combined with the patient's clinical information and laboratory test data, aiming to more accurately analyze key factors such as the type, degree, and compensatory mechanism of the patient's heart failure, and provide strong support for clinical decision-making.

[0003] Traditional methods for diagnosing heart failure have certain limitations. In terms of clinical symptom judgment, the symptoms of heart failure include dyspnea, pulmonary edema, etc. However, not all patients with dyspnea, pulmonary edema, etc. are caused by heart failure. Therefore, when initially inquiring about the patient, some conventional medical means are needed to initially judge the patient's condition. Moreover, a single imaging examination, such as a common echocardiogram, is difficult to comprehensively present the subtle structure and functional changes of the heart, and it is difficult to accurately identify some early or occult heart failures. These methods cannot comprehensively utilize multi-source data and are difficult to achieve a comprehensive and accurate diagnosis and analysis of heart failure patients.

[0004] The emergence of these current drawbacks mainly stems from the isolated application of different diagnostic methods and the neglect of the fusion analysis of multimodal data. Clinicians often make judgments based on limited information, resulting in inaccurate diagnoses. This not only delays the treatment of patients, may further deteriorate the condition, increase the patient's pain and medical costs, but also may lead to unnecessary treatments, bringing additional economic burdens and potential risks of drug side effects to the patients. At the disease management level, inaccurate and untimely diagnoses make the tracking of the condition and prognosis assessment of heart failure patients lack reliability, seriously affecting the overall medical quality and the quality of life of patients. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a method for identifying and analyzing heart failure nuclear magnetic resonance images based on multimodal deep learning, which solves the problems in the above background art.

[0006] To achieve the above object, the present invention is realized through the following technical solutions: A method for identifying and analyzing heart failure nuclear magnetic resonance images based on multi-modal deep learning, comprising the following steps,

[0007] S1. Collect relevant medical data of different types of heart failure patients and normal control individuals from the cardiology department of the hospital, and conduct non-imaging examinations on the patients to be diagnosed to preliminarily diagnose whether the current patient has heart failure;

[0008] S2. If so, use magnetic resonance imaging technology to obtain the heart failure nuclear magnetic resonance images of the patients. After image preprocessing, extract relevant functional data, relevant anatomical data, and blood flow and vascular status data. Based on the relevant functional data and relevant anatomical data, and according to the relevant functional data and blood flow and vascular status data, analyze the type of heart failure of the patients to respectively obtain the first discrimination coefficient Sxs1 and the second discrimination coefficient Sxs2, and based on the value of the first discrimination coefficient Sxs1, analyze the activation state of the patient's own compensatory mechanism, and based on the activation state, re-judge the type of heart failure of the patients;

[0009] S3. Use deep learning technology, and combine relevant functional data, relevant anatomical data, and blood flow and vascular status data to construct a heart failure recognition and evaluation model, and fit and output a comprehensive discrimination index Zx. Based on the comprehensive discrimination index Zx, comprehensively evaluate the degree of the patient suffering from mixed heart failure;

[0010] S4. Based on the content of S3, and combined with the relevant medical data of different types of heart failure patients and normal control individuals collected from the cardiology department of the hospital in S1, verify the diagnostic efficacy of the current patient.

[0011] Preferably, the specific steps of S1 include:

[0012] S11. Pre-collect relevant medical data of different types of heart failure patients and normal control individuals from the cardiology department of the hospital. Among them, the relevant medical data includes the age, gender, past medical history, symptom manifestations, heart failure nuclear magnetic resonance images, physical examination results, and laboratory examination data of different types of heart failure patients and normal control individuals;

[0013] S12. Use a physical examination tool to auscultate the lungs of the patient to be diagnosed to identify whether there are moist rales. If so, the physical examination tool will issue an alarm to indicate the presence of moist rales. When the cardiology department of the hospital receives the alarm, it indicates that the current patient has a heart failure risk. At this time, laboratory examination types will be used to collect blood samples from the patient; among them, the physical examination tool includes a smart stethoscope, and the non-imaging examination includes physical examination and laboratory examination.

[0014] Preferably, the specific steps of S1 further include:

[0015] S13. Extract the B-type natriuretic peptide concentration and N-terminal pro-B-type natriuretic peptide concentration of each normal control individual from the laboratory test data of normal control individuals in S11. After statistics, determine the maximum value of the B-type natriuretic peptide concentration of normal control individuals and the maximum value of the N-terminal pro-B-type natriuretic peptide concentration of normal control individuals.

[0016] S14. According to the blood sampling in S12, detect the biomarker levels in the patient's blood. The biomarker levels include the patient's B-type natriuretic peptide concentration and the patient's N-terminal pro-B-type natriuretic peptide concentration. Combine with the extraction of the laboratory test data of normal control individuals in S13 to preliminarily diagnose whether the current patient has heart failure. If the patient's B-type natriuretic peptide concentration exceeds the maximum value of the B-type natriuretic peptide concentration of normal control individuals, and the patient's N-terminal pro-B-type natriuretic peptide concentration exceeds the maximum value of the N-terminal pro-B-type natriuretic peptide concentration of normal control individuals, preliminarily diagnose that the current patient has heart failure. At this time, the imaging examination mechanism will be activated.

[0017] Preferably, the specific steps of S2 include:

[0018] S21. Receive and activate the imaging examination mechanism. Use the respiratory gating technology to determine the time point for obtaining the heart failure MRI image of the patient using magnetic resonance imaging technology, and trigger the acquisition operation of the heart failure MRI image at this time point to obtain the heart failure MRI image of the patient. Remove the noise in the heart failure MRI image by using the Gaussian filtering method, and align the obtained multiple groups of heart failure MRI images to the same standard space through image registration technology.

[0019] Preferably, the specific steps of S2 also include:

[0020] S22. On the basis of S21, perform feature extraction on the heart failure MRI image after image preprocessing to obtain relevant functional data, relevant anatomical data, and blood flow and vascular status data, and correct the relevant functional data, relevant anatomical data, and blood flow and vascular status data according to anthropometric techniques and body surface area. Among them, the corrected relevant functional data includes the left ventricular ejection fraction Lvef and pulmonary artery blood flow Fxz actually measured by the patient; the corrected relevant anatomical data includes the left ventricular longitudinal strain ∈ long ; the corrected blood flow and vascular status data includes the measured plasma renin activity PRA, measured angiotensin II concentration Ang II, measured aldosterone concentration ALD, and the ratio EA of the peak velocity of early diastolic blood flow to the peak velocity of late diastolic blood flow in the mitral valve when the patient is in the supine position;

[0021] S23. According to the relevant functional data and the relevant anatomical data, and in combination with the relevant medical data of normal control individuals, analyze the type of heart failure of the patient, and after dimensionless processing, obtain the first discriminant coefficient Sxs1. The first discriminant coefficient Sxs1 is obtained through the following formula:

[0022]

[0023] In the formula, Lvef represents the left ventricular ejection fraction actually measured in the patient, and Lvef max represents the maximum value of the left ventricular ejection fraction in each normal control individual, and ∈ long represents the left ventricular longitudinal strain actually measured in the patient, and ∈ long，max represents the maximum value of the left ventricular longitudinal strain in each normal control individual, and Lvef min represents the minimum value of the left ventricular ejection fraction in each normal control individual, and ∈ long，min represents the minimum value of the left ventricular longitudinal strain in each normal control individual;

[0024] S24. Preset an evaluation threshold, and by comparing the first discriminant coefficient Sxs1 with the evaluation threshold, identify whether the heart failure of the current patient belongs to systolic functional heart failure. The specific content is as follows:

[0025] If the first discriminant coefficient Sxs1 exceeds the evaluation threshold, it will be identified that the heart failure of the current patient temporarily does not belong to systolic functional heart failure, and the activation state of the patient's own compensatory mechanism will be further detected;

[0026] If the first discriminant coefficient Sxs1 does not exceed the evaluation threshold, it will be identified that the heart failure of the current patient belongs to systolic functional heart failure, and the current patient will be labeled, and the systolic functional heart failure label will be set.

[0027] Preferably, the specific steps of S2 further include:

[0028] S25. When it is identified that the heart failure of the current patient temporarily does not belong to systolic functional heart failure, according to the blood sampling of the patient in S12, detect the measured plasma renin activity PRA, the measured angiotensin II concentration Ang II, and the measured aldosterone concentration ALD in the blood when the patient is in the supine position. In combination with the relevant medical data of normal control individuals, measure the activation degree of the renin-angiotensin-aldosterone system, and after dimensionless processing, obtain the activation index Jz. The specific method for obtaining it is as follows:

[0029]

[0030] In the formula, PRA max represents the upper limit of the plasma renin activity when each normal control individual is in the supine position, and Ang II maxRepresents the upper limit of the angiotensin II concentration when each normal control individual is in the supine position, ALD max Represents the upper limit of the aldosterone concentration when each normal control individual is in the supine position;

[0031] S26. By comparing the activation index Jz with a preset threshold, if the activation index Jz does not exceed the preset threshold, it indicates that the current patient's own compensatory mechanism is not yet activated. At this time, it is still considered that the current patient's heart failure does not belong to systolic functional heart failure as the comparison result. If the activation index Jz exceeds the preset threshold, it indicates that the current patient's own compensatory mechanism is activated. At this time, the type of the patient's heart failure will be rejudged, and it will be rejudged that the current patient's heart failure temporarily belongs to systolic functional heart failure.

[0032] Preferably, the specific steps of S2 further include:

[0033] S27. According to the relevant functional data and the blood flow and vascular status data, analyze the influence of the patient on diastolic functional heart failure, and after dimensionless processing, obtain the second discriminant coefficient Sxs2. The second discriminant coefficient Sxs2 is obtained by the following formula:

[0034]

[0035] In the formula, EA represents the ratio of the peak velocity of early diastolic blood flow in the mitral valve to the peak velocity of late diastolic blood flow, Fxz represents the pulmonary blood flow, and a1 and a2 both represent weight values. Among them, the specific values of a1 and a2 are set by the user according to the situation.

[0036] Preferably, the specific steps of S3 include:

[0037] S31. Use the convolutional neural network model in deep learning technology as the initial model, and use the relevant functional data, relevant anatomical data, blood flow and vascular status data, relevant medical data of different types of heart failure patients and normal control individuals as input quantities, and input them into the initial model. The input quantities are divided into a training set and a validation set. After training and validation, a heart failure recognition and evaluation model is generated. The heart failure recognition and evaluation model refers to the initial model after training and validation. According to the heart failure recognition and evaluation model, after dimensionless processing, the comprehensive discriminant index Zx is fitted and output from the output layer of the heart failure recognition and evaluation model. The comprehensive discriminant index Zx is obtained by the following formula:

[0038]

[0039] In the formula, F1 and F2 are both weight values. Among them, the specific values of F1 and F2 are set by the user according to the situation.

[0040] Preferably, the specific steps of S3 further include:

[0041] S32. Based on the statistical analysis method, set the discrimination range. By comparing the comprehensive discrimination index Zx with the discrimination threshold G, comprehensively evaluate the degree of the patient suffering from mixed heart failure. If the comprehensive discrimination index Zx falls within the discrimination range, it is comprehensively evaluated that the patient has a risk of suffering from mixed heart failure. If the comprehensive discrimination index Zx does not fall within the discrimination range, it is comprehensively evaluated that the patient does not currently have a risk of suffering from mixed heart failure.

[0042] Preferably, the specific steps of S4 include:

[0043] S41. Extract a number of patients suffering from mixed heart failure and a number of groups of normal control individuals from the hospital's cardiology department. The number of patients suffering from mixed heart failure is the same as the number of normal control individuals. Combine the relevant medical data of a number of patients suffering from mixed heart failure and a number of groups of normal control individuals, and conduct a diagnostic test operation on the number of patients suffering from mixed heart failure and a number of groups of normal control individuals extracted from the hospital's cardiology department. By the method of obtaining the comprehensive discrimination index Zx in S31 and the comparison content in S32, determine the proportion of positive diagnostic test results for the number of patients suffering from mixed heart failure extracted from the hospital's cardiology department, and mark it as the positive proportion Yb1, and the proportion of negative diagnostic test results for a number of groups of normal control individuals, and mark it as the negative proportion Yb2;

[0044] S42. According to the positive proportion Yb1 and the negative proportion Yb2, analyze the probability ratio of positive results in the populations with and without mixed heart failure to obtain the positive likelihood ratio LR. The positive likelihood ratio LR is obtained by the following formula:

[0045]

[0046] S43. When the value of the positive likelihood ratio LR ≥ 10, it indicates that the diagnostic efficacy of the current patient is within the correct range. The present invention provides a method for identifying and analyzing heart failure nuclear magnetic resonance images based on multi-modal deep learning, which has the following beneficial effects:

[0047] (1)This method further improves the accuracy of early heart failure diagnosis through multimodal data fusion. In stage S1, various medical data of cardiology patients are collected from multiple aspects, covering clinical symptoms, medical history, physical examination, laboratory indicators, etc. At the same time, non-imaging examinations are used to preliminarily judge the possibility of heart failure. This comprehensive screening method can capture the subtle abnormal manifestations of heart failure patients. Compared with traditional single examination methods, it further reduces the missed diagnosis rate of heart failure. For example, by combining symptoms such as fatigue and mild dyspnea in patients, as well as changes in biomarker levels such as BNP / NT-proBNP, signs of heart failure can be detected earlier. Early detection gains precious time for timely intervention, helps delay the progression of the disease, and improves the long-term prognosis of patients. Once a preliminary diagnosis of heart failure is made, in stage S2, MRI technology is used for in-depth analysis. Through the preprocessing of nuclear magnetic images and multi-dimensional data extraction, the first discriminant coefficient and the second discriminant coefficient are calculated to accurately judge the type of heart failure, and the activation state of the patient's compensatory mechanism is evaluated based on the first discriminant coefficient, so as to further clarify the type of heart failure. This accurate classification method provides a key basis for the formulation of subsequent personalized treatment plans. For example, for systolic heart failure and diastolic heart failure, the treatment strategies are significantly different. Accurately judging the type can avoid inappropriate treatment, improve the treatment effect, reduce drug adverse reactions, and significantly improve the quality of life of patients. Quantify the degree of heart failure and optimize treatment decisions: In stage S3, an evaluation model constructed by deep learning technology is used to comprehensively fit and output a comprehensive discriminant index Zx based on multimodal data, realizing the quantitative evaluation of the degree of mixed heart failure. Compared with traditional subjective judgments or simple index evaluations, this method is more objective, comprehensive and timely. Doctors can more scientifically adjust drug doses and select treatment methods according to the comprehensive discriminant index, such as deciding whether cardiac resynchronization therapy (CRT) is needed, etc. This treatment decision based on quantitative evaluation can optimize the treatment plan, avoid over-treatment or under-treatment, effectively save medical resources, and at the same time improve the treatment effect and survival rate of patients. Verify the diagnostic efficacy and ensure clinical reliability: In stage S4, through the data verification of a large number of patients and normal control individuals, the positive likelihood ratio LR is calculated to test the efficacy of the diagnostic method. When the LR value ≥ 10, it proves that the diagnostic method has extremely high accuracy, providing a reliable guarantee for clinical application. This verification mechanism can timely detect potential problems in the diagnostic process, continuously optimize the model, and ensure the stability and reliability of the diagnostic method in different clinical environments, making doctors and patients more confident in the diagnostic results and promoting the wide application of this technology in clinical practice.

[0048] (2)First, auscultation and blood sampling are used to preliminarily judge the possibility of heart failure, and then cardiac imaging examination is performed by MRI technology. This can indeed reduce the risk of patients being exposed to CT radiation, especially for those heart failure patients who need regular monitoring.

[0049] (3) In S11, various types of data are widely collected. Age and gender information can help determine the incidence of heart failure in different populations. For example, the characteristics of heart failure in the elderly and women are different from those in other populations. Collecting this information provides a basic dimension for subsequent analysis. Past medical history such as hypertension and coronary heart disease are important risk factors for heart failure. Knowing these conditions can help doctors quickly locate potential causes. Advantages of multimodal data fusion: Multimodal data such as symptoms, physical examination results, laboratory test data, and heart failure MRI images are integrated to form a comprehensive diagnostic information library. Different types of data reflect the patient's physical condition from different angles, providing doctors with a three-dimensional diagnostic perspective. For example, dyspnea and edema in symptoms are intuitive manifestations of heart failure, while BNP / NT-proBNP levels in laboratory test data reflect heart function at the molecular level. Multiple data confirm each other, greatly improving the accuracy of diagnosis. Real-time sign monitoring and rapid response: The S12 uses a smart stethoscope for lung auscultation. Once wet rales are detected, an alarm is immediately sounded and the blood sampling process is triggered. This process realizes real-time monitoring of key signs of heart failure and shortens the time from abnormality detection to further examination. As one of the common signs of heart failure, the timely detection of wet rales allows doctors to capture the patient's heart failure risk at the first time, and buys valuable time for subsequent diagnosis and treatment. Compared with traditional manual auscultation and manual judgment processes, the application of smart stethoscopes significantly improves diagnostic efficiency and reduces the risk of missed diagnosis due to human negligence. The linkage mechanism improves the timeliness of diagnosis: The linkage mechanism between the smart stethoscope and the hospital's cardiology department enables abnormal information to be quickly transmitted to the professional department and the corresponding diagnostic process is initiated. This rapid response mechanism ensures that patients can receive more in-depth examinations in a timely manner when signs of heart failure risk appear, which helps to detect heart failure early, so as to take effective intervention measures to prevent the disease from worsening.

[0050] (4) The respiratory gating technology in S21 is combined with MRI imaging to effectively avoid respiratory motion artifacts. By capturing the critical time point of end-expiration to trigger acquisition, the heart is ensured to be in a state where it is least affected by breathing, and high-quality MRI images of heart failure are obtained. These images clearly present cardiac structures such as the myocardium and valves, as well as functional details such as ventricular volume changes and myocardial motion, laying a solid foundation for subsequent feature extraction and analysis, allowing doctors to observe heart lesions more accurately and improve the accuracy of diagnosis.

[0051] (5)Precisely insight into the status of the RAAS system: S25 can accurately measure the activation degree of the RAAS system by calculating the activation index Jz. Taking renin activity, angiotensin II concentration, and aldosterone concentration as indicators, combined with the data of normal control individuals, it can intuitively reflect the differences between each indicator and the normal upper limit. Optimize the heart failure diagnosis path: When systolic functional heart failure (left heart failure) is initially excluded, the calculation of Jz provides a new dimension for diagnosis. If Jz > 0, it means the RAAS system is activated, prompting doctors to re-examine the type of heart failure to prevent misdiagnosis. This dynamic diagnosis idea based on data can effectively avoid missing key information due to a single judgment error, further improving the accuracy and comprehensiveness of diagnosis. Assist in judging the compensatory mechanism: In S26, by comparing Jz with a preset threshold, the activation status of the patient's compensatory mechanism can be clearly judged. When Jz exceeds the threshold, it indicates that the patient's compensatory mechanism is activated, and it is necessary to re-judge whether it is systolic functional heart failure (left heart failure). This helps doctors deeply understand the pathophysiological changes of the patient, provides an important basis for formulating personalized treatment plans, ensures the pertinence and effectiveness of treatment, and thus improves the treatment effect and prognosis of the patient.

[0052] (6)Precisely quantify the diagnostic discrimination ability: S42 calculates the positive likelihood ratio LR using the positive and negative ratios, accurately measuring the ability of the diagnostic test to distinguish between patients with mixed heart failure and non-patients. By comparing the probabilities of positive results in the diseased and non-diseased populations, it clearly demonstrates the effectiveness of the diagnostic method in differentiating diseases. This quantitative index provides a key basis for doctors and researchers to objectively judge the value of the diagnostic test, helping to accurately evaluate the feasibility and practicality of this method in actual clinical scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic flowchart of the method for identifying and analyzing heart failure nuclear magnetic resonance images based on multi-modal deep learning of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0054] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0055] Embodiment 1

[0056] Please refer to Figure 1 , the present invention provides a method for identifying and analyzing heart failure nuclear magnetic resonance images based on multi-modal deep learning, including the following steps,

[0057] S1. Collect relevant medical data of different types of heart failure patients and normal control individuals from the cardiology department of the hospital, and conduct non-imaging examinations on the patients to be diagnosed to preliminarily diagnose whether the current patients have heart failure;

[0058] S2. If so, use magnetic resonance imaging (MRI) technology to obtain the heart failure nuclear magnetic resonance images of the patients. After image preprocessing, extract relevant functional data, relevant anatomical data, and blood flow and vascular status data. Based on the relevant functional data and relevant anatomical data, and according to the relevant functional data and blood flow and vascular status data, analyze the type of heart failure of the patients to respectively obtain the first discriminant coefficient Sxs1 and the second discriminant coefficient Sxs2, and based on the value of the first discriminant coefficient Sxs1, analyze the activation status of the patient's own compensatory mechanism, and based on the activation status, re-judge the type of heart failure of the patients;

[0059] Classify heart failure according to the functional status of the heart, including systolic functional heart failure (left heart failure) and diastolic functional heart failure (right heart failure);

[0060] S3. Use deep learning technology, combined with relevant functional data, relevant anatomical data, and blood flow and vascular status data, to construct a heart failure recognition and evaluation model, and fit and output a comprehensive discriminant index Zx. Based on the comprehensive discriminant index Zx, comprehensively evaluate the degree of the patient's mixed heart failure;

[0061] S4. Based on the content of S3, combined with the relevant medical data of different types of heart failure patients and normal control individuals collected by S1 from the cardiology department of the hospital, verify the diagnostic efficacy of the current patients.

[0062] In this embodiment, the method comprehensively analyzes through multiple steps, maximizing the accuracy of heart failure diagnosis. In S1, multi-source medical data is collected from the cardiology department of the hospital, including the age, gender, past medical history, symptoms, physical examination results, and laboratory test data of different types of heart failure patients and normal control individuals. At the same time, non-imaging examinations are performed on the patient to be diagnosed to preliminarily determine whether there is a heart failure phenomenon, which provides basic information for subsequent in-depth diagnosis. Then, in S2, MRI technology is used to obtain the heart failure nuclear magnetic resonance images of the patient. After preprocessing, relevant functional, anatomical, and blood flow and vascular status data are extracted. Through these data, the first discriminant coefficient and the second discriminant coefficient are calculated to analyze the type of heart failure in the patient. And according to the first discriminant coefficient, the activation status of the patient's own compensatory mechanism is further judged, and then the type of heart failure is rejudged. This comprehensive analysis of multi-dimensional data can reflect the patient's heart condition in more aspects compared with a single examination method, effectively reducing the possibility of misdiagnosis and missed diagnosis. For example, by analyzing the patient's heart function data (such as left ventricular ejection fraction, left ventricular longitudinal strain, etc.) and anatomical data (such as myocardial thickness, chamber size, etc.), combined with blood flow and vascular status data, it is possible to more accurately distinguish systolic functional heart failure and diastolic functional heart failure, and determine whether there is mixed heart failure. In S3, a heart failure recognition and evaluation model is constructed using deep learning technology. Combining multi-modal data, the comprehensive discriminant index Zx is fitted and output, which can comprehensively evaluate the degree of mixed heart failure in the patient. This model integrates multi-source data and explores the potential connections between data, making the evaluation results more accurate. Compared with the traditional method of judging the degree of heart failure solely based on experience or simple indicators, this method is based on big data and complex algorithms and can take into account more influencing factors. For example, by comprehensively considering various information such as the patient's heart function, anatomical structure, hemodynamics, and clinical characteristics, the model can more accurately judge the severity of the patient's mixed heart failure, providing a strong basis for doctors to develop personalized treatment plans, helping to improve the treatment effect and the prognosis of the patient. In S4, based on the model output in S3 and the data collected in S1, the diagnostic efficacy of the current patient is verified. By performing diagnostic test operations on several patients with mixed heart failure and normal control individuals, the positive likelihood ratio LR is calculated. When the LR value ≥ 10, it indicates that the diagnostic efficacy of the current patient is within the correct range. This verification mechanism guarantees the reliability of the diagnostic method. For example, by verifying the diagnostic efficacy, potential problems in the diagnostic process can be timely discovered, the model can be optimized and improved, ensuring the accuracy and stability of the diagnostic method in clinical applications, providing a more reliable diagnostic tool for clinicians, and thus better serving the treatment and management of patients. Overall, this method provides comprehensive, accurate, and verified diagnostic information for clinicians, helping doctors make more scientific clinical decisions.From the initial diagnosis to the determination of the type of heart failure, then to the assessment of the degree of mixed heart failure and the verification of the diagnostic efficacy, each step provides rich information for doctors, enabling them to formulate treatment plans, select appropriate treatment drugs, and determine the treatment timing more targeted.

[0063] Example 2

[0064] Please refer to Figure 1 , specifically: The specific steps of S1 include:

[0065] S11. Collect relevant medical data of different types of heart failure patients and normal control individuals from the cardiology department of the hospital in advance. Among them, the relevant medical data include the age, gender, past medical history, symptom manifestations (such as dyspnea, edema, fatigue, etc.), heart failure MRI images, physical examination results (abnormalities in heart auscultation, lung auscultation, etc.), and laboratory test data (blood routine, blood biochemistry, cardiac markers such as BNP / NT-proBNP levels) of different types of heart failure patients and normal control individuals;

[0066] S12. Use a physical examination tool to perform lung auscultation on the patient to be diagnosed to identify whether there are moist rales. If so, the physical examination tool will issue an alarm to indicate the presence of moist rales. When the cardiology department of the hospital receives the alarm, it indicates that the current patient has a risk of heart failure. At this time, laboratory test types will be used to collect blood samples from the patient; among them, the physical examination tool includes intelligent stethoscopes (such as Eko Devices, Thinklabs One, etc.), and the non-imaging examinations include physical examinations and laboratory tests.

[0067] Moist rales refer to the sounds produced when the liquid and air in the airway interact due to lung water accumulation.

[0068] The specific steps of S1 also include:

[0069] S13. Extract the B-type natriuretic peptide concentration (BNP) and N-terminal pro-B-type natriuretic peptide concentration (NT-proBNP) of each normal control individual from the laboratory test data of the normal control individuals in S11. After statistics, determine the maximum value of the B-type natriuretic peptide concentration of the normal control individuals and the maximum value of the N-terminal pro-B-type natriuretic peptide concentration of the normal control individuals;

[0070] S14. According to the blood sampling in S12, the biomarker levels in the patient's blood are detected, wherein the biomarker levels include the patient's B-type natriuretic peptide concentration (BNP) and the patient's N-terminal precursor B-type natriuretic peptide concentration (NT-proBNP), and combined with the extraction of laboratory test data of normal control individuals in S13, a preliminary diagnosis is made as to whether the current patient has heart failure. If the patient's B-type natriuretic peptide concentration exceeds the maximum value of the B-type natriuretic peptide concentration of the normal control individual, and the patient's N-terminal precursor B-type natriuretic peptide concentration exceeds the maximum value of the N-terminal precursor B-type natriuretic peptide concentration of the normal control individual, the current patient is preliminarily diagnosed with heart failure, and the imaging examination mechanism is activated at this time.

[0071] In this embodiment, in step S11, multi-dimensional medical data of different types of heart failure patients and normal control individuals are widely collected, covering age, gender, past medical history, symptoms, heart failure MRI images, physical examination results and laboratory test data, etc. This comprehensive data collection method provides a rich and diversified source of information for subsequent diagnostic analysis. Through comparative analysis of different individual data, some hidden characteristics and patterns related to heart failure can be discovered. For example, combined with the patient's age, previous history of cardiovascular disease and current symptoms, the patient's risk tendency for heart failure can be preliminarily judged, providing a strong background support for subsequent diagnosis, and improving the accuracy and reliability of diagnosis as much as possible. Step S12 uses physical examination tools such as smart stethoscopes to auscultate the lungs and monitor in real time whether the patient has wet rales. Once wet rales are detected, the tool immediately issues an alarm, prompting the hospital's cardiology department patients to be at risk of heart failure, and quickly starts the blood sampling process. This real-time monitoring and rapid warning mechanism can detect abnormal signs of heart failure in patients at the first time. As one of the important clinical manifestations of heart failure, wet rales are discovered in time to gain valuable time for early diagnosis. Compared with the traditional manual auscultation and subsequent manual judgment process, the time interval from the discovery of abnormalities to further examination is further shortened, which helps to detect heart failure patients as early as possible and avoid delays in the disease. Steps S13 and S14 perform statistical analysis on the B-type natriuretic peptide (BNP) concentration and N-terminal pro-B-type natriuretic peptide (NT-proBNP) concentration of normal control individuals to determine the maximum concentration, and use this as a standard to compare with the biomarker levels of the patient to be diagnosed. This standardized comparison method provides an objective and quantitative basis for the preliminary diagnosis of heart failure.

[0072] Since BNP and NT-proBNP are important biomarkers reflecting cardiac function and the severity of heart failure, when both of these two indicators in a patient exceed the maximum value of normal control individuals, it can accurately preliminarily diagnose the presence of heart failure in the patient, thereby initiating the imaging examination mechanism. This process avoids the error of subjective judgment, improves the accuracy and consistency of diagnosis, helps to screen out potential heart failure patients at an early stage, and lays a foundation for more precise diagnosis and treatment in the follow-up.

[0073] Example 3

[0074] Please refer to Figure 1 , specifically: The specific steps of S2 include:

[0075] S21. Receive and initiate the imaging examination mechanism, use respiratory gating technology to determine the time point for acquiring the heart failure MRI image of the patient using magnetic resonance imaging (MRI) technology, and trigger the acquisition operation of the heart failure MRI image at this time point to obtain the heart failure MRI image of the patient. By adopting the Gaussian filtering method to remove the noise in the heart failure MRI image, improve the quality and clarity of the image, avoid the interference of noise on subsequent analysis, and align the multiple groups of heart failure MRI images obtained through image registration technology to the same standard space for effective feature comparison and analysis.

[0076] During the breathing process, the chest and abdomen of the human body will move, and this movement will inevitably drive the change of the heart position. If MRI data acquisition is carried out randomly without considering the respiratory factor, motion artifacts will be generated, resulting in blurred heart images. Respiratory gating technology monitors the respiratory cycle and finds the relatively static stage of respiratory movement, that is, the end of expiration. At this stage, the movement amplitudes of the chest and diaphragm are the smallest, and the heart is least affected by respiratory movement. By accurately determining this time point to trigger the MRI data acquisition, it can ensure that the data collected each time is obtained under the condition that the heart is least affected by respiratory movement. For example, when the respiratory gating system detects that the patient starts to exhale and determines the end of expiration after analysis, at this specific moment or a little later (this delay is to make the respiratory movement more stable when triggering the acquisition), trigger the radiofrequency pulse emission and signal acquisition of the MRI device to obtain high-quality heart images, which is conducive to accurately observing and evaluating the heart structure (such as myocardium, valves, etc.) and function (such as ventricular volume change, myocardial movement, etc.).

[0077] The specific steps of S2 also include:

[0078] S22. On the basis of S21, extract features from the preprocessed heart failure nuclear magnetic resonance images to obtain relevant functional data, relevant anatomical data, and blood flow and vascular status data, and correct the relevant functional data, relevant anatomical data, and blood flow and vascular status data according to anthropometric techniques and body surface area; wherein, the corrected relevant functional data includes the left ventricular ejection fraction Lvef and the pulmonary artery blood flow Fxz actually measured in the patient; the corrected relevant anatomical data includes the left ventricular longitudinal strain ∈ actually measured in the patient. long The corrected blood flow and vascular status data includes the measured plasma renin activity PRA, the measured angiotensin II concentration Ang II, the measured aldosterone concentration ALD, and the ratio EA of the peak early diastolic blood flow velocity to the peak late diastolic blood flow velocity of the mitral valve in the patient when lying down.

[0079] The technical nature of the Du Bois formula: The Du Bois formula comes from anthropometric techniques and belongs to the category of anthropometric techniques. From a technical perspective, it is an important tool in anthropometry. Anthropometry is a discipline that studies human characteristics by measuring and statistically analyzing human dimensions, proportions, etc. The Du Bois formula uses a mathematical model to establish a relationship between two variables, body weight and height, and body surface area, providing a method for quantifying human body surface area in fields such as medicine and physiology. Application in the correction technology of physiological function indicators: In medical research and clinical practice, it is mainly used to correct physiological function indicators such as the left ventricular ejection fraction and left ventricular longitudinal strain mentioned above. Since an individual's body size (mainly reflected by body surface area) can affect cardiac function indicators, calculating the body surface area using the Du Bois formula and then using this body surface area to correct cardiac function indicators can make cardiac function indicators more comparable among individuals of different body sizes. This is a technical means of data standardization, which helps to more accurately evaluate cardiac function and avoid misjudgment caused by individual body size differences.

[0080] S23. According to the relevant functional data and the relevant anatomical data, and in combination with the relevant medical data of normal control individuals, analyze the type of heart failure in the patient, and after dimensionless processing, obtain a first discriminant coefficient Sxs1. The first discriminant coefficient Sxs1 is obtained through the following formula:

[0081]

[0082] In the formula, Lvef represents the left ventricular ejection fraction actually measured in the patient, Lvef max represents the maximum value of the left ventricular ejection fraction within each normal control individual, ∈ long represents the left ventricular longitudinal strain actually measured in the patient, ∈ long，maxRepresents the maximum left ventricular longitudinal strain in each normal control individual, Lvef min Represents the minimum left ventricular ejection fraction in each normal control individual, ∈ long，min Represents the minimum left ventricular longitudinal strain in each normal control individual;

[0083] The above-mentioned left ventricular ejection fraction refers to the percentage of the left ventricular stroke volume to the left ventricular end-diastolic volume, which is an important indicator reflecting the left ventricular systolic function. Simply put, it reflects the ability of the heart to pump blood out during each contraction, and the normal range is generally 50%-70%. For example, a left ventricular ejection fraction of 60% means that after the left ventricle is filled with blood at the end of diastole, 60% of the blood can be pumped out into the aorta during each contraction. The left ventricular ejection fraction can be measured by echocardiogram;

[0084] Left ventricular longitudinal strain is used to evaluate the deformation ability of the left ventricular myocardium in the longitudinal direction (from the apex to the base of the heart), reflecting the myocardial systolic function. Normal myocardium will shorten to a certain extent during contraction, and left ventricular longitudinal strain is an index to quantify this shortening degree. A negative value indicates shortening during myocardial contraction, and the normal range is about -18% to -22%. For example, a left ventricular longitudinal strain value of -20% indicates that the myocardium has a 20% shortening in the longitudinal direction. Similarly, left ventricular longitudinal strain is mainly obtained by echocardiogram. Through the speckle tracking technique of echocardiogram, the movement of tiny structures (speckles) in the myocardium can be tracked, so as to accurately measure the strain of the myocardium in the longitudinal direction. This technique can sensitively detect early changes in myocardial systolic function and is of great significance for the diagnosis of some early heart diseases.

[0085] Sub-part: (Lvef - Lvef max ) calculates the difference between the actually measured Lvef and the upper limit of normal Lvef. The larger this difference, the farther Lvef deviates from the normal range, and the greater the negative contribution to the first discriminant coefficient Sxs1; (∈ long - ∈ long，max ) Similarly, it is the difference between the actually measured longitudinal strain and the upper limit of normal longitudinal strain, which also reflects the degree of deviation of the longitudinal strain from the normal range and also has a negative impact on the first discriminant coefficient Sxs1. Denominator part: (Lvef min - Lvef max ) is the span of the normal Lvef range, used to standardize the contribution of Lvef to the first discriminant coefficient Sxs1; (∈ long，min - ∈ long，max ) is the span of the normal longitudinal strain range, used to standardize the contribution of the longitudinal strain to the first discriminant coefficient Sxs1.

[0086] S24. Preset an evaluation threshold, and identify whether the heart failure of the current patient belongs to systolic functional heart failure (left heart failure) by comparing the first discriminant coefficient Sxs1 with the evaluation threshold. The specific content is as follows:

[0087] If the first discriminant coefficient Sxs1 exceeds the evaluation threshold, it will be recognized that the current patient's heart failure does not belong to systolic functional heart failure (left heart failure) for the time being, and the activation state of the patient's own compensatory mechanism will be further detected;

[0088] Among them, the heart will initiate a compensatory mechanism in the early stage of impaired function. For example, when myocardial contractility decreases, the heart will activate the renin-angiotensin-aldosterone system (RAAS) and the sympathetic nervous system through the neuro-humoral regulation mechanism, causing myocardial cell hypertrophy and ventricular cavity dilation to maintain cardiac output. At this stage, the left ventricular ejection fraction and left ventricular longitudinal strain may change to a certain extent, but have not reached the typical heart failure diagnostic criteria. It is necessary to comprehensively evaluate the compensatory state of the heart to accurately judge whether the left heart function is truly impaired.

[0089] If the first discriminant coefficient Sxs1 does not exceed the evaluation threshold, it will be recognized that the current patient's heart failure belongs to systolic functional heart failure (left heart failure), and the current patient will be labeled, and a systolic functional heart failure label will be set.

[0090] When the value of the first discriminant coefficient Sxs1 = 0, it means that Lvef and ∈ long are both at the upper limit of the normal range, and the cardiac systolic function is hovering on the edge of the abnormal state; as the value of the first discriminant coefficient Sxs1 gradually decreases, it means that Lvef and ∈ long deviate from the normal range more and more. When the first discriminant coefficient Sxs1 is less than the threshold, it indicates the presence of systolic functional heart failure (left heart failure).

[0091] In this embodiment, in S21, respiratory gating technology is combined with MRI imaging to effectively avoid respiratory motion artifacts and obtain high-quality nuclear magnetic resonance images of heart failure. By precisely capturing the moment at the end of exhalation when the heart is least affected by respiration for acquisition, the details of the heart's structure (such as myocardium and valves) and function (such as ventricular volume changes and myocardial motion) can be clearly shown, providing a reliable image basis for subsequent feature extraction and analysis, and further improving the accuracy of observing heart diseases. Data standardization and correction to eliminate interference from individual differences: In S22, using anthropometric technology and the Du Bois formula, the body surface area is calculated based on body weight and height, and relevant functional data, anatomical data, and blood flow and vascular status data are corrected. This method effectively eliminates the influence of individual body type differences on heart function indicators, making the heart function indicators of patients with different body types comparable, helping to more accurately evaluate heart function, and reducing misdiagnosis or misjudgment caused by body type factors. Quantitatively discriminating the types of heart failure to guide precise diagnosis: In S23, by constructing a first discriminant coefficient formula, combining the relevant functional and anatomical data of patients and the data of normal control individuals, after dimensionless processing, it provides a quantitative basis for judging the types of heart failure. The numerator reflects the deviation degree of the patient's indicators from the normal upper limit, and the denominator realizes the standardized contribution. Through this coefficient, the difference between the heart function and the normal range can be intuitively reflected. Threshold comparison to accurately judge systolic functional heart failure: In S24, an evaluation threshold is preset and compared with the first discriminant coefficient. When the coefficient does not exceed the threshold, it can accurately identify that the patient's heart failure belongs to systolic functional heart failure (left heart failure) and make a label. This process provides a clear standard for the diagnosis of systolic functional heart failure, helps doctors quickly and accurately judge the condition, lays a foundation for formulating targeted treatment plans subsequently, and improves the efficiency and accuracy of heart failure diagnosis.

[0092] Example 4

[0093] Please refer to Figure 1 , specifically: The specific steps of S2 also include:

[0094] S25. When it is identified that the current patient's heart failure does not belong to systolic functional heart failure (left heart failure) for the time being, according to the blood sampling of the patient in S12, the measured plasma renin activity (PRA), measured angiotensin II concentration (Ang II), and measured aldosterone concentration (ALD) in the patient's blood when lying down are detected. Combining the relevant medical data of normal control individuals, the activation degree of the renin-angiotensin-aldosterone system (RAAS) is measured, and after dimensionless processing, the activation index Jz is obtained. The specific way to obtain it is as follows:

[0095]

[0096] In the formula, PRA max represents the upper limit of the renin activity of each normal control individual when lying down, Ang IImax represents the upper limit of the angiotensin II concentration when each normal control individual is in the supine position, ALD max represents the upper limit of the aldosterone concentration when each normal control individual is in the supine position;

[0097] It should be noted that: when both the left ventricular ejection fraction and the left ventricular longitudinal strain decrease, it is not necessarily possible to determine that it is systolic functional heart failure (left heart failure), because there is a timely "protection" function in the human body (the body's own compensatory mechanism), but this protection may only be temporary. However, it is precisely this temporary situation that may prevent the timely detection of systolic functional heart failure (left heart failure) in patients; the body's own compensatory mechanism will play a role when the heart function is initially damaged. This compensatory mechanism is a temporary protective measure. At this stage, although some cardiac function indicators such as the left ventricular ejection fraction and the left ventricular longitudinal strain have become abnormal, the heart can still maintain the cardiac output within a certain range through adaptive changes in the cardiac structure (such as myocardial cell hypertrophy and ventricular cavity dilation) and function regulation (such as increased heart rate), so that the patient has not shown typical heart failure symptoms.

[0098] The measured plasma renin activity PRA in the above-mentioned blood refers to the rate at which renin catalyzes the conversion of angiotensinogen to angiotensin I, reflecting the activation degree of the renin-angiotensin-aldosterone system (RAAS). Renin is a proteolytic enzyme secreted by the juxtaglomerular cells of the kidney. When situations such as decreased renal perfusion pressure and reduced blood volume occur, renin secretion increases, thereby activating the RAAS system. It is measured by collecting venous blood and using laboratory detection methods such as radioimmunoassay (RIA) or enzyme-linked immunosorbent assay (ELISA).

[0099] The measured angiotensin II concentration Ang II is a key active substance in the renin-angiotensin-aldosterone system, which is converted from angiotensin I under the action of angiotensin-converting enzyme. It has a strong vasoconstrictive effect, can increase blood pressure, and stimulate the secretion of aldosterone, playing an important role in maintaining blood pressure and water-salt balance. Similarly, it is detected by collecting venous blood and using radioimmunoassay (RIA) or enzyme-linked immunosorbent assay (ELISA).

[0100] The measured aldosterone concentration ALD is a steroid hormone secreted by the zona glomerulosa of the adrenal cortex, mainly acting to retain sodium and excrete potassium, maintaining water-salt balance and blood volume. When the RAAS system is activated, aldosterone secretion increases, promoting the reabsorption of sodium and the excretion of potassium by the kidneys. It is also measured by collecting venous blood and using laboratory detection methods such as radioimmunoassay (RIA) and chemiluminescence immunoassay (CLIA) to determine the aldosterone concentration. These methods can specifically detect the content of aldosterone in the blood, helping doctors understand the water-salt metabolism and endocrine regulation in patients.

[0101] Among them, represents the ratio of the measured renin activity to the upper limit of the normal range. When the renin activity exceeds the upper limit of the normal range, this ratio is greater than 1, indicating an increase in renin activity, which is a signal that the RAAS system may be activated; in this formula, the renin activity (PRA) part uses in the form of, mainly because renin activity is the starting link of RAAS system activation. When the renin activity increases and exceeds the upper limit of the normal range, this ratio is greater than 1, directly reflecting the relative increase in renin activity and serving as a preliminary signal for RAAS activation.

[0102] calculates the ratio of the degree by which the measured angiotensin II concentration exceeds the normal upper limit to the normal upper limit. If this value is greater than 0, it indicates an increase in the angiotensin II concentration, further suggesting RAAS system activation;

[0103] has a similar meaning to the angiotensin II part. It is the ratio of the degree by which the measured aldosterone level exceeds the normal upper limit to the normal upper limit. When it is greater than 0, it indicates an increase in the aldosterone level, which is also one of the manifestations of RAAS system activation;

[0104] For the angiotensin II concentration and aldosterone level parts, the forms of and are used to more prominently show the relative increase degree after these two parts exceed the upper limit of the normal range. Because angiotensin II and aldosterone are downstream products after renin activation, the increase in the production of angiotensin II and aldosterone depends not only on renin activity but also on other factors (such as the activity of angiotensin-converting enzyme, regulation of aldosterone synthesis, etc.). By this way of subtracting the normal upper limit and then dividing by the upper limit, the abnormal increase degree of angiotensin II and aldosterone beyond the normal range can be more intuitively reflected.

[0105] When the activation index Jz = 0, it indicates that all indicators are within the normal range, and the RAAS system may not be activated. When the activation index Jz > 0, it means that at least one indicator exceeds the normal range, and the larger the index, the higher the activation degree of the RAAS system. The RAAS system is involved in the regulation process of cardiac function. Under normal physiological conditions, the situation of activation index Jz < 0 is relatively rare because the renin-angiotensin-aldosterone system (RAAS) of the human body usually maintains a relatively stable working state. Even in situations such as rest, these indicators generally do not fall too far below the lower limit of the normal range to ensure basic physiological functions such as maintaining blood pressure and electrolyte balance. Certain rare diseases or special physiological conditions may cause the RAAS system to be in a low-activity state. For example, in some patients with congenital adrenal cortical hypofunction, due to the impaired function of the adrenal cortex in secreting aldosterone, the aldosterone level may be significantly reduced, and at the same time, it may affect the feedback regulation of the renin-angiotensin system, resulting in a decrease in renin activity and angiotensin II concentration, thus leading to Jz < 0.

[0106] S26. By comparing the activation index Jz with a preset threshold, if the activation index Jz does not exceed the preset threshold, it indicates that the current patient's own compensatory mechanism is not yet activated. At this time, it is still determined that the current patient's heart failure does not belong to systolic functional heart failure (left heart failure) as the comparison result. If the activation index Jz exceeds the preset threshold, it indicates that the current patient's own compensatory mechanism is activated. At this time, the type of the patient's heart failure will be rejudged, and it will be rejudged whether the current patient's heart failure temporarily belongs to systolic functional heart failure (left heart failure).

[0107] In this embodiment, in S25, by detecting the plasma renin activity (PRA), angiotensin II concentration, and aldosterone concentration (ALD) in the patient's blood in the lying position and combining with the data of normal control individuals, an activation index Jz is constructed. This method can accurately measure the activation degree of the RAAS system. In the formula, the processing of each index, such as the ratio of plasma renin activity to the upper normal limit, and the ratio of angiotensin II and aldosterone to the upper normal limit beyond the normal range, comprehensively and intuitively reflects the abnormality of each index, providing a quantitative basis for accurately judging the status of the RAAS system; providing key clues for heart failure diagnosis: when it is initially judged that the patient's heart failure does not belong to systolic functional heart failure (left heart failure), the calculation of Jz provides a new direction for further diagnosis. By analyzing the Jz value, doctors can understand whether the RAAS system is involved. If Jz > 0, it indicates that the RAAS system is activated, which is closely related to cardiac function regulation, prompting doctors to re-examine the type of heart failure in the patient, providing key clues for subsequent diagnosis, and helping to more comprehensively understand the patient's pathophysiological state. Assisting in judging the activation state of the compensatory mechanism: In S26, by comparing the activation index Jz with a preset threshold, the activation state of the patient's own compensatory mechanism can be effectively judged. When Jz exceeds the threshold, it means that the patient's compensatory mechanism is activated. At this time, re-judging the type of heart failure may include the originally excluded systolic functional heart failure (left heart failure) in the consideration range. This process provides doctors with a dynamic diagnostic idea, avoiding ignoring important disease types due to the first misjudgment, improving the accuracy and timeliness of heart failure diagnosis, making the formulation of treatment plans more targeted, and thus better improving the patient's treatment effect and prognosis.

[0108] Example 5

[0109] Please refer to Figure 1 , specifically: The specific steps of S2 further include:

[0110] S27. According to the relevant functional data and the blood flow and vascular status data, analyze the influence of the patient on diastolic functional heart failure (right heart failure), and after dimensionless processing, obtain a second discriminant coefficient Sxs2. The second discriminant coefficient Sxs2 is obtained through the following formula:

[0111]

[0112] In the formula, EA represents the ratio of the peak velocity of early diastolic blood flow in the mitral valve to the peak velocity of late diastolic blood flow, Fxz represents the pulmonary blood flow, and a1 and a2 both represent weight values, where 0 < a1 < 1, 0 < a2 < 1, and the specific values of a1 and a2 are set by the user according to the situation.

[0113] The above-mentioned peak early diastolic mitral blood flow velocity can be measured by the pulsed Doppler technique of echocardiography. By placing the Doppler sampling volume at the mitral valve tip, the blood flow spectrum at the mitral valve orifice can be recorded, and thus the velocity of the E wave can be measured.

[0114] The peak late diastolic blood flow velocity can be measured by the pulsed Doppler technique of echocardiography. By recording the blood flow spectrum at the mitral valve tip, the velocity of the A wave can be further measured.

[0115] The pulmonary artery blood flow refers to the amount of blood flowing through the pulmonary artery per unit time, which reflects the ability of the right ventricle to transport blood to the lungs and is crucial for maintaining normal gas exchange and blood circulation in the lungs. The pulmonary artery blood flow is calculated by measuring the velocity-time integral (VTI) of the blood flow velocity at the pulmonary valve orifice by echocardiography and then combining it with the cross-sectional area of the pulmonary valve annulus.

[0116] In the early diastolic phase of the heart, the blood in the left atrium rapidly flows into the left ventricle. At this time, the blood flow velocity at the mitral valve orifice reaches its peak, which is the E wave. It reflects the diastolic function of the left ventricle and the pressure gradient between the left atrium and the left ventricle. Under normal circumstances, the velocity of the E wave is relatively high. When the diastolic function of the left ventricle is impaired, the velocity of the E wave may decrease. In the late diastolic phase of the heart, atrial contraction causes the remaining blood in the left atrium to further flow into the left ventricle. At this time, the blood flow velocity at the mitral valve orifice rises again and reaches its peak, which is called the A wave. The A wave mainly reflects the systolic function of the left atrium. Under normal circumstances, the velocity of the A wave is relatively lower than that of the E wave. When the diastolic dysfunction of the left ventricle gradually worsens, the velocity of the A wave may relatively increase and even exceed that of the E wave.

[0117] The specific steps of S3 include:

[0118] S31: Use the convolutional neural network model in deep learning technology as the initial model, and use relevant functional data, relevant anatomical data, blood flow and vascular status data, relevant medical data of different types of heart failure patients and normal control individuals as input quantities, and input them into the initial model. Then divide the input quantities into a training set and a validation set. After training and validation, a heart failure recognition and evaluation model is generated. The heart failure recognition and evaluation model refers to the initial model after training and validation. According to the heart failure recognition and evaluation model, after dimensionless processing, a comprehensive discrimination index Zx is fitted and output from the output layer of the heart failure recognition and evaluation model. The comprehensive discrimination index Zx is obtained through the following formula:

[0119]

[0120] In the formula, both F1 and F2 are weight values, where 0 < F1 < 1, 0 < F2 < 1, and the specific values of F1 and F2 are set by the user according to the situation.

[0121] The specific steps of S3 also include:

[0122] S32. Based on the statistical analysis method, a discrimination range is set, and the comprehensive discrimination index Zx is compared with the discrimination threshold G to comprehensively evaluate the degree of mixed heart failure in the patient. If the comprehensive discrimination index Zx falls within the discrimination range, the comprehensive evaluation shows that the patient is at risk of suffering from mixed heart failure. If the comprehensive discrimination index Zx does not fall within the discrimination range, the comprehensive evaluation shows that the patient is not at risk of suffering from mixed heart failure.

[0123] In this embodiment, S26 calculates the second discriminant coefficient by a specific formula, and deeply analyzes the impact of the patient's diastolic functional heart failure (right heart failure) based on relevant functional data and blood flow and vascular status data. Dimensionless processing makes the data more comparable, and different weight values ​​can be flexibly adjusted according to actual conditions to accurately reflect the degree of diastolic dysfunction. The acquisition of this coefficient provides key information for the comprehensive diagnosis of heart failure types, avoids misdiagnosis and missed diagnosis caused by only focusing on systolic function and ignoring diastolic dysfunction, and improves the heart failure diagnosis system. Use deep learning to build a powerful evaluation model: S31 uses a convolutional neural network model to integrate multi-source data, including functional, anatomical, blood flow and other data as well as medical information of different patients and normal control individuals. The input is divided into a training set and a validation set for training and verification. The generated heart failure recognition and evaluation model has strong learning and generalization capabilities. The model can mine the complex intrinsic connections between data and provide a reliable basis for subsequent accurate evaluation. Quantitative comprehensive discrimination to guide clinical decision-making: S32 calculates the comprehensive discrimination index Zx, sets the discrimination range based on statistical analysis, and compares Zx with the discrimination threshold G to achieve a quantitative assessment of the patient's mixed heart failure. This quantitative assessment method enables doctors to judge the patient's condition more intuitively and accurately. If Zx falls into the discrimination range, it indicates the risk of mixed heart failure, and doctors can formulate targeted treatment plans in a timely manner; if it does not fall into the range, the risk can be ruled out and unnecessary treatment can be avoided. This provides a strong basis for clinical decision-making, helps to improve treatment effects, optimize the allocation of medical resources, and improve patient prognosis.

[0124] Example 6

[0125] Please refer to Figure 1 , specifically: S4 specific steps include:

[0126] S41. Extract a number of patients with mixed heart failure and several groups of normal control individuals from the Department of Cardiology of the hospital. The number of patients with mixed heart failure is the same as the number of normal control individuals. Combine the relevant medical data of a number of patients with mixed heart failure and several groups of normal control individuals, and conduct a diagnostic test operation on the number of patients with mixed heart failure and several groups of normal control individuals extracted from the Department of Cardiology of the hospital. Through the method of obtaining the comprehensive discrimination index Zx in S31 and the comparison content in S32, determine the proportion of the diagnostic test results of a number of patients with mixed heart failure extracted from the Department of Cardiology of the hospital being positive (patients with mixed heart failure are diagnosed with mixed heart failure through the diagnostic test operation), and mark it as the positive proportion Yb1, and the proportion of the diagnostic test results of several groups of normal control individuals being negative (normal control individuals are diagnosed as normal individuals through the diagnostic test operation), and mark it as the negative proportion Yb2;

[0127] S42. According to the positive proportion Yb1 and the negative proportion Yb2, analyze the probability ratio of positive results in the populations with and without mixed heart failure, so as to measure the ability of the diagnostic test to distinguish patients from non-patients, and obtain the positive likelihood ratio LR. The positive likelihood ratio LR is obtained through the following formula:

[0128]

[0129] The numerator part in the formula refers to the proportion of positive diagnostic test results in the population with mixed heart failure, which belongs to the true positive number (TP), that is, the number of patients who actually have the disease and have positive diagnostic test results. For example, if there are 100 patients with a certain disease, and after being tested by a certain diagnostic test, 80 patients have positive results, then the positive proportion Yb1 of this diagnostic test for this disease is So from a probability perspective, the numerator is the probability of a positive diagnostic test in the population with the disease.

[0130] The denominator part refers to the proportion of negative diagnostic test results in the population without the disease, which belongs to the true negative number (TN), that is, the number of individuals who do not have the disease and have negative diagnostic test results. For example, if there are 100 people without a certain disease, and after being tested by a certain diagnostic test, 90 have negative results, then the specificity of this diagnostic test for this disease is And 1 - Yb2 = 0.1. From a probability perspective, the denominator is the probability of a positive diagnostic test in the population without the disease.

[0131] S43. When the value of the positive likelihood ratio LR ≥ 10, it indicates that the diagnostic efficacy of the current patient is within the correct range.

[0132] When the positive likelihood ratio is greater than 10, it is a very strong diagnostic evidence, which means that the possibility of a positive result in a population with the disease is more than 10 times the possibility of a positive result in a population without the disease. For example, for a certain disease, the LR of a diagnostic test is 20, indicating that the positive result of the test has a strong suggestive effect on the diagnosis of the disease, and the diagnostic accuracy is relatively high at this time.

[0133] When the LR is greater than 10, under the framework of Bayes' theorem, the posterior probability (the probability that the patient actually has the disease after receiving a positive test result) will be significantly improved. This helps doctors make correct diagnostic decisions and reduce uncertainty when faced with positive test results. For example, in a population with a disease prevalence of 10%, if the LR of a diagnostic test is greater than 10, when the patient tests positive, the posterior probability that he or she actually has the disease will increase significantly, making doctors more confident in making subsequent treatment decisions.

[0134] In this embodiment, in S41, by conducting diagnostic tests on a number of patients with mixed heart failure and an equal number of normal control individuals, the positive ratio and negative ratio are obtained. This process provides a solid data basis for subsequent analysis. This rigorous sample selection and test method can comprehensively and truly reflect the performance of the diagnostic method in different populations, avoid evaluation errors caused by sample bias, and ensure the reliability of the evaluation of diagnostic accuracy. Quantitatively distinguish between patients and non-patients: S42 accurately measures the ability of the diagnostic test to distinguish between patients and non-patients by calculating the positive likelihood ratio LR. Comparing the positive probability in the population with mixed heart failure with the positive probability in the non-diseased population intuitively shows the discriminatory efficacy of the diagnostic test, which enables doctors and researchers to clearly understand the effectiveness of the diagnostic method in practical applications and provides a quantitative basis for judging the value of the diagnostic test. Powerfully guide clinical decision-making: S43 clearly states that when the LR value is ≥10, the diagnostic efficacy is in the correct range. A high positive likelihood ratio not only means that the diagnostic test is highly accurate, but also plays a key role in clinical decision-making. Based on Bayes' theorem, when LR is greater than 10, the posterior probability corresponding to the patient's positive test result is significantly improved. When doctors face a positive result, they can be more confident in making a correct diagnosis and then decisively carry out follow-up treatment, reducing delays or wrong decisions caused by diagnostic uncertainty, further improving the timeliness and accuracy of clinical treatment, gaining precious time for patients' treatment, and effectively improving patients' prognosis.

[0135] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying and analyzing heart failure nuclear magnetic resonance images based on multimodal deep learning, characterized in that: Including the following steps, S1. Collect relevant medical data of different types of heart failure patients and normal control individuals from the department of cardiology in the hospital, and conduct non-imaging examinations on the patients to be diagnosed to preliminarily diagnose whether the current patient has heart failure; S2. If so, use magnetic resonance imaging technology to obtain the heart failure nuclear magnetic resonance images of the patients. After image preprocessing, extract relevant functional data, relevant anatomical data, and blood flow and vascular status data. Based on the relevant functional data and relevant anatomical data, and according to the relevant functional data and blood flow and vascular status data, analyze the type of heart failure of the patients to obtain the first discriminant coefficient Sxs1 and the second discriminant coefficient Sxs2 respectively. And based on the value of the first discriminant coefficient Sxs1, analyze the activation status of the patient's own compensatory mechanism. Based on the activation status, re-judge the type of heart failure of the patient; S3. Use deep learning technology, and combine relevant functional data, relevant anatomical data, and blood flow and vascular status data to construct a heart failure recognition and evaluation model, and fit and output a comprehensive discriminant index Zx. Evaluate the degree of mixed heart failure of the patient based on the comprehensive discriminant index Z; S4. Based on the content of S3, and combined with the relevant medical data of different types of heart failure patients and normal control individuals collected from the department of cardiology in the hospital in S1, verify the diagnostic efficacy of the current patient.

2. The method for identifying and analyzing heart failure nuclear magnetic resonance images based on multimodal deep learning according to claim 1, wherein: The specific steps of S1 include: S11. Pre-collect relevant medical data of different types of heart failure patients and normal control individuals from the department of cardiology in the hospital. Among them, the relevant medical data include the age, gender, past medical history, symptom manifestations, heart failure nuclear magnetic resonance images, physical examination results, and laboratory examination data of different types of heart failure patients and normal control individuals; S12. Use a physical examination tool to auscultate the lungs of the patient to be diagnosed to identify whether there are moist rales. If so, the physical examination tool will issue an alarm to indicate the presence of moist rales. When the department of cardiology in the hospital receives the alarm, it means that the current patient has a heart failure risk. At this time, laboratory examination types will be used to collect blood samples from the patient; Among them, the physical examination tool includes a smart stethoscope, and the non-imaging examination includes physical examination and laboratory examination.

3. The method for identifying and analyzing heart failure nuclear magnetic resonance images based on multimodal deep learning according to claim 2, wherein: The specific steps of S1 also include: S13. Extract the B-type natriuretic peptide concentration and N-terminal pro-B-type natriuretic peptide concentration of each normal control individual from the laboratory examination data of the normal control individuals in S11. After statistics, determine the maximum value of the B-type natriuretic peptide concentration of the normal control individuals and the maximum value of the N-terminal pro-B-type natriuretic peptide concentration of the normal control individuals; S14. Detect the levels of biomarkers in the patient's blood based on the blood sampling in S12. The biomarker levels include the concentration of B-type natriuretic peptide and the concentration of N-terminal pro-B-type natriuretic peptide in the patient. Combine with the extraction of laboratory examination data of normal control individuals in S13 to preliminarily diagnose whether the current patient has heart failure. If the concentration of B-type natriuretic peptide in the patient exceeds the maximum value of the B-type natriuretic peptide concentration of normal control individuals, and the concentration of N-terminal pro-B-type natriuretic peptide in the patient exceeds the maximum value of the N-terminal pro-B-type natriuretic peptide concentration of normal control individuals, preliminarily diagnose that the current patient has heart failure. At this time, the imaging examination mechanism will be activated.

4. The method for identifying and analyzing heart failure nuclear magnetic resonance images based on multimodal deep learning according to claim 3, wherein: The specific steps of S2 include: S21. Receive and activate the imaging examination mechanism. Use the respiratory gating technique to determine the time point for obtaining the heart failure MRI image of the patient using magnetic resonance imaging technology, and trigger the acquisition operation of the heart failure MRI image at this time point to obtain the heart failure MRI image of the patient. Remove the noise in the heart failure MRI image by using the Gaussian filtering method, and align the obtained multiple groups of heart failure MRI images to the same standard space through image registration technology.

5. The method for identifying and analyzing heart failure nuclear magnetic resonance images based on multi-modal deep learning according to claim 4, characterized in that: The specific steps of S2 also include: S22. On the basis of S21, perform feature extraction on the preprocessed heart failure nuclear magnetic resonance images to obtain relevant functional data, relevant anatomical data, and blood flow and vascular status data, and correct the relevant functional data, relevant anatomical data, and blood flow and vascular status data according to anthropometric techniques and body surface area. Among them, the corrected relevant functional data includes the left ventricular ejection fraction Lvef and the pulmonary artery blood flow Fxz actually measured by the patient; the corrected relevant anatomical data includes the left ventricular longitudinal strain ∈ actually measured by the patient. long The corrected blood flow and vascular status data includes the measured plasma renin activity PRA, the measured angiotensin II concentration Ang II, the measured aldosterone concentration ALD, and the ratio EA of the peak early diastolic blood flow velocity to the peak late diastolic blood flow velocity of the mitral valve when the patient is in the supine position. S23. According to the relevant functional data and the relevant anatomical data, and combine with the relevant medical data of normal control individuals, analyze the type of the patient's heart failure. After dimensionless processing, obtain the first discriminant coefficient Sxs1. The first discriminant coefficient Sxs1 is obtained through the following formula: Wherein, Lvef represents the left ventricular ejection fraction actually measured in the patient, Lvef max represents the maximum value of the left ventricular ejection fraction in each normal control individual, ∈ long represents the left ventricular longitudinal strain actually measured in the patient, ∈ long，max represents the maximum value of the left ventricular longitudinal strain in each normal control individual, Lvef min represents the minimum value of the left ventricular ejection fraction in each normal control individual, ∈ long，min represents the minimum value of the left ventricular longitudinal strain in each normal control individual; S24. Preset an evaluation threshold. By comparing the first discriminant coefficient Sxs1 with the evaluation threshold, identify whether the heart failure of the current patient belongs to systolic functional heart failure. The specific content is as follows: If the first discriminant coefficient Sxs1 exceeds the evaluation threshold, at this time, it will be identified that the heart failure of the current patient temporarily does not belong to systolic functional heart failure, and further detect the activation state of the patient's own compensatory mechanism; If the first discriminant coefficient Sxs1 does not exceed the evaluation threshold, at this time, it will be identified that the heart failure of the current patient belongs to systolic functional heart failure, and label the current patient, and set the systolic functional heart failure label.

6. The method for identifying and analyzing heart failure nuclear magnetic resonance images based on multi-modal deep learning according to claim 5, characterized in that: The specific steps of S2 also include: S25. When it is identified that the heart failure of the current patient temporarily does not belong to systolic functional heart failure, based on the blood sampling of the patient in S12, detect the measured plasma renin activity (PRA), the measured concentration of angiotensin II (Ang II), and the measured concentration of aldosterone (ALD) in the blood when the patient is in the supine position. Combine with the relevant medical data of normal control individuals to measure the activation degree of the renin-angiotensin-aldosterone system. After dimensionless processing, obtain the activation index Jz. The specific method for obtaining it is as follows: where PRA max represents the upper limit of renin activity of each normal control individual in the supine position, AngⅡ max represents the upper limit of angiotensin II concentration of each normal control individual in the supine position, ALD max represents the upper limit of aldosterone concentration of each normal control individual in the supine position; S26. By comparing the activation index Jz with a preset threshold, if the activation index Jz does not exceed the preset threshold, it indicates that the current patient's own compensatory mechanism is not yet in an activated state. At this time, it is still considered that the current patient's heart failure does not belong to systolic functional heart failure as the comparison result. If the activation index Jz exceeds the preset threshold, it indicates that the current patient's own compensatory mechanism is in an activated state. At this time, the type of the patient's heart failure will be rejudged, and it is rejudged that the current patient's heart failure temporarily belongs to systolic functional heart failure.

7. The method for identifying and analyzing heart failure nuclear magnetic resonance images based on multimodal deep learning according to claim 6, wherein: The specific steps of S2 also include: S27. According to the relevant functional data and the blood flow and vascular status data, analyze the influence of the patient on diastolic functional heart failure, and after dimensionless processing, obtain the second discriminant coefficient Sxs2. The second discriminant coefficient Sxs2 is obtained through the following formula: In the formula, EA represents the ratio of the peak velocity of early diastolic blood flow in the mitral valve to the peak velocity of late diastolic blood flow, Fxz represents the pulmonary artery blood flow, and a1 and a2 both represent weight values. Among them, the specific values of a1 and a2 are set by the user according to the situation.

8. The method for identifying and analyzing heart failure nuclear magnetic resonance images based on multimodal deep learning according to claim 7, wherein: The specific steps of S3 include: S31. Use the convolutional neural network model in deep learning technology as the initial model, and use the relevant functional data, relevant anatomical data, blood flow and vascular status data, relevant medical data of different types of heart failure patients and normal control individuals as input quantities, and input them into the initial model. The input quantities are divided into a training set and a validation set. After training and validation, a heart failure recognition and evaluation model is generated. The heart failure recognition and evaluation model refers to the initial model after training and validation. According to the heart failure recognition and evaluation model, after dimensionless processing, the comprehensive discriminant index Zx is fitted and output from the output layer of the heart failure recognition and evaluation model. The comprehensive discriminant index Zx is obtained through the following formula: In the formula, F1 and F2 are both weight values. Among them, the specific values of F1 and F2 are set by the user according to the situation.

9. The method for identifying and analyzing heart failure nuclear magnetic resonance images based on multi-modal deep learning according to claim 8, wherein: The specific steps of S3 also include: S32. Based on the statistical analysis method, set the discriminant range. By comparing the comprehensive discriminant index Zx with the discriminant threshold G, comprehensively evaluate the degree of the patient suffering from mixed heart failure. If the comprehensive discriminant index Zx falls within the discriminant range, it is comprehensively evaluated that the patient has a risk of suffering from mixed heart failure. If the comprehensive discriminant index Zx does not fall within the discriminant range, it is comprehensively evaluated that the patient temporarily does not have a risk of suffering from mixed heart failure.

10. The method for identifying and analyzing heart failure nuclear magnetic resonance images based on multimodal deep learning according to claim 9, characterized in that: The specific steps of S4 include: S41. Extract a number of patients with mixed heart failure and several groups of normal control individuals from the department of cardiology in the hospital. The number of patients with mixed heart failure is the same as the number of normal control individuals. Combine the relevant medical data of a number of patients with mixed heart failure and several groups of normal control individuals, and conduct a diagnostic test operation on the extracted patients with mixed heart failure and several groups of normal control individuals from the department of cardiology in the hospital. Determine the proportion of positive diagnostic test results for the extracted patients with mixed heart failure from the department of cardiology in the hospital, which is marked as the positive proportion Yb1, and the proportion of negative diagnostic test results for several groups of normal control individuals, which is marked as the negative proportion Yb2, through the method of obtaining the comprehensive discrimination index Zx in S31 and the comparison content in S32. S42. According to the positive proportion Yb1 and the negative proportion Yb2, analyze the probability ratio of positive results in the populations with and without mixed heart failure to obtain the positive likelihood ratio LR. The positive likelihood ratio LR is obtained through the following formula: S43. When the value of the positive likelihood ratio LR ≥ 10, it indicates that the diagnostic efficacy of the current patient is within the correct range.

Images