Heart valve anomaly analysis method based on deep learning

Abstract

The invention mainly aims to provide a heart valve anomaly analysis method based on deep learning, and the abnormal condition of a heart valve, especially the severity of aortic stenosis, is accurately detected through automatic image processing and stenosis rate calculation. The method can reduce errors of manual operation and improve the diagnosis efficiency of heart valve diseases.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing and intelligent diagnosis, and particularly to a method for analyzing cardiac valve abnormalities based on deep learning, mainly used for the detection and evaluation of aortic stenosis. Background Art

[0002] Cardiac valve diseases are common types of heart diseases, among which aortic stenosis is one of the most serious lesions in cardiac valve abnormalities. Aortic stenosis is caused by the thickening, hardening or calcification of the aortic valve, resulting in valve stenosis, restricting blood flow, and leading to heart failure or even death.

[0003] The existing detection of aortic stenosis mainly relies on echocardiography or magnetic resonance imaging (MRI), and parameters such as the aortic valve orifice area and transvalvular pressure gradient are measured manually. However, this method requires professional doctors for evaluation, and subjective factors may lead to misdiagnosis and missed diagnosis.

[0004] However, these traditional methods have the following defects:

[0005] 1. Strong subjectivity: The analysis of image data depends on the experience of doctors, and there are errors in subjective judgment, which affects the accuracy of diagnosis.

[0006] 2. Time-consuming: Complex image data requires doctors to analyze frame by frame, which is time-consuming and laborious, and has low efficiency.

[0007] 3. Misdiagnosis rate: Due to the different experiences and film reading levels of different doctors, missed diagnosis and misdiagnosis are likely to occur, especially in the detection of early cardiac valve lesions.

[0008] In recent years, the development of deep learning technology has provided new ideas for the automated analysis of medical images, and it can automatically identify the lesion sites in images by training neural network models based on a large amount of medical image data. However, the current deep learning methods for detecting cardiac valve abnormalities are still in the preliminary research stage and lack mature solutions. Summary of the Invention

[0009] The main purpose of the present invention is to provide a method for analyzing cardiac valve abnormalities based on deep learning, which can accurately detect the abnormalities of cardiac valves, especially the severity of aortic stenosis, through automated image processing and stenosis rate calculation. This method can reduce the errors of manual operations and improve the diagnostic efficiency of cardiac valve diseases.

[0010] The present invention proposes a method for analyzing cardiac valve abnormalities based on deep learning, including the following steps:

[0011] Step 1. Image and data acquisition: Obtain the cardiac image of the patient through echocardiogram, and perform preprocessing operations such as denoising, normalization, and segmentation on the image to ensure the clear visibility of the valve area; through the Doppler mode of echocardiogram, obtain the patient's physiological parameters, including cardiac output CO and transvalvular pressure difference ΔP.

[0012] Step 2. Image feature extraction: Based on the convolutional neural network CNN model, analyze the preprocessed cardiac image and automatically extract the key pathological features of the valve, including:

[0013] Degree of calcification: Identify the calcified area through image segmentation technology, and calculate the calcification area and distribution;

[0014] Valve thickness: Extract the thickness information of the valve through edge detection algorithm;

[0015] Valve opening and closing state: Detect the kinematic parameters of the valve through motion tracking algorithm.

[0016] Step 3. Image feature scoring: Score the above image features, specifically including:

[0017] Calcification score: Calculate the score according to the calcification area combined with the calcification density;

[0018] Thickening score: Calculate the score according to the thickening volume of the valve;

[0019] Opening and closing score: Calculate the score according to the dynamic kinematic parameters of the valve;

[0020] On this basis, generate the image feature score TAS from the calcification score, thickening score, and opening and closing score.

[0021] Step 4. Physiological parameter calculation: Based on the patient's cardiac output CO, transvalvular pressure difference ΔP and other physiological parameters, calculate the effective orifice area EOA of the valve, using the following formula:

[0022]

[0023] where CO is the cardiac output, HR is the heart rate, SEP is the effective systolic time, ΔP is the transvalvular pressure difference, and ρ is the blood density.

[0024] Step 5. Stenosis rate calculation: By combining physiological parameters and image feature scores, comprehensively calculate the aortic valve stenosis rate SR using the following formula:

[0025]

[0026] where EOA is the effective orifice area of the valve, TAS is the image feature score, α and β are weight coefficients, EOA normal is the normal reference orifice area of the valve, TASnormal The normal reference score for imaging features.

[0027] Step 6, stenosis rate output: According to the comprehensively calculated stenosis rate, the system outputs a reference diagnosis result of the stenosis degree to provide decision support for doctors.

[0028] The present invention proposes a cardiac valve abnormality analysis system based on deep learning. This system is used to implement the above-mentioned cardiac valve abnormality analysis method based on deep learning and includes the following modules:

[0029] An image and data acquisition module, which is used to obtain the cardiac medical images of patients and obtain the physiological parameters of patients, including cardiac output CO and transvalvular pressure difference ΔP;

[0030] An imaging feature extraction module, which is used to automatically extract the key pathological features of the valve based on the convolutional neural network CNN model;

[0031] An imaging feature scoring module, which is used to score the imaging features and generate an imaging feature score TAS;

[0032] A physiological parameter calculation module, which is used to calculate the effective orifice area EOA based on the cardiac output CO and transvalvular pressure difference ΔP of the patient;

[0033] A stenosis rate calculation module, which is used to calculate the stenosis rate SR of the aortic valve by combining physiological parameters and imaging feature scores. The calculation formula is as follows:

[0034]

[0035] Among them, EOA is the effective orifice area, TAS is the imaging feature score, α and β are weight coefficients, EOA normal is the normal reference orifice area, and TAS normal is the normal reference score for imaging features;

[0036] A stenosis rate output module, which is used to output a reference diagnosis result of the stenosis degree according to the comprehensively calculated stenosis rate.

[0037] The present invention proposes a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned cardiac valve abnormality analysis method based on deep learning are implemented.

[0038] The technical solution of the present invention has the following advantages:

[0039] High degree of automation: Based on deep learning technology, it can automatically extract imaging features and calculate the stenosis rate, reduce human intervention, and reduce the risk of misdiagnosis.

[0040] High-precision detection: By combining a convolutional neural network and a time-series analysis network, it can efficiently detect structural and functional abnormalities of heart valves and improve the diagnostic accuracy.

[0041] High clinical value: By combining imaging features and physiological parameters, it can more accurately evaluate the degree of aortic valve stenosis, providing a more comprehensive diagnostic basis. Through the automatically generated diagnostic report, it provides important reference for clinicians, especially significant in the early diagnosis and intervention of aortic stenosis. Brief Description of the Drawings

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0043] Figure 1 It is a schematic flow chart of the method for analyzing heart valve abnormalities based on deep learning of the present invention;

[0044] Figure 2 It is a schematic structural diagram of the CNN model in the process of extracting imaging features of the present invention;

[0045] Figure 3 It is a schematic diagram of a system for analyzing heart valve abnormalities based on deep learning proposed by the present invention. Detailed Embodiments

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0047] Embodiment 1

[0048] Refer to the attached Figure 1 , the present invention proposes a method for analyzing heart valve abnormalities based on deep learning, including the following steps:

[0049] Step 1. Image and data acquisition: Obtain the heart image of the patient through echocardiogram, and perform preprocessing operations such as denoising, normalization, and segmentation on the image to ensure the clear visibility of the valve area; through the Doppler mode of echocardiogram, obtain the physiological parameters of the patient, including cardiac output CO and transvalvular pressure difference ΔP.

[0050] Step 2. Image feature extraction: Based on the convolutional neural network (CNN) model, analyze the preprocessed cardiac images to automatically extract the key pathological features of the valves, including:

[0051] Degree of calcification: Through image segmentation technology, identify the calcified areas and calculate the calcification area and distribution.

[0052] Valve thickness: Extract the thickness information of the valve through edge detection algorithms.

[0053] Valve opening and closing state: Detect the kinematic parameters of the valve through motion tracking algorithms.

[0054] Specifically, establish a convolutional neural network (CNN) model for extracting features from cardiac images, including:

[0055] 1. Data preparation: Collect a large number of cardiac ultrasound images that contain pathological features of different degrees; preprocess the images, including denoising, standardizing the size, enhancing the contrast, etc., to improve the recognition ability of the model.

[0056] 2. Labeling data: Professional doctors label the calcified areas, valve thickness, and valve opening and closing states in the images, and these labels will be used as the "ground truth" for model training.

[0057] 3. Model design: Refer to Appendix Figure 2 , the designed CNN model includes:

[0058] Input layer, used to input the image to be judged;

[0059] Convolutional layer, used to extract the calcified areas, valve thickness, and valve opening and closing states in the image;

[0060] Pooling layer, used to reduce the dimensionality of the high-dimensional features of the image and retain the key features;

[0061] Activation layer, used to enhance the extraction ability of non-linear features using the ReLU function;

[0062] Scoring layer, used to predict the structural abnormalities of the valve (such as calcification, thickening, opening and closing disorders).

[0063] 4. Model training: Use the labeled image data to train the CNN model; optimize the model weights through the backpropagation algorithm to minimize the difference between the prediction results and the true labels.

[0064] 5. Model evaluation: Evaluate the performance of the model on an independent test set using metrics such as accuracy, recall rate, and F1 score.

[0065] 6. Identification of key pathological features: Input the new cardiac image into the trained CNN model, and the model will output the identification results of medical record features, including the calcification area, valve thickness, and valve dynamic kinematic parameters.

[0066] Step 3. Image feature scoring: Score the above image features, specifically including:

[0067] Calcification score: Calculate the score based on the calcification area combined with the calcification density;

[0068] Thickening score: Calculate the score based on the valve thickening volume;

[0069] Opening and closing score: Calculate the score based on the valve dynamic kinematic parameters;

[0070] On this basis, generate the image feature score from the calcification score, thickening score, and opening and closing score.

[0071] Specifically, the scoring can be carried out with reference to the following implementation methods:

[0072] 1. Calculation of calcification score:

[0073] The Agatstion scoring method is a commonly used calcification score calculation method in the prior art. The calcification score can be scored according to the size and density of the calcification area. The higher the proportion of the calcification area in the total valve area, the higher the score.

[0074] The present invention not only evaluates the density of calcification, but also judges its stability by combining the "density" of calcification, thereby making the calcification score more comprehensive and accurate. The risks of soft plaques (low HU value) and hard calcifications with poor stability (high HU value) may be different. The plaques with HU values in different ranges are processed separately. The calculation formula of the calcification score (Ca_Score) is as follows:

[0075] Ca_Score = ∑Area i ×(HU i ×W i )

[0076] Among them, Area i is the area of the i-th calcification spot, HU i is the HU value of the i-th calcification spot, and W i is the density coefficient corresponding to the HU value of the i-th calcification spot.

[0077] For example, the correspondence between HU value and density coefficient is as follows:

[0078] 130 ≤ HU < 200, the density coefficient is 0.0025;

[0079] 200 ≤ HU < 300, the density coefficient is 0.005;

[0080] 300 ≤ HU < 400, with a density coefficient of 0.0075;

[0081] HU ≥ 400, with a density coefficient of 0.01.

[0082] For example, if the patient's image shows 2 calcification spots with an area of 8 mm2 and a HU value of 420, and an area of 10 mm2 and a HU value of 340, the calcification score is calculated as follows:

[0083] Ca_Score = 8 × 420 × 0.01 + 10 × 340 × 0.005 = 50.6

[0084] 2. Calculation of thickening score

[0085] The CNN model can automatically detect the edges of the valve and calculate its thickness. By automatically detecting the thickness changes of each leaflet, the model can determine whether the valve has significant thickening.

[0086] In the prior art, the conventional valve thickening score is calculated based on the mean value of the valve thickness as the difference between the actual thickness and the normal reference value, and the score is:

[0087]

[0088] In the prior art, the actual thickness is usually obtained by averaging the thickness measured at several points, which cannot fully reflect the thickening of the entire valve, and the thickness of the valve may vary significantly in different regions.

[0089] In this application, by using volume integration for the valve, the changes in thickness on the valve surface are comprehensively integrated to generate a more accurate thickening score. The formula is as follows:

[0090] V thickening = ∫ ValveSurface Thickness(x,y) × dA;

[0091] where V thickening is the total volume of thickening;

[0092] Valve Surface refers to the three-dimensional outer surface of the heart valve;

[0093] Thickness(x,y) is the degree of thickening at a certain point on the valve surface (usually represented by the thickness value measured by imaging);

[0094] dA is the differential area element of the surface.

[0095] Then, based on the total volume of thickening V thickeningCalculate the Thickening_Score using the following formula:

[0096]

[0097] Where V reference is a reference volume, representing the standard volume considered clinically as severely thickened, and can refer to the clinical threshold calculated from the weighted volume in severely thickened cases in the sample data.

[0098] This method can accurately quantify the thickening of the entire valve by performing three-dimensional integration on the valve surface, rather than just local thickness changes; it takes into account the overall distribution of thickening on the valve surface and can reflect the heterogeneity and regional differences of thickening.

[0099] 3. Calculation of the opening and closing score

[0100] In the prior art, the valve opening and closing score is determined by calculating the ratio of the maximum opening area of the valve to the normal opening area, combined with the characteristics of incomplete valve closure:

[0101]

[0102] The above calculation method is based on two-dimensional imaging data, cannot comprehensively reflect the movement of the valve, relies on only one parameter, and ignores the dynamic changes of the entire valve structure. Therefore, in this application, based on a three-dimensional geometric model, the dynamic kinematic parameters of the valve are extracted, so as to be able to evaluate its movement ability. The formula for calculating its ValveOpening_Score is:

[0103] ValveOpening_Score = 100×(α S ·S + α F ·F + α A ·A)

[0104] Where: α S 、α F 、α A are weight coefficients, α S : Synchronization weight, reflecting the importance of leaflet coordination, α F : Deformation flexibility weight, reflecting the influence of valve flexibility, α A : Opening area weight, reflecting the importance of the maximum opening ability of the valve. They satisfy α S +α F +α A = 1; S is the synchronization score, F is the deformation flexibility score, and A is the opening area score, all ranging from 0 to 1.

[0105] Synchronization reflects the degree of coordination among multiple valve leaflets during the opening and closing process. If the movements of the valve leaflets are inconsistent, it may affect the normal function of the valve. The calculation formula for the synchronization score (S) is:

[0106]

[0107] where: θ i (t) is the opening and closing angle of the i-th valve leaflet, is the average opening and closing angle of all valve leaflets, n is the number of valve leaflets, and the range of S is from 0 to 1. The closer it is to 1, the better the synchronization.

[0108] Deformation flexibility measures the softness of the valve leaflets and whether there is excessive deformation or stiffness during the opening and closing process, which affects the function of the valve. The calculation formula for the deformation flexibility score (F) is:

[0109]

[0110] where: C(t) is the curvature of the valve at time t, is the normal curvature of the valve, and the range of F is from 0 to 1. The closer it is to 1, the better the flexibility and deformation state of the valve leaflets.

[0111] The opening area reflects the maximum opening ability of the valve and is a core indicator of the valve function. The calculation formula for the opening area score (A) is:

[0112]

[0113] where: Measured Area is the maximum opening area of the current valve, Normal Area is the standard opening area of a healthy valve, and the range of A is from 0 to 1. The closer it is to 1, the closer the opening area is to normal.

[0114] 4. Calculation of Image Feature Score

[0115] The image feature score (TAS) is generated from the calcification score, thickening score, and opening and closing score. The specific formula is as follows:

[0116] TAS = w1 × calcification score + w2 × thickening score + w3 × opening and closing score

[0117] where w1, w2, and w3 are the weights of each feature score.

[0118] Assume the weights are:

[0119] The weight of the calcification score w1 = 0.4

[0120] The weight of the thickening score w2 = 0.3

[0121] The weight of the opening and closing score w3 = 0.3

[0122] Then, assuming a calcification score of 40, a thickening score of 150, and an opening / closing score of 40, the imaging feature score is:

[0123] TAS = 0.4×40 + 0.3×150 + 0.3×40 = 16 + 45 + 12 = 73.

[0124] Step Four: Physiological parameter calculation: Based on physiological parameters of the patient such as cardiac output CO and transvalvular pressure difference ΔP, calculate the effective orifice area EOA using the following formula:

[0125]

[0126] where CO is the cardiac output, HR is the heart rate, SEP is the effective systolic time, ΔP is the transvalvular pressure difference, and ρ is the blood density.

[0127] Step Five: Stenosis rate calculation: By combining the imaging feature score and physiological parameter calculation, comprehensively calculate the aortic valve stenosis rate using the following formula:

[0128]

[0129] where EOA is the effective orifice area, TAS is the imaging feature score, α and β are the balance coefficients between physiological parameters and imaging features, EOA normal is the normal reference orifice area, and TAS normal is the normal reference score of the imaging features.

[0130] Assume:

[0131] EOA = 0.8 cm 2 , EOA normal = 2.0 cm 2

[0132] TAS normal = 0 (no abnormality in the imaging features of the normal valve)

[0133] α = 0.5, β = 0.5

[0134] Then:

[0135]

[0136] The stenosis rate is 66.5%.

[0137] Step Six: Stenosis rate output: According to the comprehensively calculated stenosis rate, the system outputs a reference diagnosis result of the stenosis degree to provide decision support for doctors.

[0138] Specifically, according to the comprehensively calculated stenosis rate, the system outputs a reference diagnostic result of the stenosis degree to provide decision-making support for doctors. The stenosis degree can be divided into: mild stenosis (SR < 30%), moderate stenosis (30% ≤ SR < 70%), severe stenosis (70% ≤ SR < 99%), and complete occlusion (99% ≤ SR).

[0139] For mild stenosis, emergency intervention may not be required, but regular monitoring is needed.

[0140] For moderate stenosis, medical treatment and lifestyle changes may be required.

[0141] For stenosis above severe, further interventional treatment or surgery may be required.

[0142] In this way, the calculation and output of the stenosis rate not only provide a quantitative stenosis degree index, but also provide a powerful tool for doctors to support their clinical decision-making.

[0143] See the appendix Figure 3 , the present invention proposes a deep learning-based cardiac valve abnormality analysis system, which is used to implement the above-mentioned deep learning-based cardiac valve abnormality analysis method, including the following modules:

[0144] An image and data acquisition module, which is used to obtain the cardiac medical images of the patient and obtain the physiological parameters of the patient, including the cardiac output CO and the transvalvular pressure difference ΔP;

[0145] An image feature extraction module, which is used to automatically extract the key pathological features of the valve based on a convolutional neural network (CNN) model;

[0146] An image feature scoring module, which is used to score the image features and generate an image feature score TAS;

[0147] A physiological parameter calculation module, which is used to calculate the effective orifice area EOA based on the patient's cardiac output CO and transvalvular pressure difference ΔP;

[0148] A stenosis rate calculation module, which is used to calculate the stenosis rate SR of the aortic valve by combining physiological parameters and image feature scores. The calculation formula is as follows:

[0149]

[0150] where EOA is the effective orifice area, TAS is the image feature score, α and β are weight coefficients, EOA normal is the normal reference orifice area of the valve, and TAS normal is the normal reference score of the image features;

[0151] The stenosis rate output module is used to output a reference diagnosis result of the stenosis degree according to the comprehensively calculated stenosis rate.

[0152] The present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned method for analyzing heart valve abnormalities based on deep learning are implemented.

[0153] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for analyzing cardiac valve abnormalities based on deep learning, comprising the following steps: Step 1: Image and data acquisition: Obtain the cardiac images of the patient and acquire the patient's physiological parameters, including cardiac output CO and transvalvular pressure difference ΔP; Step 2: Image feature extraction: Automatically extract the key pathological features of the valve based on the convolutional neural network CNN model; Step 3: Image feature scoring: Score the image features to generate an image feature score TAS; Step 4: Physiological parameter calculation: Calculate the effective orifice area EOA based on the patient's cardiac output CO and transvalvular pressure difference ΔP; Step 5: Stenosis rate calculation: Combine the physiological parameters and the image feature score to calculate the stenosis rate SR of the aortic valve. The calculation formula is as follows: Among them, EOA is the effective valve orifice area, TAS is the imaging feature score, α and β are weight coefficients, and EOA normal is the normal reference valve orifice area, and TAS nornal is the normal reference score of imaging features; Step 6: Stenosis rate output: According to the comprehensively calculated stenosis rate, the system outputs a reference diagnosis result of the stenosis degree.

2. The analysis method according to claim 1, wherein The image features include the degree of calcification, valve thickness, and valve opening and closing status.

3. The analysis method according to claim 2, characterized in that The image feature score TAS is generated from a calcification score, a thickening score, and an opening and closing score. Among them, The calcification score is calculated based on the calcification area combined with the calcification density; The thickening score is calculated based on the valve thickening volume; The opening and closing score is calculated based on the valve dynamic kinematic parameters; TAS = w1 × calcification score + w2 × thickening score + w3 × opening and closing score. Where w1, w2, and w3 are the weights of each feature score.

4. The analysis method according to claim 3, characterized in that The calculation formula for the calcification score (Ca_Score) is as follows: Ca_Score = ∑Area i × (HU i × W i ) where Area i is the area of the i-th calcification spot, HU i is the HU value of the i-th calcification spot, and W i is the density coefficient corresponding to the HU value of the i-th calcification spot.

5. The analysis method according to claim 3, wherein The formula for the thickening score (Thickening_Score) is as follows: V thickening = ∫ ValveSurface Thickness(x,y) × dA; Among them, Valve Surface refers to the three-dimensional outer surface of the heart valve; Thickness(x,y) is the degree of thickening at a certain point on the valve surface; dA is the differential area element of the surface; V thickening is the total volume of thickening; V reference is a reference volume, representing the standard volume considered clinically as severe thickening.

6. The analysis method according to claim 3, characterized in that The calculation formula for the opening and closing score (ValveOpening_Score) is: ValveOpening_Score = 100×(α S ·S + α F ·F + α A ·A) where: α S , α F , α A are weight coefficients, α S : synchronization weight, reflecting the importance of leaflet coordination, α F : deformation flexibility weight, reflecting the influence of valve flexibility, α A : opening area weight, reflecting the importance of the maximum opening ability of the valve, and they satisfy α S +α F +α A = 1; S is the synchronization score, F is the deformation flexibility score, and A is the opening area score, and the ranges are all from 0 to 1; Where: θ i (t) is the opening and closing angle of the i-th leaflet, is the average opening and closing angle of all leaflets, n is the number of leaflets, and the range of S is from 0 to 1; Where: C(t) is the curvature of the valve at time t, is the normal curvature of the valve, and the range of F is from 0 to 1; Where: Measured Area is the maximum opening area of the current valve, and NormalArea is the standard opening area of a healthy valve.

7. The analysis method according to claim 1, wherein The calculation formula for the effective orifice area EOA: Where CO is the cardiac output, HR is the heart rate, SEP is the effective systolic time, ΔP is the transvalvular pressure difference, and ρ is the blood density.

8. A system for analyzing cardiac valve abnormalities based on deep learning, which is used to implement the method for analyzing cardiac valve abnormalities based on deep learning according to any one of claims 1-7, and includes the following modules: An image and data acquisition module, which is used to obtain the cardiac medical images of the patient and acquire the patient's physiological parameters, including cardiac output CO and transvalvular pressure difference ΔP; An image feature extraction module, which is used to automatically extract the key pathological features of the valve based on the convolutional neural network CNN model; An image feature scoring module, which is used to score the image features to generate an image feature score TAS; A physiological parameter calculation module, which is used to calculate the effective orifice area EOA based on the patient's cardiac output CO and transvalvular pressure difference ΔP; A stenosis rate calculation module, which is used to combine the physiological parameters and the image feature score to calculate the stenosis rate SR of the aortic valve. The calculation formula is as follows: Among them, EOA is the effective valve orifice area, TAS is the imaging feature score, α and β are weight coefficients, and EOA normal is the normal reference valve orifice area, and TAS normal is the normal reference score of imaging features; A stenosis rate output module, which is used to output a reference diagnosis result of the stenosis degree according to the comprehensively calculated stenosis rate.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the method for analyzing cardiac valve abnormalities based on deep learning according to any one of claims 1-7 are implemented.

Images