Aortic valve regurgitation degree assessment method, device and medium based on deep learning

Abstract

The present application discloses a method, device and medium for evaluating the degree of aortic valve regurgitation based on deep learning, which is suitable for semi-quantitative evaluation of the degree of aortic valve regurgitation in echocardiography. The method includes preprocessing and labeling parasternal long-axis two-dimensional images and color Doppler images, and training key point prediction models and aortic valve prediction models. The positions of the upper and lower key points of the left ventricular outflow tract are obtained through the parasternal long-axis two-dimensional image, and the aortic valve position information is obtained through the color Doppler image. The left ventricular outflow tract width is calculated based on the Euclidean distance of the key points, and the regurgitant area is cropped to extract the abnormal contour, and the regurgitant beam width and the contraction neck width are obtained. Finally, the regurgitant degree is inferred based on the regurgitant degree grading standard. This method realizes the semi-automatic evaluation of the degree of aortic valve regurgitation, and improves the diagnostic efficiency, accuracy and repeatability.

Description

Technical Field

[0001] The present application relates to the field of medical image processing technology, and in particular to a method, device, and medium for evaluating the degree of aortic valve regurgitation based on deep learning. Background Art

[0002] Aortic regurgitation (AR) is a common heart valve disorder characterized by the failure of the aortic valve to close completely during diastole, resulting in the backflow of blood from the aorta into the left ventricle. This condition can increase the workload on the left ventricle, leading to heart failure and other cardiovascular complications. Effectively assessing the severity of aortic regurgitation is crucial for clinical decision-making, disease monitoring, and treatment planning.

[0003] Traditional methods for assessing aortic regurgitation rely primarily on techniques such as cardiac ultrasound (such as two-dimensional ultrasound and Doppler ultrasound), magnetic resonance imaging (MRI), and cardiac catheterization. While these methods can provide quantitative assessments, they typically require interpretation by specialized physicians and can be affected by various factors during operation, such as patient positioning, physician experience, and device performance. Furthermore, traditional imaging methods are often subjective when assessing the degree of regurgitation, and assessment results can vary between operators, leading to inconsistent clinical decision-making. Summary of the Invention

[0004] The present application provides a method, device and medium for assessing the degree of aortic regurgitation based on deep learning to solve the problems of strong subjectivity, low efficiency and high misdiagnosis rate in the current diagnosis of aortic regurgitation. By extracting key points and valve features through an automated deep learning model, the influence of human factors on the assessment results is reduced, thereby improving the accuracy of the diagnosis. At the same time, the method can quickly process ultrasound images and automatically calculate regurgitation-related parameters, significantly shortening the diagnosis time and improving clinical work efficiency. In addition, clear quantitative indicators (such as left ventricular outflow tract width, contraction neck width and regurgitation jet width) are provided to provide clinicians with a more intuitive evaluation basis and promote more effective treatment decisions. Through early and accurate assessment of the degree of regurgitation, doctors can take timely intervention measures to improve patient prognosis and reduce the incidence of serious complications.

[0005] The technical solution of this application is as follows:

[0006] In one aspect, the present application provides a method for assessing the degree of aortic regurgitation based on deep learning, comprising the following steps:

[0007] Step 1: Perform image preprocessing and image annotation on the parasternal long-axis 2D images to obtain the parasternal long-axis 2D dataset PLAX-2D; perform image preprocessing and image annotation on the parasternal long-axis color Doppler images to obtain the parasternal long-axis color Doppler dataset PLAX-CDFI;

[0008] Step 2: Use the PLAX-2D dataset to train the YOLOv11 model. Use the trained model to extract keypoint features from the target patient's parasternal long-axis 2D image to obtain the positions of the upper and lower keypoints of the left ventricular outflow tract. Calculate the Euclidean distance between the upper and lower keypoints of the left ventricular outflow tract. Calculate the distance on the image based on the Euclidean distance to obtain the left ventricular outflow tract width.

[0009] Step 3: Use the PLAX-CDFI dataset to train a Unet++ model. The trained model is used to extract aortic valve features from the target patient's parasternal long-axis color Doppler images to output aortic valve position information. The regurgitant region of the aortic valve position information is processed to extract the contour of the abnormal regurgitant region. The upper and lower contour points of the systolic neck width are then extracted to calculate the systolic neck width and regurgitant jet width.

[0010] Step 4: Determine the severity of the target patient's heart valve regurgitation based on the target patient's left ventricular outflow tract width, contraction neck width, and regurgitant jet width.

[0011] Furthermore, the step 1 includes:

[0012] Step 1.1: Acquire clear parasternal long-axis 2D images and parasternal long-axis color Doppler images of patients aged 18 years or older. Save patient information to a file, including image ID, left ventricular outflow tract width, systolic neck width, and regurgitant jet width. Preprocess the images to remove all text information, including the patient's name, gender, hospital, and serial number.

[0013] Step 1.2: Label the parasternal long-axis 2D image using LabelImg. Label the upper and lower key points corresponding to the LVOT width and draw a rectangular frame around the LVOT to obtain the parasternal long-axis 2D dataset, PLAX-2D.

[0014] Step 1.3: Label the parasternal long-axis color Doppler image using LabelImg as the annotation tool to annotate the visible aortic valve region and obtain the parasternal long-axis color Doppler dataset PLAX-CDFI.

[0015] Step 1.4: Divide the datasets PLAX-2D and PLAX-CDFI into training, test, and validation sets in proportion.

[0016] Furthermore, the step 2 includes:

[0017] Step 2.1: Use the training set and validation set of the PLAX-2D dataset to train the YOLOv11 deep learning model to obtain an automatic inference model for the upper and lower key points of the left ventricular outflow tract;

[0018] Step 2.2: Calculate the left ventricular outflow tract image ratio;

[0019] Step 2.2.1: Save the upper keypoint coordinates (X1, Y1) and lower keypoint coordinates (X2, Y2) of each image in the PLAX-2D dataset and calculate the Euclidean distance of each image.

[0020] Step 2.2.2: Obtain the actual left ventricular outflow tract width from the document, and divide the left ventricular outflow tract width data into m categories with different depths, where m is the set depth at which the left ventricular outflow tract width is measured, m∈{11cm, 12cm, 13cm, 14cm};

[0021] Step 2.2.3: Find the image ratio corresponding to each depth Where m is the image depth, D i The left ventricular outflow tract width was actually measured for the parasternal long-axis two-dimensional image;

[0022] Step 2.3: Use the automatic inference model of the upper and lower key points of the left ventricular outflow tract to infer the parasternal long axis 2D image, obtain the coordinates of the upper key point (x1, y1), the coordinates of the lower key point (x2, y2), and calculate the Euclidean distance between the upper and lower key points

[0023] Step 2.4: Find the actual distance using Euclidean distance and graph scale Where m is the depth corresponding to the input parasternal long axis 2D image.

[0024] Furthermore, the step 3 includes:

[0025] Step 3.1: Use the training set and validation set of the PLAX-CDFI dataset to train the Unet++ deep learning model to obtain an automatic aortic valve segmentation model;

[0026] Step 3.2: Use the aortic valve automatic segmentation model to infer the parasternal long axis color Doppler image and obtain the coordinate points of the aortic valve region (x i ,y i );

[0027] Step 3.3: Expand the aortic valve area;

[0028] Step 3.3.1: Obtain the horizontal coordinate x1 of the leftmost point and the vertical coordinate y1 of the uppermost point of the aortic valve region, and establish the upper left coordinate point (x1, y1) of the rectangular frame. Obtain the horizontal coordinate x2 of the rightmost point and the vertical coordinate y2 of the lowermost point of the aortic valve region, and establish the lower right coordinate point (x2, y2) of the rectangular frame to establish the minimum rectangular frame.

[0029] Step 3.3.2: h = x2 - x1, w = y2 - y1;

[0030] Step 3.3.3: If h > 2 cm, skip to step 3.3.4; otherwise, skip to step 3.3.5.

[0031] Step 3.3.4: If w > 2 cm, skip to step 3.4, otherwise skip to step 3.3.6;

[0032] Step 3.3.5: Expand the width of the rectangle to 2cm;

[0033] Step 3.3.6: Expand the height of the rectangle to 2cm;

[0034] Step 3.4: Extract the abnormal area rectangle L(x,y,w,h);

[0035] Step 3.5: Use traditional image processing methods to obtain the abnormal area, the width of the contraction neck and the width of the regurgitant jet;

[0036] Step 3.5.1: Convert the read image from BGR color space to HSV color space using color space conversion technology;

[0037] Step 3.5.2: Create the first color mask: Define the lower and upper bounds of the HSV color values ​​and use threshold-based image processing techniques to generate a binary mask. The pixel values ​​in the mask indicate which pixels are within the specified color range and are marked as the target color. Extract the first specified color region.

[0038] Step 3.5.3: Create a second color mask: Define the lower and upper bounds of the HSV color values ​​and use threshold-based image processing techniques to generate a binary mask; the pixel values ​​in the mask indicate which pixels are within the specified color range and are marked as the target color, extracting the second specified color area.

[0039] Step 3.5.4: Use bitwise operation techniques to combine the two masks mask1 and mask2 into a combined mask to extract all target color areas;

[0040] Step 3.5.5: Erosion operation is performed on the combined mask to remove small noise;

[0041] Step 3.5.6: Dilate the eroded mask to restore the size of the target area;

[0042] Step 3.5.7: Extract the mask of the aortic valve region L;

[0043] Step 3.5.8: Convert the processed image to a grayscale image and generate a binary image through thresholding. Next, extract the contour points in the image.

[0044] Step 3.5.9: Divide the contour into n regions according to its width;

[0045] Step 3.5.10: For each divided area, filter out the upper and lower contour points and calculate the distance between them to find the closest point pair;

[0046] Step 3.5.11: The rightmost point pair is the upper and lower contour points corresponding to the vena contracta width. Calculate the Euclidean distance between the two contour points. Calculate the actual distance based on steps 2.2 to 2.3, which is the vena contracta width.

[0047] Step 3.5.12: The length of the line segment corresponding to the line segment with a distance of 1 cm between the line segments corresponding to the width of the vena contracta is the width of the backflow jet.

[0048] Furthermore, the step 4 includes:

[0049] Step 4.1: Calculate RJW_ratio = contraction width / left ventricular outflow tract width;

[0050] Step 4.2: If the RJW_ratio is less than 25% and the contraction width is less than 0.3 cm, the degree of reflux is judged to be mild; if the RJW_ratio is 25% to 64% and the contraction width is 0.3 cm to 0.6 cm, the degree of reflux is judged to be moderate; if the RJW_ratio is greater than 64% and the contraction width is greater than 0.6 cm, the degree of reflux is judged to be severe.

[0051] In another aspect, the present application provides a device for assessing the degree of aortic regurgitation based on deep learning, comprising:

[0052] The data set acquisition unit is used to perform image preprocessing and image annotation on the parasternal long axis two-dimensional image to obtain the parasternal long axis two-dimensional data set PLAX-2D; and perform image preprocessing and image annotation on the parasternal long axis color Doppler image to obtain the parasternal long axis color Doppler data set PLAX-CDFI;

[0053] The first model training and testing unit is used to train the YOLOv11 model using the PLAX-2D dataset. The trained model is used to extract key point features from the target patient's parasternal long-axis two-dimensional image to obtain the positions of the upper and lower key points of the left ventricular outflow tract; the Euclidean distance between the upper and lower key points of the left ventricular outflow tract is calculated, and the distance on the map is calculated based on the Euclidean distance to obtain the left ventricular outflow tract width;

[0054] The first model training and testing unit is used to train the Unet++ model using the PLAX-CDFI dataset. The trained model is used to extract aortic valve features from the target patient's parasternal long-axis color Doppler image to output aortic valve position information. The regurgitant region of the aortic valve position information is processed to extract the contour of the abnormal regurgitant region, and then the upper and lower contour points of the systolic neck width are extracted to calculate the systolic neck width and the regurgitant jet width.

[0055] The heart valve regurgitation severity judgment unit is used to judge the severity of the heart valve regurgitation of the target patient according to the left ventricular outflow tract width, the contraction neck width and the regurgitation jet width of the target patient.

[0056] On the other hand, the present application provides a cardiac image data processing terminal, including a processor and a memory, wherein the memory stores a computer-readable program, and when the computer-readable program is called and executed by the processor, the method described above is implemented.

[0057] On the other hand, the present application provides a computer-readable medium, which stores a computer-readable program. When the computer-readable program is called and executed by a computer, it implements the method described above.

[0058] In summary, the beneficial effects of the present application are: First, the use of deep learning models (such as YOLOv11 and Unet++) significantly improves the accuracy of reflux degree assessment and reduces errors caused by human factors. Secondly, the rapid processing of ultrasound images and automatic calculation of reflux parameters greatly improve the diagnostic efficiency and reduce the time doctors spend on image analysis. In addition, the quantitative indicators provided (such as left ventricular outflow tract width, contraction neck width and regurgitant jet width) enable clinicians to more intuitively assess the patient's condition and support the formulation of scientific treatment plans. Finally, by identifying potential heart problems early and reducing the misdiagnosis rate, patient safety and overall treatment effects are enhanced. Therefore, the present invention provides significant gains for the diagnosis and management of aortic valve regurgitation and has important clinical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 Overall flow chart of the semi-quantitative assessment method for the degree of aortic regurgitation

[0060] Figure 2Flowchart for prediction of contraction neck width and backflow jet width

[0061] Figure 3 Prediction results of upper and lower key points of left ventricular outflow tract

[0062] Figure 4 Aortic valve segmentation results DETAILED DESCRIPTION

[0063] The specific implementation of the present application is described in detail below with reference to the accompanying drawings.

[0064] A specific embodiment of the present application provides a method for evaluating the degree of aortic regurgitation based on deep learning, such as Figure 1 , including the following steps:

[0065] Step 1: Perform image preprocessing and image annotation on the parasternal long-axis 2D images to obtain the parasternal long-axis 2D dataset PLAX-2D; perform image preprocessing and image annotation on the parasternal long-axis color Doppler images to obtain the parasternal long-axis color Doppler dataset PLAX-CDFI;

[0066] Step 2: Use the PLAX-2D dataset to train the YOLOv11 model. Use the trained model to extract keypoint features from the target patient's parasternal long-axis 2D image to obtain the positions of the upper and lower keypoints of the left ventricular outflow tract. Calculate the Euclidean distance between the upper and lower keypoints of the left ventricular outflow tract. Calculate the distance on the image based on the Euclidean distance to obtain the left ventricular outflow tract width.

[0067] Step 3: Use the PLAX-CDFI dataset to train a Unet++ model. The trained model is used to extract aortic valve features from the target patient's parasternal long-axis color Doppler images to output aortic valve position information. The regurgitant region of the aortic valve position information is processed to extract the contour of the abnormal regurgitant region. The upper and lower contour points of the systolic neck width are then extracted to calculate the systolic neck width and regurgitant jet width.

[0068] Step 4: Determine the severity of the target patient's heart valve regurgitation based on the target patient's left ventricular outflow tract width, contraction neck width, and regurgitant jet width.

[0069] Furthermore, the step 1 includes:

[0070] Step 1.1: Acquire clear parasternal long-axis 2D images and parasternal long-axis color Doppler images of patients aged 18 years or older. Save patient information to a file, including image ID, left ventricular outflow tract width, systolic neck width, and regurgitant jet width. Preprocess the images to remove all text information, including the patient's name, gender, hospital, and serial number.

[0071] Step 1.2: Label the parasternal long-axis 2D image using LabelImg. Label the upper and lower key points corresponding to the LVOT width and draw a rectangular frame around the LVOT to obtain the parasternal long-axis 2D dataset, PLAX-2D.

[0072] Step 1.3: Label the parasternal long-axis color Doppler image using LabelImg as the annotation tool to annotate the visible aortic valve region and obtain the parasternal long-axis color Doppler dataset PLAX-CDFI.

[0073] Step 1.4: Divide the datasets PLAX-2D and PLAX-CDFI into training, test, and validation sets in proportion.

[0074] The step 2 includes:

[0075] Step 2.1: Use the training set and validation set of the dataset PLAX-2D to train the YOLOv11 deep learning model to obtain the automatic reasoning model of the upper and lower key points of the left ventricular outflow tract; use the automatic reasoning model of the upper and lower key points of the left ventricular outflow tract to extract key point features from the parasternal long axis two-dimensional image of the target patient. The recognition results are as follows: Figure 3 shown.

[0076] Step 2.2: Calculate the left ventricular outflow tract image ratio;

[0077] Step 2.2.1: Save the upper keypoint coordinates (X1, Y1) and lower keypoint coordinates (X2, Y2) of each image in the PLAX-2D dataset and calculate the Euclidean distance of each image.

[0078] Step 2.2.2: Obtain the actual left ventricular outflow tract width from the document, and divide the left ventricular outflow tract width data into m categories with different depths, where m is the set depth at which the left ventricular outflow tract width is measured, m∈{11cm, 12cm, 13cm, 14cm};

[0079] Step 2.2.3: Find the image ratio corresponding to each depth Where m is the image depth, D i The left ventricular outflow tract width was actually measured for the parasternal long-axis two-dimensional image;

[0080] Step 2.3: Use the automatic inference model of the upper and lower key points of the left ventricular outflow tract to infer the parasternal long axis 2D image, obtain the coordinates of the upper key point (x1, y1), the coordinates of the lower key point (x2, y2), and calculate the Euclidean distance between the upper and lower key points

[0081] Step 2.4: Find the actual distance using Euclidean distance and graph scale Where m is the depth corresponding to the input parasternal long axis 2D image.

[0082] like Figure 2 , the step 3 comprises:

[0083] Step 3.1: Use the training set and validation set of the PLAX-CDFI dataset to train the Unet++ deep learning model to obtain an automatic aortic valve segmentation model;

[0084] Step 3.2: Use the aortic valve automatic segmentation model to infer the parasternal long axis color Doppler image. The result is as follows Figure 4 As shown, the coordinate point of the aortic valve region (x i ,y i );

[0085] Step 3.3: Expand the aortic valve area;

[0086] Step 3.3.1: Obtain the horizontal coordinate x1 of the leftmost point and the vertical coordinate y1 of the uppermost point of the aortic valve region, and establish the upper left coordinate point (x1, y1) of the rectangular frame. Obtain the horizontal coordinate x2 of the rightmost point and the vertical coordinate y2 of the lowermost point of the aortic valve region, and establish the lower right coordinate point (x2, y2) of the rectangular frame to establish the minimum rectangular frame.

[0087] Step 3.3.2: h = x2 - x1, w = y2 - y1;

[0088] Step 3.3.3: If h > 2 cm, skip to step 3.3.4; otherwise, skip to step 3.3.5.

[0089] Step 3.3.4: If w > 2 cm, skip to step 3.4, otherwise skip to step 3.3.6;

[0090] Step 3.3.5: Expand the width of the rectangle to 2cm;

[0091] Step 3.3.6: Expand the height of the rectangle to 2cm;

[0092] Step 3.4: Extract the abnormal area rectangle L(x,y,w,h);

[0093] Step 3.5: Use traditional image processing methods to obtain the abnormal area, the width of the contraction neck and the width of the regurgitant jet;

[0094] Step 3.5.1: Convert the read image from BGR color space to HSV color space using color space conversion technology;

[0095] Step 3.5.2: Create the first color mask: Define the lower and upper bounds of the HSV color values ​​and use threshold-based image processing techniques to generate a binary mask. The pixel values ​​in the mask indicate which pixels are within the specified color range and are marked as the target color. Extract the first specified color region.

[0096] Step 3.5.3: Create a second color mask: Define the lower and upper bounds of the HSV color values ​​and use threshold-based image processing techniques to generate a binary mask; the pixel values ​​in the mask indicate which pixels are within the specified color range and are marked as the target color, extracting the second specified color area.

[0097] Step 3.5.4: Use bitwise operation techniques to combine the two masks mask1 and mask2 into a combined mask to extract all target color areas;

[0098] Step 3.5.5: Erosion operation is performed on the combined mask to remove small noise;

[0099] Step 3.5.6: Dilate the eroded mask to restore the size of the target area;

[0100] Step 3.5.7: Extract the mask of the aortic valve region L;

[0101] Step 3.5.8: Convert the processed image to a grayscale image and generate a binary image through thresholding. Next, extract the contour points in the image.

[0102] Step 3.5.9: Divide the contour into n regions according to its width;

[0103] Step 3.5.10: For each divided area, filter out the upper and lower contour points and calculate the distance between them to find the closest point pair;

[0104] Step 3.5.11: The rightmost point pair is the upper and lower contour points corresponding to the vena contracta width. Calculate the Euclidean distance between the two contour points. Calculate the actual distance based on steps 2.2 to 2.3, which is the vena contracta width.

[0105] Step 3.5.12: The length of the line segment corresponding to the line segment with a distance of 1 cm (actual distance) from the width of the contraction neck is the width of the reverse flow beam.

[0106] The step 4 comprises:

[0107] Step 4.1: Calculate RJW_ratio = contraction width / left ventricular outflow tract width;

[0108] Step 4.2: If the RJW_ratio is less than 25% and the systolic width is less than 0.3 cm, the regurgitation is considered mild. If the RJW_ratio is between 25% and 64% and the systolic width is between 0.3 cm and 0.6 cm, the regurgitation is considered moderate. If the RJW_ratio is greater than 64% and the systolic width is greater than 0.6 cm, the regurgitation is considered severe. The following table lists the grading criteria for aortic regurgitation.

[0109] Aortic regurgitation severity grading standard

[0110]

[0111] Another specific embodiment of the present application provides a device for assessing the degree of aortic regurgitation based on deep learning, comprising:

[0112] The data set acquisition unit is used to perform image preprocessing and image annotation on the parasternal long axis two-dimensional image to obtain the parasternal long axis two-dimensional data set PLAX-2D; and perform image preprocessing and image annotation on the parasternal long axis color Doppler image to obtain the parasternal long axis color Doppler data set PLAX-CDFI;

[0113] The first model training and testing unit is used to train the YOLOv11 model using the PLAX-2D dataset. The trained model is used to extract key point features from the target patient's parasternal long-axis two-dimensional image to obtain the positions of the upper and lower key points of the left ventricular outflow tract; the Euclidean distance between the upper and lower key points of the left ventricular outflow tract is calculated, and the distance on the map is calculated based on the Euclidean distance to obtain the left ventricular outflow tract width;

[0114] The first model training and testing unit is used to train the Unet++ model using the PLAX-CDFI dataset. The trained model is used to extract aortic valve features from the target patient's parasternal long-axis color Doppler image to output aortic valve position information. The regurgitant region of the aortic valve position information is processed to extract the contour of the abnormal regurgitant region, and then the upper and lower contour points of the systolic neck width are extracted to calculate the systolic neck width and the regurgitant jet width.

[0115] The heart valve regurgitation severity judgment unit is used to judge the severity of the heart valve regurgitation of the target patient according to the left ventricular outflow tract width, the contraction neck width and the regurgitation jet width of the target patient.

[0116] Another specific embodiment of the present application provides a cardiac image data processing terminal, including a processor and a memory, wherein the memory stores a computer-readable program, and when the computer-readable program is called and executed by the processor, the method described above is implemented.

[0117] Another specific embodiment of the present application provides a computer-readable medium, which stores a computer-readable program. When the computer-readable program is called and executed by a computer, the method described above is implemented.

[0118] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this field, several variations and improvements can be made without departing from the creative concept of the present application, and these all fall within the scope of protection of the present application.

Claims

1. A method for assessing the degree of aortic regurgitation based on deep learning, characterized in that: The steps include: Step 1: Perform image preprocessing and image annotation on the parasternal long-axis 2D images to obtain the parasternal long-axis 2D dataset PLAX-2D; perform image preprocessing and image annotation on the parasternal long-axis color Doppler images to obtain the parasternal long-axis color Doppler dataset PLAX-CDFI; Step 2: Use the PLAX-2D dataset to train the YOLOv11 model. Use the trained model to extract keypoint features from the target patient's parasternal long-axis 2D image to obtain the positions of the upper and lower keypoints of the left ventricular outflow tract. Calculate the Euclidean distance between the upper and lower keypoints of the left ventricular outflow tract. Calculate the distance on the image based on the Euclidean distance to obtain the left ventricular outflow tract width. Step 3: Use the PLAX-CDFI dataset to train a Unet++ model. The trained model is used to extract aortic valve features from the target patient's parasternal long-axis color Doppler images to output aortic valve position information. The regurgitant region of the aortic valve position information is processed to extract the contour of the abnormal regurgitant region. The upper and lower contour points of the systolic neck width are then extracted to calculate the systolic neck width and regurgitant jet width. Step 4: Determine the severity of the target patient's heart valve regurgitation based on the target patient's left ventricular outflow tract width, contraction neck width, and regurgitant jet width.

2. The method for assessing aortic regurgitation based on deep learning according to claim 1, characterized in that: The step 1 comprises: Step 1.1: Acquire clear parasternal long-axis 2D images and parasternal long-axis color Doppler images of patients aged 18 years or older. Save patient information to a file, including image ID, left ventricular outflow tract width, systolic neck width, and regurgitant jet width. Preprocess the images to remove all text information, including the patient's name, gender, hospital, and serial number. Step 1.2: Label the parasternal long-axis 2D image using LabelImg. Label the upper and lower key points corresponding to the LVOT width and draw a rectangular frame around the LVOT to obtain the parasternal long-axis 2D dataset, PLAX-2D. Step 1.3: Label the parasternal long-axis color Doppler image using LabelImg as the annotation tool to annotate the visible aortic valve region and obtain the parasternal long-axis color Doppler dataset PLAX-CDFI. Step 1.4: Divide the datasets PLAX-2D and PLAX-CDFI into training, test, and validation sets in proportion.

3. The method for assessing aortic regurgitation based on deep learning according to claim 2, characterized in that: The step 2 includes: Step 2.1: Use the training set and validation set of the PLAX-2D dataset to train the YOLOv11 deep learning model to obtain an automatic inference model for the upper and lower key points of the left ventricular outflow tract; Step 2.2: Calculate the left ventricular outflow tract image ratio; Step 2.2.1: Save the upper keypoint coordinates (X1, Y1) and lower keypoint coordinates (X2, Y2) of each image in the PLAX-2D dataset and calculate the Euclidean distance of each image. Step 2.2.2: Obtain the actual left ventricular outflow tract width from the document, and divide the left ventricular outflow tract width data into m categories with different depths, where m is the set depth at which the left ventricular outflow tract width is measured, m∈{11cm, 12cm, 13cm, 14cm}; Step 2.2.3: Find the image ratio corresponding to each depth Where m is the image depth, D i The left ventricular outflow tract width was actually measured for the parasternal long-axis two-dimensional image; Step 2.3: Use the automatic inference model of the upper and lower key points of the left ventricular outflow tract to infer the parasternal long axis 2D image, obtain the coordinates of the upper key point (x1, y1), the coordinates of the lower key point (x2, y2), and calculate the Euclidean distance between the upper and lower key points Step 2.4: Find the actual distance using Euclidean distance and graph scale Where m is the depth corresponding to the input parasternal long axis 2D image.

4. The method for assessing aortic regurgitation based on deep learning according to claim 3, characterized in that: The step 3 comprises: Step 3.1: Use the training set and validation set of the PLAX-CDFI dataset to train the Unet++ deep learning model to obtain an automatic aortic valve segmentation model; Step 3.2: Use the aortic valve automatic segmentation model to infer the parasternal long axis color Doppler image and obtain the coordinate points of the aortic valve region (x i ,y i ); Step 3.3: Expand the aortic valve area; Step 3.3.1: Obtain the horizontal coordinate x1 of the leftmost point and the vertical coordinate y1 of the uppermost point of the aortic valve region, and establish the upper left coordinate point (x1, y1) of the rectangular frame. Obtain the horizontal coordinate x2 of the rightmost point and the vertical coordinate y2 of the lowermost point of the aortic valve region, and establish the lower right coordinate point (x2, y2) of the rectangular frame to establish the minimum rectangular frame. Step 3.3.2: h = x2 - x1, w = y2 - y1; Step 3.3.3: If h > 2 cm, skip to step 3.3.4; otherwise, skip to step 3.3.

5. Step 3.3.4: If w > 2 cm, skip to step 3.4, otherwise skip to step 3.3.6; Step 3.3.5: Expand the width of the rectangle to 2cm; Step 3.3.6: Expand the height of the rectangle to 2cm; Step 3.4: Extract the abnormal area rectangle L(x,y,w,h); Step 3.5: Use traditional image processing methods to obtain the abnormal area, the width of the contraction neck and the width of the regurgitant jet; Step 3.5.1: Convert the read image from BGR color space to HSV color space using color space conversion technology; Step 3.5.2: Create the first color mask: Define the lower and upper bounds of the HSV color values ​​and use threshold-based image processing techniques to generate a binary mask. The pixel values ​​in the mask indicate which pixels are within the specified color range and are marked as the target color. Extract the first specified color region. Step 3.5.3: Create a second color mask: Define the lower and upper bounds of the HSV color values ​​and use threshold-based image processing techniques to generate a binary mask. The pixel values ​​in the mask indicate which pixels are within the specified color range and are marked as the target color. Extract the second specified color region. Step 3.5.4: Use bitwise operation techniques to combine the two masks mask1 and mask2 into a combined mask to extract all target color areas; Step 3.5.5: Erosion operation is performed on the combined mask to remove small noise; Step 3.5.6: Dilate the eroded mask to restore the size of the target area; Step 3.5.7: Extract the mask of the aortic valve region L; Step 3.5.8: Convert the processed image to a grayscale image and generate a binary image through thresholding. Next, extract the contour points in the image. Step 3.5.9: Divide the contour into n regions according to its width; Step 3.5.10: For each divided area, filter out the upper and lower contour points and calculate the distance between them to find the closest point pair; Step 3.5.11: The rightmost point pair is the upper and lower contour points corresponding to the vena contracta width. Calculate the Euclidean distance between the two contour points. Calculate the actual distance based on steps 2.2 to 2.3, which is the vena contracta width. Step 3.5.12: The length of the line segment corresponding to the line segment with a distance of 1 cm between the line segments corresponding to the width of the vena contracta is the width of the reverse flow jet.

5. The method for assessing aortic regurgitation based on deep learning according to claim 4, characterized in that: The step 4 comprises: Step 4.1: Calculate RJW_ratio = contraction width / left ventricular outflow tract width; Step 4.2: If the RJW_ratio is less than 25% and the contraction width is less than 0.3 cm, the degree of reflux is judged to be mild; if the RJW_ratio is 25% to 64% and the contraction width is 0.3 cm to 0.6 cm, the degree of reflux is judged to be moderate; if the RJW_ratio is greater than 64% and the contraction width is greater than 0.6 cm, the degree of reflux is judged to be severe.

6. A device for assessing the degree of aortic regurgitation based on deep learning, characterized in that: include: The data set acquisition unit is used to perform image preprocessing and image annotation on the parasternal long axis two-dimensional image to obtain the parasternal long axis two-dimensional data set PLAX-2D; and perform image preprocessing and image annotation on the parasternal long axis color Doppler image to obtain the parasternal long axis color Doppler data set PLAX-CDFI; The first model training and testing unit is used to train the YOLOv11 model using the PLAX-2D dataset, and use the trained model to extract key point features from the target patient's parasternal long-axis two-dimensional image to obtain the positions of the upper and lower key points of the left ventricular outflow tract; Calculate the Euclidean distance of the upper and lower key points of the left ventricular outflow tract, calculate the distance on the map based on the Euclidean distance, and obtain the left ventricular outflow tract width; The first model training and testing unit is used to train the Unet++ model using the PLAX-CDFI dataset. The trained model is used to extract aortic valve features from the target patient's parasternal long-axis color Doppler image to output aortic valve position information. The regurgitant region of the aortic valve position information is processed to extract the contour of the abnormal regurgitant region, and then the upper and lower contour points of the systolic neck width are extracted to calculate the systolic neck width and the regurgitant jet width. The heart valve regurgitation severity judgment unit is used to judge the severity of the heart valve regurgitation of the target patient according to the left ventricular outflow tract width, the contraction neck width and the regurgitation jet width of the target patient.

7. A cardiac image data processing terminal, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer-readable program, and when the computer-readable program is called and executed by the processor, the method according to any one of claims 1 to 5 is implemented.

8. A computer-readable medium, characterized in that The computer-readable medium stores a computer-readable program, and when the computer-readable program is called and executed by a computer, the method according to any one of claims 1 to 5 is implemented.

Images