A prognostic method and system for aortic arch restenosis

Abstract

The present invention proposes a prognosis method and system for aortic arch restenosis, which relates to the technical field of prognosis of heart disease, and solves the problem that the current prognosis method for aortic arch restenosis that relies on treatment experience ignores the characteristic data of the aortic arch image, resulting in poor prognosis effect. A three-dimensional CT heart image is obtained, a segmentation data neural network model is constructed and trained, the three-dimensional CT heart image is input into the trained segmentation data neural network model for segmentation, the center line of the three-dimensional aortic arch image is extracted from the segmentation result, and then the part to be measured is measured at the measurement point to obtain the characteristic data of the image, a regression model is constructed based on the selected characteristic data, and the regression model is used to evaluate the experimental data to be detected. The present invention does not rely on treatment experience, combines the characteristic data on the aortic arch image obtained by computer segmentation to evaluate the prognosis detection of aortic arch restenosis, and has a good prognosis effect.

Description

Technical Field

[0001] The present invention relates to the technical field of heart disease prognosis, and in particular to a method and system for prognosing aortic arch restenosis. Background Art

[0002] Coarctation of the aorta refers to localized narrowing of the aorta, which causes a reduction in the lumen and reduces blood flow. It is a common congenital heart disease, accounting for about 4-8% of all congenital heart diseases. With the development of surgical technology, the surgical effect of coarctation of the aorta is generally satisfactory. Due to inherent risk factors such as low age and weight, severe dysplasia of the arch, reconstruction methods of the aortic arch, and associated complex malformations, it is not uncommon for the aortic arch to re-coarct after surgery. Re-coarctation of the aortic arch is an important factor leading to long-term hypertension and seriously affects the long-term prognosis of patients with coarctation of the aortic arch.

[0003] Existing methods for evaluating aortic coarctation mainly use machine learning models to segment cardiac CT images to obtain aortic images, then measure the aortic images, calculate the characteristic data of the measured cardiac CT images, and obtain the aortic diameter ratio. The severity of aortic coarctation is evaluated based on the aortic diameter ratio, but it is impossible to evaluate the occurrence of re-coarctation of the aortic arch. Currently, there is limited research on the prognosis of aortic arch re-coarctation, and the characteristics of the disease that causes aortic arch re-coarctation are still unclear. The prognosis of aortic arch re-coarctation is mainly based on traditional medical clinical data and relies on treatment experience, but ignores the characteristic data on the aortic arch image, resulting in poor prognosis. Summary of the invention

[0004] In order to solve the problem that the current prognostic method for aortic arch restenosis relies on treatment experience and ignores the characteristic data of the aortic arch image, resulting in poor prognostic effect, the present invention proposes a prognostic method and system for aortic arch restenosis, which does not rely on treatment experience, combines the characteristic data on the aortic arch image obtained by computer segmentation to evaluate the prognosis detection of aortic arch restenosis, and has a good prognostic effect.

[0005] In order to achieve the above technical effects, the technical solution of the present invention is as follows:

[0006] A method for prognosis of aortic arch restenosis comprises the following steps:

[0007] S1. Acquire a certain number of three-dimensional CT cardiac images and divide them to obtain a first image data set and a second image data set;

[0008] S2. Constructing a segmentation data neural network model, and training the segmentation data neural network model using the first image data set to obtain a trained segmentation data neural network model;

[0009] S3. Input the three-dimensional CT cardiac image in the second image data set into the trained segmentation data neural network model, and output the segmentation result of the three-dimensional CT cardiac image in the second image data set;

[0010] S4. extracting a three-dimensional aortic arch image of the heart from the segmentation result, and extracting a center line in the three-dimensional aortic arch image;

[0011] S5. Determine the part to be measured in the three-dimensional aortic arch image, collect measurement points of the part to be measured on the center line, and then measure the part to be measured at the measurement points to obtain the characteristic data of the three-dimensional CT heart image described in S3, and pre-process the characteristic data;

[0012] S6. Acquire physiological characteristic data of volunteers, use the data as medical clinical characteristic data, merge the medical clinical characteristic data with the characteristic data of three-dimensional CT cardiac images, and select characteristic data for evaluating the prognosis of aortic arch restenosis from the merged characteristic data;

[0013] S7. construct a regression model using the selected feature data;

[0014] S8. Obtain the experimental data to be tested, input the experimental data to be tested into the constructed regression model, test the prognosis of aortic arch restenosis, and evaluate the prognosis detection of aortic arch restenosis.

[0015] In the present technical solution, a certain number of three-dimensional CT heart images are first divided into a first image data set and a second image data set, and a segmentation data neural network model is constructed. The segmentation data neural network model is trained using the first image data set to obtain a segmentation data neural network model with high segmentation accuracy. The three-dimensional CT heart images in the second image data set are then input into the segmentation data neural network model for segmentation. A three-dimensional aortic arch image of the heart is extracted from the segmentation result, a center line is extracted from the three-dimensional aortic arch image, and measurement points of the part to be measured are collected on the center line for measurement to obtain feature data of the three-dimensional CT heart image in the second image data set, thereby avoiding ignoring the feature data of the aortic arch image. , which leads to the disadvantage of poor prognosis. The feature data of three-dimensional CT cardiac images are merged with the feature data of medical clinics, the merged feature data are selected, and then the regression model is constructed using the selected feature data. Since there are many merged feature data, selecting effective feature data can make the regression model have a good regression effect, and the detection accuracy of the prognosis of aortic arch restenosis is high. Finally, the regression model is used to detect the prognosis of aortic arch restenosis on the experimental data to be tested, and the prognosis detection of aortic arch restenosis is evaluated. This scheme does not rely on treatment experience, but combines the feature data on the aortic arch image obtained by computer segmentation to evaluate the prognosis detection of aortic arch restenosis, and has a good prognosis effect.

[0016] Preferably, in step S2, the specific steps of training the segmentation data neural network model are:

[0017] S21. uniformly sampling the size of the three-dimensional CT cardiac image in the first image data set to the same size;

[0018] S22. Set the number of training batches batchsize, learning rate and training round n, and input the three-dimensional CT heart images of the same size into the constructed segmentation data neural network model;

[0019] S23. Use the gradient descent training method to train the segmentation data neural network model, and when the training round reaches n, output the trained segmentation data neural network model.

[0020] Preferably, in step S4, the step of extracting the center line is:

[0021] S41. Determine the blood vessel position of the aortic arch by a three-dimensional aortic arch image;

[0022] S42. Skeletonize the blood vessels of the aortic arch;

[0023] S43. Based on the skeletonized blood vessel, the center of the blood vessel is determined, and several lines extending along the blood vessel toward the center are determined, and the several lines are set as center lines.

[0024] Here, there are multiple blood vessels in the three-dimensional aortic arch image, and the blood vessel positions are determined to facilitate finding the target of centerline extraction. Since the blood vessels in the segmented three-dimensional aortic arch image are hollow in the middle, it is necessary to skeletonize the blood vessels of the aortic arch to facilitate extracting the centerline from the blood vessels of the aortic arch.

[0025] Preferably, in step S5, the parts to be measured are the ascending aorta, the proximal aorta, the distal aorta, the isthmus, the constriction and the descending aorta, which plays a role in clarifying the measurement parts.

[0026] Preferably, in step S5, the step of collecting the measurement points includes:

[0027] S51. Obtaining a starting point and an end point of a center line along the direction of a blood vessel with the largest cross-sectional area in the three-dimensional aortic arch image, determining a connection path between the starting point and the end point according to the direction of the blood vessel between the starting point and the end point, and setting the connection path as a main path of the center line in the three-dimensional aortic arch image, wherein the main path is composed of a plurality of voxel points in the three-dimensional aortic arch image;

[0028] S52. Obtain all voxel points on the trunk path, determine the number of center lines connected to each voxel point, that is, the degree of the voxel point, and obtain the cross-sectional area of ​​the blood vessel connected to the voxel point at each voxel point;

[0029] S53. Compare the cross-sectional areas of the blood vessels connected to each voxel point to obtain the point of the voxel with the smallest cross-sectional area on the trunk path;

[0030] S54. Set the starting point as the measurement point of the ascending aorta, set the end point as the measurement point of the descending aorta, collect all voxel points with a degree of n on the main path, set all voxel points with a degree of n on the main path as the measurement points of the proximal aorta, the distal aorta and the isthmus, and set the point with the smallest cross-sectional area on the main path as the measurement point of the stenosis.

[0031] Here, the ascending aorta, proximal aorta, distal aorta, isthmus, constriction and descending aorta all have corresponding measurement points. The collection of measurement points facilitates the measurement of characteristic data of three-dimensional CT cardiac images at the above-mentioned measurement locations, making the measurement accuracy more accurate and obtaining reliable measurement of characteristic data of three-dimensional CT cardiac images.

[0032] Preferably, in step S5, the measuring step is to obtain a cross-section of the blood vessel at the positions of the ascending aorta, the proximal aorta, the distal aorta, the isthmus, the stenosis and the descending aorta, extract the contour of the cross-section of the blood vessel using the Sobel operator, fit the contour of the cross-section into an ellipse using the least squares method, measure the major axis, minor axis, diameter, circumference and area of ​​the ellipse, and use the measurement results as the major axis, minor axis, diameter, circumference and area of ​​the contour of the cross-section corresponding to the ellipse.

[0033] Here, cross-sections of the blood vessels are obtained at the positions of the ascending aorta, proximal aorta, distal aorta, isthmus, coarctation and descending aorta. Since the cross-section is an image, it cannot be measured directly. It is necessary to extract the contour of the cross-section of the blood vessel, and then fit the contour of the cross-section into an ellipse using the least squares method. The major axis, minor axis, diameter, circumference and area of ​​the ellipse are measured, and the above measurement data are used as the characteristic data of the measurement part.

[0034] Preferably, in step S7, the regression model is constructed using a random forest algorithm.

[0035] Preferably, in step S8, the experimental data to be detected include feature data of three-dimensional CT heart images measured manually by a doctor, feature data of three-dimensional CT heart images measured by a computer, feature data combining feature data of three-dimensional CT heart images measured by a computer with feature data from medical clinics, feature data from medical clinics, and feature data combining feature data of three-dimensional CT heart images measured manually by a doctor with feature data from medical clinics.

[0036] Preferably, in step S8, the evaluation process adopts a five-fold cross validation method, randomly dividing the experimental data to be tested into a training set and a test set five times, using the training set to train the regression model, and using the test set to evaluate the prognosis detection of aortic arch restenosis by the regression model.

[0037] In addition, the present invention also discloses a prognosis system for aortic arch restenosis, comprising the following system:

[0038] An image data set acquisition module, used for acquiring a certain number of three-dimensional CT cardiac images and dividing them to obtain a first image data set and a second image data set;

[0039] A segmentation data neural network model construction processing module is used to construct a segmentation data neural network model, and train the segmentation data neural network model using the first image data set to obtain a trained segmentation data neural network model;

[0040] a segmentation module, used for inputting the three-dimensional CT heart image in the second image data set into the trained segmentation data neural network model, and outputting the segmentation result of the three-dimensional CT heart image in the second image data set;

[0041] An extraction module, used for extracting an aortic arch image of the heart from the segmentation result, and extracting a center line in the aortic arch image;

[0042] a measurement processing module, used for determining a portion to be measured in the aortic arch image, collecting measurement points of the portion to be measured on the center line, and then measuring the portion to be measured at the measurement points, obtaining feature data of the three-dimensional CT heart image in the second image data set, and preprocessing the feature data;

[0043] A feature data selection module is used to obtain physiological feature data of volunteers, use the data as medical clinical feature data, merge the medical clinical feature data with feature data of three-dimensional CT cardiac images, and select feature data for evaluating the prognosis of aortic arch restenosis from the merged feature data;

[0044] A regression model building module, used to build a regression model using selected feature data;

[0045] The evaluation module is used to obtain the experimental data to be detected, input the experimental data to be detected into the constructed regression model, detect the prognosis of aortic arch restenosis, and evaluate the prognosis detection of aortic arch restenosis.

[0046] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0047] The present invention provides an aortic arch restenosis prognosis method and system. First, a certain number of three-dimensional CT heart images are divided into a first image data set and a second image data set, and a segmentation data neural network model is constructed. The segmentation data neural network model is trained using the first image data set to obtain a segmentation data neural network model with high segmentation accuracy. Then, the three-dimensional CT heart image in the second image data set is input into the segmentation data neural network model for segmentation. The three-dimensional aortic arch image of the heart is extracted from the segmentation result. The center line is extracted from the three-dimensional aortic arch image. The measurement points of the to-be-measured part are collected on the center line for measurement to obtain feature data of the three-dimensional CT heart image in the second image data set. The drawback of ignoring the feature data of the aortic arch image, which leads to a poor prognosis effect, is avoided. Feature data obtained by combining the feature data of the three-dimensional CT heart image with the feature data of medical clinic is selected. Then, a regression model is used to detect the prognosis of aortic arch restenosis on the experimental data to be detected, and the prognosis detection of aortic arch restenosis is evaluated. The scheme does not rely on treatment experience, but combines the feature data on the aortic arch image obtained by computer segmentation to evaluate the prognosis detection of aortic arch restenosis, and has a good prognosis effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram showing a flow chart of a prognosis method for aortic arch restenosis proposed in Example 1 of the present invention;

[0049] Figure 2 A schematic diagram showing the centerline position of the aortic arch proposed in Example 2 of the present invention;

[0050] Figure 3 ROC curve graph showing the characteristic data of 3D CT cardiac images measured manually by doctors;

[0051] Figure 4 A ROC curve diagram representing characteristic data of a three-dimensional CT cardiac image measured by a computer;

[0052] Figure 5 ROC curve diagram representing the characteristic data of medical clinical practice;

[0053] Figure 6 A ROC curve diagram showing the characteristic data of the three-dimensional CT cardiac image measured manually by the doctor and the characteristic data of the medical clinic combined;

[0054] Figure 7 A ROC curve diagram showing characteristic data obtained by combining characteristic data of a three-dimensional CT cardiac image measured by a computer with characteristic data from medical clinics;

[0055] Figure 8 A structural diagram showing a prognostic system for aortic arch restenosis proposed in Example 4 of the present invention. DETAILED DESCRIPTION

[0056] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0057] In order to better illustrate the present embodiment, some parts of the drawings may be omitted, enlarged or reduced, and do not represent actual sizes. The description of the directions of parts such as "upper" and "lower" does not limit the present patent;

[0058] It is understandable to those skilled in the art that some well-known contents may be omitted in the drawings;

[0059] The positional relationships described in the drawings are only for illustrative purposes and should not be construed as limiting the present patent.

[0060] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0061] Example 1

[0062] Figure 1 A schematic diagram showing a prognosis method for aortic arch restenosis proposed in Example 1 of the present invention, see Figure 1 , including the following steps:

[0063] S1. Acquire a certain number of three-dimensional CT cardiac images and divide them to obtain a first image data set and a second image data set;

[0064] S2. Construct a segmentation data neural network model, and use the first image data set to train the segmentation data neural network model. The specific steps of the training are:

[0065] S21. uniformly sampling the size of the three-dimensional CT cardiac image in the first image data set to the same size of 256*256*256;

[0066] S22. Set the batch size of training to 2, the learning rate to 0.001, and the number of training rounds to 400, and input the 256*256*256 sized three-dimensional CT cardiac image into the constructed segmentation data neural network model;

[0067] S23. Use the gradient descent training method to train the segmentation data neural network model. When the training round reaches 400, output the trained segmentation data neural network model.

[0068] S3. Input the three-dimensional CT cardiac image in the second image data set into the trained segmentation data neural network model, and output the segmentation result of the three-dimensional CT cardiac image in the second image data set;

[0069] S4. extracting a three-dimensional aortic arch image of the heart from the segmentation result, and extracting a center line in the three-dimensional aortic arch image;

[0070] S5. Determine the part to be measured in the three-dimensional aortic arch image, collect measurement points of the part to be measured on the center line, and then measure the part to be measured at the measurement points to obtain the feature data of the three-dimensional CT heart image described in S3, pre-process the feature data, and normalize each three-dimensional CT heart image by row so that the feature data of the three-dimensional CT heart image becomes a unit vector, which is convenient for constructing a ROC curve graph of the feature data.

[0071] S6. Obtain the physiological characteristic data of the volunteers and use it as the medical clinical characteristic data. Merge the medical clinical characteristic data with the characteristic data of the three-dimensional CT cardiac image. Use the wrapper select feature selection method to select the characteristic data for evaluating the prognosis of aortic arch restenosis from the merged characteristic data. Determine the importance of the characteristic data by sorting the characteristic data after the selection, and select the top 9 characteristic data in terms of importance.

[0072] S7. construct a regression model using the selected feature data;

[0073] S8. Obtain the experimental data to be detected, wherein the experimental data to be detected include feature data of three-dimensional CT heart images measured manually by a doctor, feature data of three-dimensional CT heart images measured by a computer, medical clinical feature data, feature data combined with feature data of three-dimensional CT heart images measured by a computer and medical clinical feature data, and feature data combined with feature data of three-dimensional CT heart images measured manually by a doctor and medical clinical feature data, input the experimental data to be detected into the constructed regression model, detect the prognosis of aortic arch restenosis, and evaluate the detection of the prognosis of aortic arch restenosis.

[0074] Example 2

[0075] See also Figure 2 , in step S4, the steps of extracting the center line are:

[0076] S41. Determine the vascular position of the aortic arch from the three-dimensional aortic arch image, where AAO represents the ascending aorta, DA represents the descending aorta, IA represents the innominate artery, LCCA represents the left common carotid artery, and LSA represents the left subclavian artery;

[0077] S42. Skeletonize the blood vessels of the aortic arch;

[0078] S43. Based on the skeletonized blood vessel, the center of the blood vessel is determined, and five lines extending along the blood vessel toward the center are determined, and all five lines are set as center lines.

[0079] The parts to be measured are the ascending aorta, the proximal aorta, the distal aorta, the isthmus, the stenosis and the descending aorta. The steps for collecting the measurement points include:

[0080] S51. Obtaining the starting point and the end point of the center line from the descending aorta DA to the ascending aorta AAO along the direction of the blood vessel with the largest cross-sectional area in the three-dimensional aortic arch image, determining a connection path between the starting point and the end point according to the direction of the blood vessel between the starting point and the end point, and setting the connection path as the main path of the center line in the three-dimensional aortic arch image, wherein the main path is composed of a plurality of voxel points in the three-dimensional aortic arch image;

[0081] S52. Obtain all voxel points on the trunk path, determine the number of center lines connected to each voxel point, that is, the degree of the voxel point, and obtain the cross-sectional area of ​​the blood vessel connected to the voxel point at each voxel point;

[0082] S53. Compare the cross-sectional areas of the blood vessels connected to each voxel point to obtain the point of the voxel with the smallest cross-sectional area on the trunk path;

[0083] S54. Set the starting point as the measurement point of the ascending aorta, set the end point as the measurement point of the descending aorta, collect all voxel points with a degree of 3 on the main path, there are 3 voxel points with a degree of 3 in total, set the 3 voxel points with a degree of 3 on the main path as the measurement points of the proximal aorta, the distal aorta and the isthmus, respectively, and set the point with the smallest cross-sectional area on the main path as the measurement point of the stenosis;

[0084] The measurement steps are as follows: obtain the cross-section of the blood vessel at the positions of the ascending aorta, proximal aorta, distal aorta, isthmus, coarctation and descending aorta, extract the contour of the cross-section of the blood vessel using the sobel operator, fit the contour of the cross-section into an ellipse using the least squares method, measure the major axis, minor axis, diameter, circumference and area of ​​the ellipse, and use the measurement results as the major axis, minor axis, diameter, circumference and area of ​​the contour of the cross-section corresponding to the ellipse; the regression model is constructed using the random forest algorithm, and the algorithm uses the sklean library function in pytorch.

[0085] Example 3

[0086] See also Figure 1In step S8, the evaluation process adopts a five-fold cross validation method, and the experimental data to be tested are randomly divided into a training set and a test set five times. The experimental data to be tested include 30 feature data of three-dimensional CT heart images measured manually by doctors, 30 feature data of three-dimensional CT heart images measured by computers, 71 feature data of medical clinics, 101 feature data of three-dimensional CT heart images measured by computers and medical clinics, and 101 feature data of three-dimensional CT heart images measured manually by doctors and medical clinics. Each experimental data to be tested is divided into a training set and a test set. The data were randomly divided into 1st, 2nd, 3rd, 4th and 5th folds. The data of 2nd, 3rd, 4th and 5th folds were used as training sets to train the regression model for the first time, and the data of 1st fold was used as the test set to evaluate the prognosis detection of aortic arch restenosis by the regression model. The data of 1st, 3rd, 4th and 5th folds were used as training sets to train the regression model for the second time, and the data of 2nd fold was used as the test set, and so on. The area under the ROC curve of the 5 test sets, that is, the AUC value, was used as the evaluation standard to evaluate the prediction effect of the model. The higher the AUC value, the larger the area under the curve, which means the higher the prediction accuracy.

[0087] Figure 3 ROC curve diagram showing the characteristic data of 3D CT cardiac images measured manually by doctors, see Figure 3 ,The top 9 feature data of importance selected are isthmus area, CoA-I, distal short diameter, constriction short diameter, diaphragm circumference, constriction diameter, constriction circumference, ascending area and diaphragm horizontal diameter, and the average value of five-fold cross validation is 0.72;

[0088] Figure 4 ROC curve diagram showing characteristic data of computer-measured 3D CT cardiac images, see Figure 4 The top 9 feature data of importance selected are ascending area, isthmus area, proximal short diameter, proximal long diameter, constricted perimeter, proximal area, ascending long diameter, isthmus perimeter and constricted diameter, and the average value of five-fold cross validation is 0.73;

[0089] Figure 5 ROC curve diagram showing clinical medical feature data, see Figure 5 The top 9 feature data selected in terms of importance are heart rate, blood oxygen saturation, right upper limb systolic pressure, aortic valve flow velocity, maximum pressure difference between upper and lower limbs, right lower limb diastolic pressure, body weight, mitral valve early diastolic velocity and pulmonary valve flow velocity. The average AUC of five-fold cross validation is 0.75;

[0090] Figure 6 The ROC curve of the feature data of the three-dimensional CT heart image measured manually by the doctor and the feature data of the medical clinic is shown in Figure 6The top 9 feature data of importance were selected as heart rate, short diameter of the stenosis, aortic valve flow velocity, body weight, distal short diameter, diameter of the stenosis, circumference of the stenosis, early diastolic velocity of the mitral valve, and diastolic pressure of the left lower limb. The average value of the five-fold cross validation was 0.80;

[0091] Figure 7 ROC curve diagram showing the characteristic data of the computer-measured three-dimensional CT cardiac image combined with the characteristic data of medical clinic, see Figure 7 The top 9 feature data selected in terms of importance are heart rate, ascending area, aortic valve flow velocity, body weight, isthmus area, proximal circumference, right upper limb systolic pressure, mitral valve early diastolic velocity and right lower limb diastolic pressure. The average value of the five-fold cross validation is 0.79.

[0092] From the above experimental results, we can see that Figure 3 and Figure 4 , the regression model evaluates the prognosis of aortic arch restenosis detection using only the feature data of three-dimensional CT heart images measured manually by doctors and the evaluation results using only the feature data of three-dimensional CT heart images measured by computers. The difference between the two is not large, indicating that the feature data of three-dimensional CT heart images measured by computers can achieve the evaluation results of the feature data of CT images measured manually by doctors; see Figure 5 , Figure 6 and Figure 7 The regression model's evaluation of the prognosis of aortic arch restenosis using only medical clinical feature data is worse than the evaluation result using only the combined data of medical clinical feature data and image feature data, indicating that the regression model's evaluation result of the prognosis of aortic arch restenosis using the combined data of medical clinical feature data and image feature data is better.

[0093] Example 4

[0094] See also Figure 8 , a prognostic system for aortic arch restenosis, comprising the following:

[0095] An image data set acquisition module 11 is used to acquire a certain number of three-dimensional CT cardiac images and divide them into a first image data set and a second image data set;

[0096] The segmentation data neural network model construction processing module 12 is used to construct the segmentation data neural network model, and train the segmentation data neural network model using the first image data set to obtain a trained segmentation data neural network model;

[0097] A segmentation module 13, used for inputting the three-dimensional CT heart image in the second image data set into the trained segmentation data neural network model, and outputting the segmentation result of the three-dimensional CT heart image in the second image data set;

[0098] An extraction module 14, used for extracting an aortic arch image of the heart from the segmentation result, and extracting a center line in the aortic arch image;

[0099] The measurement processing module 15 is used to determine the part to be measured in the aortic arch image, collect measurement points of the part to be measured on the center line, and then measure the part to be measured at the measurement points to obtain feature data of the three-dimensional CT heart image in the second image data set, and pre-process the feature data;

[0100] The feature data selection module 16 is used to obtain the physiological feature data of the volunteer, use it as the medical clinical feature data, merge the medical clinical feature data with the feature data of the three-dimensional CT heart image, and select the feature data for evaluating the prognosis of aortic arch restenosis from the merged feature data;

[0101] A regression model building module 17, used to build a regression model using the selected feature data;

[0102] The evaluation module 18 is used to obtain the experimental data to be detected, input the experimental data to be detected into the constructed regression model, detect the prognosis of aortic arch restenosis, and evaluate the prognosis detection of aortic arch restenosis.

[0103] Obviously, the above embodiments of the present invention are only examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A prognosis method for aortic arch restenosis, characterized in that: The following steps are involved: S1. Acquire a certain number of three-dimensional CT cardiac images and divide them to obtain a first image data set and a second image data set; S2. Constructing a segmentation data neural network model, and training the segmentation data neural network model using the first image data set to obtain a trained segmentation data neural network model; S3. Input the three-dimensional CT cardiac image in the second image data set into the trained segmentation data neural network model, and output the segmentation result of the three-dimensional CT cardiac image in the second image data set; S4. extracting a three-dimensional aortic arch image of the heart from the segmentation result, and extracting a center line in the three-dimensional aortic arch image; S5. Determine the part to be measured in the three-dimensional aortic arch image, collect measurement points of the part to be measured on the center line, and then measure the part to be measured at the measurement points to obtain the characteristic data of the three-dimensional CT heart image described in S3, and pre-process the characteristic data; S6. Acquire physiological characteristic data of volunteers, use the data as medical clinical characteristic data, merge the medical clinical characteristic data with the characteristic data of three-dimensional CT cardiac images, and select characteristic data for evaluating the prognosis of aortic arch restenosis from the merged characteristic data; S7. construct a regression model using the selected feature data; S8. Obtain the experimental data to be tested, input the experimental data to be tested into the constructed regression model, test the prognosis of aortic arch restenosis, and evaluate the prognosis detection of aortic arch restenosis.

2. The prognosis method for aortic arch restenosis according to claim 1, characterized in that: In step S2, the specific steps of segmentation data neural network model training are: S21. uniformly sampling the size of the three-dimensional CT cardiac image in the first image data set to the same size; S22. Set the number of training batches batchsize, learning rate and training round n, and input the three-dimensional CT heart images of the same size into the constructed segmentation data neural network model; S23. Use the gradient descent training method to train the segmentation data neural network model, and when the training round reaches n, output the trained segmentation data neural network model.

3. The prognosis method for aortic arch restenosis according to claim 1, characterized in that: In step S4, the steps of extracting the center line are: S41. Determine the blood vessel position of the aortic arch by a three-dimensional aortic arch image; S42. Skeletonize the blood vessels of the aortic arch; S43. Based on the skeletonized blood vessel, the center of the blood vessel is determined, and several lines extending along the blood vessel toward the center are determined, and the several lines are set as center lines.

4. The prognosis method for aortic arch restenosis according to claim 3, characterized in that: In step S5, the parts to be measured are the ascending aorta, the proximal aorta, the distal aorta, the isthmus, the constriction and the descending aorta.

5. The prognosis method for aortic arch restenosis according to claim 4, characterized in that: In step S5, the step of collecting the measurement points includes: S51. Obtaining a starting point and an end point of a center line along the direction of a blood vessel with the largest cross-sectional area in the three-dimensional aortic arch image, determining a connection path between the starting point and the end point according to the direction of the blood vessel between the starting point and the end point, and setting the connection path as a main path of the center line in the three-dimensional aortic arch image, wherein the main path is composed of a plurality of voxel points in the three-dimensional aortic arch image; S52. Obtain all voxel points on the trunk path, determine the number of center lines connected to each voxel point, that is, the degree of the voxel point, and obtain the cross-sectional area of ​​the blood vessel connected to the voxel point at each voxel point; S53. Compare the cross-sectional areas of the blood vessels connected to each voxel point to obtain the point of the voxel with the smallest cross-sectional area on the trunk path; S54. Set the starting point as the measurement point of the ascending aorta, set the end point as the measurement point of the descending aorta, collect all voxel points with a degree of n on the main path, set all voxel points with a degree of n on the main path as the measurement points of the proximal aorta, the distal aorta and the isthmus, and set the point with the smallest cross-sectional area on the main path as the measurement point of the stenosis.

6. The prognosis method for aortic arch restenosis according to claim 5, characterized in that: In step S5, the measuring step is to obtain a cross-section of the blood vessel at the positions of the ascending aorta, the proximal aorta, the distal aorta, the isthmus, the stenosis and the descending aorta, extract the contour of the cross-section of the blood vessel using the Sobel operator, fit the contour of the cross-section into an ellipse using the least squares method, measure the major axis, minor axis, diameter, circumference and area of ​​the ellipse, and use the measurement results as the major axis, minor axis, diameter, circumference and area of ​​the contour of the cross-section corresponding to the ellipse.

7. The prognosis method for aortic arch restenosis according to claim 6, characterized in that: In step S7, the regression model is constructed using a random forest algorithm.

8. The prognosis method for aortic arch restenosis according to claim 1, characterized in that: In step S8, the experimental data to be tested include feature data of three-dimensional CT heart images measured manually by a doctor, feature data of three-dimensional CT heart images measured by a computer, feature data of medical clinics, feature data of three-dimensional CT heart images measured by a computer and medical clinics combined, and feature data of three-dimensional CT heart images measured manually by a doctor and medical clinics combined.

9. The prognosis method for aortic arch restenosis according to claim 1, characterized in that: In step S8, the evaluation process adopts a five-fold cross-validation method, randomly dividing the experimental data to be tested into a training set and a test set five times, using the training set to train the regression model, and using the test set to evaluate the prognosis detection of aortic arch restenosis by the regression model.

10. A prognostic system for aortic arch restenosis, characterized in that: Includes the following systems: An image data set acquisition module, used for acquiring a certain number of three-dimensional CT cardiac images and dividing them to obtain a first image data set and a second image data set; A segmentation data neural network model construction processing module is used to construct a segmentation data neural network model, and train the segmentation data neural network model using the first image data set to obtain a trained segmentation data neural network model; a segmentation module, used for inputting the three-dimensional CT heart image in the second image data set into the trained segmentation data neural network model, and outputting the segmentation result of the three-dimensional CT heart image in the second image data set; An extraction module, used for extracting an aortic arch image of the heart from the segmentation result, and extracting a center line in the aortic arch image; a measurement processing module, used for determining a portion to be measured in the aortic arch image, collecting measurement points of the portion to be measured on the center line, and then measuring the portion to be measured at the measurement points, obtaining feature data of the three-dimensional CT heart image in the second image data set, and preprocessing the feature data; A feature data selection module is used to obtain physiological feature data of volunteers, use the data as medical clinical feature data, merge the medical clinical feature data with feature data of three-dimensional CT cardiac images, and select feature data for evaluating the prognosis of aortic arch restenosis from the merged feature data; A regression model building module, used to build a regression model using selected feature data; The evaluation module is used to obtain the experimental data to be detected, input the experimental data to be detected into the constructed regression model, detect the prognosis of aortic arch restenosis, and evaluate the prognosis detection of aortic arch restenosis.

Images