Aortic expansion risk prediction system after endovascular repair of type b aortic dissection

Abstract

The application discloses a B-type aortic dissection endoluminal repair postoperative aortic expansion risk prediction system, which is characterized in that: through a deep learning network, the whole aorta, true cavity and false cavity of preoperative and postoperative CTA images are segmented, spatial angle, morphology, density and texture and vascular branch imaging features are extracted in a preset interest region, optimal feature sets are formed by fusing clinical full-dimension data, hemodynamic parameters and postoperative follow-up data through feature screening, a prognosis prediction model is constructed based on a survival prognosis analysis algorithm to output risk prediction results, and dynamic change features are extracted according to postoperative imaging and clinical data in different follow-up stages to update the prediction results and trigger early warning, so that the problems of difficult prognosis analysis of the B-type aortic dissection, fuzzy boundaries of 301 typing subtypes, great heterogeneity, single imaging feature dimension, insufficient data fusion and lack of postoperative dynamic monitoring are solved, and the precision level of aortic expansion risk stratified management after TEVAR is improved.

Description

Technical Field

[0001] This invention relates to the field of medical imaging artificial intelligence and vascular surgery clinical prognostic assessment technology, and in particular to a system for predicting the risk of aortic dilation after endovascular repair of type B aortic dissection. Background Technology

[0002] With the widespread use of thoracic endovascular aortic repair (TEVAR) in the treatment of type B aortic dissection (TBAD), risk assessment of poor aortic remodeling and the long-term adverse event of thoracic aortic dilarion (TAD) has become an important clinical concern. Currently, there is no reliable model for predicting TAD risk after TEVAR in TBAD patients in clinical practice; most methods rely on empirical analysis combined with limited clinical indicators. Based on research experience, a 301-type classification for predicting TAD risk after TEVAR in TBAD patients has been developed. This classification categorizes TBAD into subtypes B1, B2, and B3 based on the relative positions of the true and false lumens in CTA images. Experiments have demonstrated its effectiveness in assessing TAD risk after TEVAR. Simultaneously, existing imaging techniques have enabled automatic segmentation of the true and false lumens in CTA images of TBAD. The segmentation results show a Dice similarity coefficient of 0.93±0.01 compared to manually segmented images by experts, laying the foundation for image feature extraction. Based on this, existing studies have constructed predictive models by extracting imaging parameters such as cumulative angle change value, cumulative absolute angle value, and corrected cumulative absolute angle value. Some models combine algorithms such as Cox, Random Survival Forest (RSF), and Survival Support Vector Machine (SVM) to achieve prognostic analysis. At the clinical level, simplified clinical predictive models are constructed using a few indicators such as the status of false lumen thrombosis in the thoracic aorta and the number of visceral arteries supplied by the false lumen of the abdominal aorta.

[0003] However, existing technologies still have significant limitations in clinical applications.

[0004] First, there are limitations to the classification and feature quantification: the traditional 301 classification is a qualitative classification system with blurred boundaries between subtypes, especially the B2 subtype which has the largest proportion and is highly heterogeneous, making it impossible to achieve refined risk stratification.

[0005] Secondly, the model data integration is poor: some existing predictive studies focus on a certain imaging feature, while others focus on clinical indicators. They have not achieved deep integration of imaging features, hemodynamic parameters, clinical data across all dimensions, and follow-up data. They have ignored the impact of key hemodynamic factors such as the location / number of distal ruptures, false lumen blood flow velocity, and distribution of spinal cord blood supply arteries on prognosis, thus limiting the accuracy and generalization of model predictions.

[0006] Furthermore, the algorithms and models have low levels of intelligence: existing prediction methods mainly rely on traditional manual image interpretation, lacking the ability to adaptively extract features from data and model nonlinear relationships, and have not introduced a model interpretability module. Clinicians cannot understand the core basis of the model's predictions, making it difficult to implement in clinical decision-making.

[0007] Furthermore, the dynamic nature of risk prediction is lacking: existing models are static predictions and do not take into account the dynamic changes in imaging and clinical indicators during the postoperative follow-up stage, thus failing to achieve dynamic postoperative risk monitoring and real-time early warning.

[0008] Finally, the population adaptability is limited: most existing research conclusions are based on regional patient databases and do not consider the influence of factors such as race, region, and lifestyle on the pathological characteristics of TBAD. They also lack cross-population adaptive calibration modules, resulting in poor applicability of the model in different populations.

[0009] Therefore, there is an urgent need to develop a novel TEVAR postoperative TAD risk prognosis prediction system that is based on preoperative CTA image analysis, has intelligent prediction capabilities, and is highly interpretable, in order to overcome the above-mentioned shortcomings of existing technologies and improve the accuracy of risk stratification management for TBAD patients after TEVAR. Summary of the Invention

[0010] To address the aforementioned technical problems, this invention provides a system for predicting the risk of aortic dilation after endovascular repair of type B aortic dissection.

[0011] In a first aspect, the present invention provides a system for predicting the risk of aortic dilation after endovascular repair of type B aortic dissection, the technical solution of which is as follows: The CTA image preprocessing module is used to acquire CTA image data of patients with type B aortic dissection during the preoperative and postoperative follow-up stages, and to perform standardized preprocessing on the CTA image data to obtain preprocessed CTA image data. The fully automated aortic segmentation module is used to perform fully automated segmentation of the aorta, true lumen, and false lumen in the preprocessed CTA image data based on a deep learning network, and output a segmentation mask. The imaging feature extraction module is used to extract spatial angle features, morphological features, density and texture features, and vascular branch features within a preset region of interest based on the segmentation mask, and to normalize the extracted features to form a multi-dimensional imaging feature set. The multi-source data fusion module is used to fuse the multi-dimensional imaging feature set, clinical full-dimensional data, hemodynamic parameters and postoperative follow-up data to construct a fused dataset, and to perform feature filtering on the fused dataset to form the optimal feature set; The intelligent prognostic prediction model module is used to construct a prognostic prediction model based on the optimal feature set and through a survival prognostic analysis algorithm, and output the risk prediction results of TAD after TEVAR in patients with type B aortic dissection. The postoperative dynamic risk monitoring module is used to extract dynamic change features from the imaging and clinical data of the patients with type B aortic dissection at different follow-up stages after surgery and input them into the prognostic prediction model to update the TEVAR postoperative TAD risk prediction results and trigger an early warning when the updated risk prediction results exceed a threshold.

[0012] The beneficial effects of the aortic dilation risk prediction system after endovascular repair of type B aortic dissection according to the present invention are as follows: The system of this invention performs fully automatic segmentation of the aorta, its true lumen, and its false lumen using a deep learning network. It extracts multi-dimensional imaging features of spatial angle, morphology, density, texture, and vascular branches within a preset region of interest. It integrates clinical data, hemodynamic parameters, and postoperative follow-up data to form an optimal feature set through feature selection. Based on a survival prognostic analysis algorithm, it constructs a prognostic prediction model and outputs risk prediction results. It also extracts dynamic change features from imaging and clinical data at different postoperative follow-up stages to update the prediction results and trigger early warnings. This system solves the problems of the traditional 301 subtype classification, such as ambiguous subtype boundaries, high heterogeneity, single-dimensional imaging features, insufficient fusion of multi-source data, and lack of postoperative dynamic monitoring. It can improve the accuracy of risk stratification management of adverse events after TEVAR.

[0013] Based on the above scheme, the aortic dilation risk prediction system after endovascular repair of type B aortic dissection of the present invention can be further improved as follows.

[0014] In one alternative approach, the CTA image preprocessing module is specifically used for: The CTA imaging data of the patients with type B aortic dissection were obtained through the medical imaging system interface during the preoperative and postoperative follow-up phases. The CTA image data is subjected to voxel value normalization, Gaussian filtering for noise removal, and grayscale histogram equalization correction. The corrected CTA image data is automatically located for regions of interest, and then the CTA image data after locating the regions of interest is resampled, voxel spacing is unified, and label is encoded to obtain the preprocessed CTA image data.

[0015] The beneficial effects of adopting the above-mentioned optional methods are as follows: CTA image data in the preoperative and postoperative follow-up stages are further obtained through the medical imaging system interface, and after voxel value normalization, Gaussian filtering noise removal and gray-level histogram equalization correction, automatic location of region of interest, resampling and voxel spacing are implemented to improve the standardization of image data, provide a high-quality data foundation for subsequent aortic segmentation and feature extraction, and reduce the impact of noise interference and data heterogeneity.

[0016] In one alternative approach, the fully automated aortic segmentation module is specifically used for: The preprocessed CTA image data is input into a deep learning network built on the nnU-Netv2 framework. Through the adaptive network architecture search mechanism and automatic hyperparameter tuning function of the deep learning network, feature maps are extracted from the whole aorta, true lumen and false lumen in the preprocessed CTA image data. The feature map is processed through the residual connections and batch normalization mechanism of the deep learning network, and the processed feature map is restored to the same spatial resolution as the CTA image data through the adaptive upsampling layer of the deep learning network, and the segmentation mask is output.

[0017] The advantages of adopting the above-mentioned optional method are as follows: the preprocessed CTA image data is further input into a deep learning network built on the nnU-Netv2 framework. With the help of adaptive network architecture search and automatic hyperparameter tuning functions, combined with residual connection and batch normalization processing, the spatial resolution is restored through adaptive upsampling layer, realizing fully automatic and accurate segmentation of the aorta, the true lumen and the false lumen, improving segmentation accuracy, reducing manual intervention, and providing accurate segmentation masks for subsequent feature extraction.

[0018] In one alternative approach, the imaging feature extraction module is specifically used for: Within the preset region of interest, based on the region corresponding to the segmentation mask in the CTA image data, the spatial angle features, morphological features, density and texture features, and vascular branch features are calculated; The spatial angle features, morphological features, density and texture features, and vascular branching features are normalized to form the multidimensional imaging feature set.

[0019] The beneficial effects of adopting the above-mentioned optional methods are as follows: further calculate the spatial angle, morphology, density and texture and vascular branch features based on the segmentation mask in the preset region of interest, and perform normalization processing to form a multi-dimensional imaging feature set, which can characterize the anatomical structure and pathological changes of the aorta, supplement the limitations of the traditional 301 classification which only focuses on angular features, provide rich imaging biomarkers for prognostic prediction models, and improve the comprehensiveness and accuracy of risk assessment.

[0020] In one alternative approach, the multi-source data fusion module is specifically used for: The multi-dimensional imaging feature set, the full-dimensional clinical data, the hemodynamic parameters, and the postoperative follow-up data are spliced ​​and fused to construct a fused dataset. A feature selection algorithm combining LASSO regression was used to reduce the dimensionality of the fused dataset, remove redundant features and retain the core features related to adverse events after TEVAR, thus forming the optimal feature set.

[0021] The beneficial effects of adopting the above-mentioned optional methods are as follows: further splicing and fusing multi-dimensional imaging feature sets, clinical full-dimensional data, hemodynamic parameters and postoperative follow-up data, using a feature selection algorithm combined with LASSO regression to perform feature dimensionality reduction, eliminating redundant features and retaining core features related to adverse events after TEVAR, forming the optimal feature set, reducing data dimensionality and noise interference, and improving training efficiency and prediction accuracy.

[0022] In one alternative approach, the intelligent prognostic prediction model module is specifically used for: The optimal feature set is randomly divided into a training set and a validation set according to a preset ratio; At least one of the random survival forest algorithm, survival support vector machine algorithm, or Cox proportional hazards model is used as the survival prognostic analysis algorithm to construct an initial prognostic prediction model on the training set. The hyperparameters of the initial prognostic prediction model were optimized using cross-validation and grid search to obtain the optimized prognostic prediction model. The validation set is input into the optimized prognostic prediction model, and the TAD risk prediction results after TEVAR are output for patients with type B aortic dissection.

[0023] The beneficial effects of adopting the above optional methods are as follows: the optimal feature set is further randomly divided into training set and validation set according to a preset ratio, and random survival forest, survival support vector machine or Cox proportional hazards model are used as survival prognostic analysis algorithms. An initial prognostic prediction model is constructed on the training set, and the hyperparameters are optimized by cross-validation and grid search method to improve the model's generalization ability and accurately output the TAD risk prediction results after TEVAR.

[0024] In one alternative approach, the postoperative dynamic risk monitoring module is specifically used for: Acquire imaging and clinical data of the patients with type B aortic dissection at different follow-up stages after surgery; The imaging data and clinical data are compared and analyzed with the corresponding preoperative baseline data to extract dynamic change features; The dynamic change characteristics are input into the optimized prognostic prediction model to obtain the updated risk prediction result; An early warning is triggered when the updated risk prediction result exceeds a preset threshold.

[0025] The beneficial effects of adopting the above-mentioned optional methods are as follows: further acquisition of imaging data and clinical data at different follow-up stages after surgery, comparative analysis with preoperative baseline data to extract dynamic change features, input of dynamic change features into the optimized prognostic prediction model to update risk prediction results, triggering early warning when the results exceed a preset threshold, realizing dynamic monitoring and real-time early warning of postoperative risks, timely capturing changes in the condition, assisting clinical intervention, and reducing the risk of adverse events.

[0026] Secondly, the present invention provides a method for predicting the risk of aortic dilation after endovascular repair of type B aortic dissection, the technical solution of which is as follows: CTA image data of patients with type B aortic dissection were acquired during the preoperative and postoperative follow-up stages, and the CTA image data were standardized and preprocessed to obtain preprocessed CTA image data. Based on a deep learning network, the aorta, true lumen, and false lumen in the preprocessed CTA image data are automatically segmented, and a segmentation mask is output. Within a preset region of interest, spatial angle features, morphological features, density and texture features, and vascular branch features are extracted based on the segmentation mask, and the extracted features are normalized to form a multi-dimensional imaging feature set. The multi-dimensional imaging feature set, clinical full-dimensional data, hemodynamic parameters and postoperative follow-up data are fused to construct a fused dataset, and the fused dataset is then subjected to feature filtering to form the optimal feature set; Based on the optimal feature set, a prognostic prediction model is constructed using a survival prognostic analysis algorithm, and the risk prediction results of TAD after TEVAR in patients with type B aortic dissection are output. Dynamic change features are extracted from imaging and clinical data of patients with type B aortic dissection at different follow-up stages after surgery and input into the prognostic prediction model to update the TEVAR postoperative TAD risk prediction results. An early warning is triggered when the updated risk prediction results exceed the threshold.

[0027] The beneficial effects of the method for predicting the risk of aortic dilation after endovascular repair of type B aortic dissection according to the present invention are as follows: The method of this invention uses a deep learning network to automatically segment the aorta, its true lumen, and its false lumen. Within a preset region of interest, it extracts multi-dimensional imaging features, including spatial angles, morphology, density, texture, and vascular branches. It integrates clinical data, hemodynamic parameters, and postoperative follow-up data, and forms an optimal feature set through feature selection. Based on a survival prognostic analysis algorithm, it constructs a prognostic prediction model to output risk prediction results. The prediction results are updated and warnings are triggered based on dynamic changes in imaging and clinical data at different postoperative follow-up stages. This method solves the problems of ambiguous and heterogeneous subtype boundaries, single-dimensional imaging features, insufficient fusion of multi-source data, and lack of postoperative dynamic monitoring in the traditional 301 classification. It can improve the accuracy of risk stratification management of adverse events after TEVAR.

[0028] Thirdly, the technical solution of an electronic device according to the present invention is as follows: The invention includes a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the method for predicting the risk of aortic dilation after endovascular repair of type B aortic dissection as described in this invention.

[0029] Fourthly, the technical solution of a computer-readable storage medium provided by the present invention is as follows: The computer-readable storage medium stores instructions that, when read, cause the computer-readable storage medium to perform the steps of the method for predicting the risk of aortic dilatation after endovascular repair of type B aortic dissection as described in the present invention.

[0030] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0031] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a schematic diagram of an embodiment of the aortic dilation risk prediction system after endovascular repair of type B aortic dissection according to the present invention. Figure 2 This is a flowchart illustrating an embodiment of a method for predicting the risk of aortic dilation after endovascular repair of type B aortic dissection according to the present invention. Figure 3This is a schematic diagram of an embodiment of an electronic device according to the present invention. Detailed Implementation

[0032] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0033] Figure 1 This diagram illustrates a structural schematic of an embodiment of a risk prediction system for aortic dilation after endovascular repair of type B aortic dissection provided by the present invention. Figure 1 As shown, the aortic dilation risk prediction system 100 after endovascular repair of type B aortic dissection includes: The CTA image preprocessing module 101 is used to acquire CTA image data of patients with type B aortic dissection during the preoperative and postoperative follow-up stages, and to perform standardized preprocessing on the CTA image data to obtain preprocessed CTA image data.

[0034] Type B aortic dissection refers to individuals diagnosed with Stanford type B aortic dissection via imaging examination. According to the 301 classification, it can be further divided into subtypes B1, B2, and B3, where the aortic dissection involves the descending aorta but not the ascending aorta. For example, Patient A, a 55-year-old male, was admitted to the hospital due to sudden chest and back pain and was diagnosed with type B aortic dissection via CTA, with a 301 classification of type B2. The preoperative and postoperative follow-up phases refer to the time period before endovascular aortic repair surgery for type B aortic dissection patients and the regular follow-up examinations after surgery. For example, if Patient A undergoes TEVAR surgery on March 1, 2025, the preoperative phase is before March 1, 2025, and the postoperative follow-up phase includes follow-up examinations at 1 month, 3 months, 6 months, and 12 months postoperatively.

[0035] CTA image data refers to raw medical image data acquired through computed tomography angiography, stored in DICOM format. For example, patient A underwent CTA examinations before surgery and at 1, 3, 6, and 12 months post-surgery, generating approximately 500 layers of raw DICOM image data each time. Preprocessed CTA image data refers to data obtained after standardizing the raw CTA image data, including voxel value normalization, noise removal, grayscale correction, resampling, and voxel spacing uniformity. For example, after preprocessing, patient A's preoperative CTA raw image data had voxel values ​​normalized to between 0 and 1, image noise removed by Gaussian filtering, and the voxel spacing of all images uniformly set to 0.5 mm × 0.5 mm × 0.5 mm.

[0036] Specifically, during the model building phase, the CTA image preprocessing module 101 acquires CTA image data of patients with type B aortic dissection at various follow-up stages before and after surgery for subsequent segmentation and feature extraction; during the model usage phase, it acquires CTA image data of patients at different follow-up stages after surgery for dynamic risk monitoring.

[0037] The fully automatic aortic segmentation module 102 is used to perform fully automatic segmentation of the aorta, the true lumen, and the false lumen in the preprocessed CTA image data based on a deep learning network, and output a segmentation mask.

[0038] Deep learning networks refer to computational models built on deep neural networks that can automatically learn and extract features from medical images and perform target segmentation. For example, preprocessed CTA image data of patient A is input into a deep learning network built on the nnU-Netv2 framework. This network contains encoder and decoder structures and can accurately segment the aortic region.

[0039] The term "complete aorta" refers to the intact thoracic and abdominal aortic structures displayed on CTA images, including the aortic wall, lumen, and branches. For example, in patient A's preoperative CTA images, the complete aorta includes the entire aortic structure from the origin of the aortic arch to the bifurcation of the iliac arteries. The "true lumen" refers to the original vascular cavity in type B aortic dissection with faster blood flow and higher pressure, usually separated from the false lumen by an intimal flap. For example, in patient A's CTA images, the true lumen appears as a well-filled, small-diameter cavity located in the medial portion of the aorta. The "false lumen" refers to the abnormal vascular cavity in type B aortic dissection with slower blood flow and lower pressure, formed by a tear in the aortic wall. For example, in patient A's CTA images, the false lumen appears as a delayed-filled, large-diameter cavity located in the lateral portion of the aorta, with thrombus formation visible in some areas.

[0040] Here, the segmentation mask refers to a three-dimensional binary image with the same spatial resolution as the original CTA image, in which each voxel is labeled as belonging to or not belonging to a specific anatomical structure; for example, in the segmentation mask output by the deep learning network of patient A, voxels of the whole aortic region are labeled as 1, voxels of the true lumen region are labeled as 2, voxels of the false lumen region are labeled as 3, and voxels of the background region are labeled as 0.

[0041] Specifically, the fully automatic aortic segmentation module 102 outputs segmentation masks for each layer, which are used for subsequent feature extraction and volume calculation.

[0042] The imaging feature extraction module 103 is used to extract spatial angle features, morphological features, density and texture features, and vascular branch features based on the segmentation mask within a preset region of interest, and to normalize the extracted features to form a multi-dimensional imaging feature set.

[0043] The preset region of interest refers to the anatomical region in the CTA image that needs to be extracted for features. It is usually defined based on the vertebral body position. For example, in the CTA image of patient A, the preset region of interest is set to the aortic segment within the T4 to T11 vertebral body level. This area is a key area for aortic remodeling after endovascular repair of the thoracic aorta.

[0044] Spatial angular features refer to quantitative parameters describing the relative position and angular relationship between the true lumen and false lumen of the aorta in three-dimensional space. These parameters refine the description of the spatial relationship between the true and false lumens in the 301 classification. For example, within patient A's predefined region of interest, the extracted spatial angular features include a three-dimensional angle of 15° between the centerline of the true lumen and the centerline of the false lumen, an angle variation rate of 0.8° / cm, and a helical curvature of 0.02. Morphological features refer to quantitative parameters describing the shape, size, and structural morphology of the true and false lumens of the aorta. These parameters supplement the quantitative assessment of dissection morphology in the 301 classification. For example, within patient A's predefined region of interest, the extracted morphological features include a true lumen cross-sectional area of ​​120 mm². 2 The cross-sectional area of ​​the false cavity is 350 mm. 2 The true lumen compression rate was 65%, and the false lumen expansion rate was 42%. Density and texture features refer to quantitative parameters describing the thrombus density distribution and calcification of the false lumen wall, used to assess thrombus burden and calcification degree in the 301 classification. For example, within patient A's predefined region of interest, the extracted density and texture features include a mean CT value of 45 HU for the false lumen thrombus, a CT value variance of 12 HU, a gray-level co-occurrence matrix contrast of 0.35, and a false lumen wall calcification density of 0.8 calcifications / cm. Vascular branch features refer to quantitative parameters describing the relationship between vascular branches such as intercostal arteries and spinal cord feeding arteries and the false lumen, used to assess branch vessel involvement in the 301 classification. For example, within patient A's predefined region of interest, the extracted vascular branch features include 8 intercostal artery openings, a false lumen feeding branch artery diameter of 2.5 mm, and a distance of 3.2 mm between the spinal cord feeding artery and the false lumen.

[0045] Among them, the multidimensional imaging feature set refers to the feature set after combining and normalizing spatial angle features, morphological features, density and texture features and vascular branch features, which is used to transform the qualitative classification of the 301 classification into quantitative indicators. For example, after normalization, all 128 dimensions of imaging features of patient A form a multidimensional imaging feature set containing 128 values.

[0046] Specifically, the imaging feature extraction module 103 obtains imaging feature sets such as the geometric center point of the aortic true and false lumens, the center point of the vertebral body, and the vectors of the true and false lumens within a preset region of interest based on a segmentation mask.

[0047] It should be noted that, in this embodiment, the system also includes: a true and false cavity vector angle accumulation module and a follow-up data analysis module; The true and false cavity vector angle accumulation module is used to automatically calculate the accumulated value of the true and false cavity vector direction angles based on the relative position of the true and false cavity planes and their position relative to the center point of the vertebral body, thereby obtaining a mathematical representation feature set of the spatial relative position of the true and false cavities.

[0048] Among them, the mathematical representation feature set of the spatial relative position of the true and false cavities refers to: a high-dimensional feature set reflecting the spatial configuration of the true and false cavities, based on the geometric center point of the true and false cavities, the center point of the vertebral body, and the vectors of the true and false cavities, by calculating the cumulative value of the vector direction angle; for example, the mathematical representation feature set output by the vector angle accumulation module of the true and false cavities of patient A includes 32-dimensional features such as the cumulative value of the direction angle and the rate of change of the vectors of the true and false cavities in three-dimensional space.

[0049] The follow-up data analysis module is used to: during the model construction period, based on bony landmarks in CTA transverse images, compare the changes in the overall aortic area at the same level between preoperative CTA images and each postoperative follow-up CTA images, and obtain the changes in aortic volume in the region of interest and the occurrence and development of TAD by integrating the data. Aortic volume increase of ≥20% is defined as TAD.

[0050] The occurrence and development of TAD refers to whether the patient experiences an increase in aortic volume of ≥20% during the follow-up period after TEVAR surgery, which serves as the outcome label for the prognostic prediction model; for example, if patient A's CTA 6 months after surgery shows an increase in aortic volume of 25% compared to before surgery, it is determined that TAD has occurred.

[0051] The multi-source data fusion module 104 is used to fuse the multi-dimensional imaging feature set, clinical full-dimensional data, hemodynamic parameters and postoperative follow-up data to construct a fused dataset, and to perform feature screening on the fused dataset to form an optimal feature set.

[0052] Clinical comprehensive data refers to a full range of clinical indicators obtained from the electronic medical record system, including patient demographic information, underlying medical history, laboratory tests, and surgical parameters. For example, Patient A's clinical comprehensive data includes age 55, male, 10-year history of hypertension, 30-year smoking history, preoperative systolic blood pressure of 145 mmHg, operation duration of 120 min, and stent length of 200 mm. Hemodynamic parameters refer to quantitative indicators describing the hemodynamic state and characteristics within the aorta, including the location and number of distal ruptures and the blood flow velocity in the false lumen. For example, Patient A's hemodynamic parameters include a preoperative false lumen blood flow velocity of 0.3 m / s, a true lumen blood flow velocity of 0.8 m / s, three distal ruptures, and a rupture location 60 mm from the left subclavian artery. Postoperative follow-up data refers to clinical outcome event data recorded at different time points after a patient's surgery. For example, postoperative follow-up data for patient A includes whether adverse events such as thoracic aortic dilatation, aortic rupture, reoperation, or death occurred at 1 month, 3 months, 6 months, and 12 months after surgery.

[0053] The fusion dataset refers to a collection of data formed by combining multi-dimensional imaging feature sets, comprehensive clinical data, hemodynamic parameters, and postoperative follow-up data. For example, Patient A's multi-dimensional imaging feature set contains 128 features, comprehensive clinical data contains 15 features, hemodynamic parameters contain 8 features, and postoperative follow-up data contains 4 outcome variables, resulting in a dataset with 155 features. The optimal feature set refers to the set of core features highly correlated with adverse events after TEVAR surgery, retained after feature selection from the fusion dataset. This set is used to construct an accurate prognostic prediction model. For example, after LASSO regression feature selection, Patient A's fusion dataset retained 12 core features from the 155 features, including false lumen expansion rate, true lumen compression rate, false lumen thrombus density variance, number of distal ruptures, and false lumen blood flow velocity.

[0054] Specifically, the multi-source data fusion module 104 fuses the multi-dimensional imaging feature set, the mathematical representation feature set of the relative spatial positions of the true and false cavities, the clinical full-dimensional data, the hemodynamic parameters, and the TAD occurrence status output by the follow-up data analysis module 108 to construct a fused dataset.

[0055] The intelligent prognostic prediction model module 105 is used to construct a prognostic prediction model based on the optimal feature set and through a survival prognostic analysis algorithm, and output the risk prediction results of TAD after TEVAR in patients with type B aortic dissection.

[0056] Among them, survival prognostic analysis algorithms refer to statistical learning methods used to analyze time-to-event data and predict the probability of event occurrence; for example, the intelligent prognostic prediction model module for patient A uses the random survival forest algorithm as the survival prognostic analysis algorithm to model 12 core features and predict the probability of postoperative thoracic aortic dilatation. Prognostic prediction models refer to computational models built based on survival prognostic analysis algorithms that can output the probability of postoperative adverse event risks for patients; for example, the prognostic prediction model built by the random survival forest algorithm for patient A on the training set can calculate the probability of thoracic aortic dilatation one year after surgery based on the patient's optimal feature set.

[0057] In this context, TAV postoperatively refers to adverse events of thoracic aortic dilation that occur after endovascular repair of the thoracic aorta in patients with type B aortic dissection. For example, TAV postoperatively in patient A includes progressive dilation of the thoracic aortic lumen, an increase in diameter exceeding 5 mm, or aortic dilation events requiring secondary intervention that occur during postoperative follow-up. Risk prediction results refer to the probability value or risk stratification level of a patient developing TAV postoperatively after TAV postoperatively, output by the prognostic prediction model. For example, the risk prediction result output by the prognostic prediction model for patient A is a probability of thoracic aortic dilation of 0.65 one year postoperatively, which is classified as high-risk.

[0058] The postoperative dynamic risk monitoring module 106 is used to extract dynamic change features from the imaging data and clinical data of the patients with type B aortic dissection at different follow-up stages after surgery and input them into the prognostic prediction model to update the TEVAR postoperative TAD risk prediction results and trigger an early warning when the updated risk prediction results exceed a threshold.

[0059] Postoperative follow-up stages refer to the specific time points at which patients undergo regular check-ups after TEVAR surgery; for example, patient A's postoperative follow-up stages include four specific time points: 1 month, 3 months, 6 months, and 12 months postoperatively. Imaging data refers to medical imaging data collected via CTA examinations during these different follow-up stages; for example, CTA imaging data collected at each of patient A's follow-up examinations at 1 month, 3 months, 6 months, and 12 months postoperatively, with approximately 500 slices of images at each time point. Clinical data refers to the patient's clinical indicators obtained from the electronic medical record system during these different follow-up stages; for example, clinical indicators such as blood pressure, heart rate, symptom changes, and medication use recorded at each of patient A's follow-up examinations at 1 month, 3 months, 6 months, and 12 months postoperatively.

[0060] The dynamic change characteristics refer to the quantitative changes extracted by comparing and analyzing postoperative imaging and clinical data with preoperative baseline data. For example, CTA images of patient A 6 months postoperatively show a 25% increase in false lumen volume, a 15% decrease in true lumen diameter, and a 10 mmHg increase in blood pressure compared to preoperative levels; these changes constitute the dynamic change characteristics. The updated risk prediction result refers to the risk probability value for the patient's current stage, recalculated after inputting the dynamic change characteristics into the prognostic prediction model. For example, after inputting patient A's dynamic change characteristics 6 months postoperatively into the optimized prognostic prediction model, the model outputs an updated risk prediction result of 0.78 for the probability of thoracic aortic dilatation at 12 months postoperatively.

[0061] Specifically, during the model usage period, the postoperative dynamic risk monitoring module 106 automatically analyzes the imaging data and TAD occurrence of patients at different follow-up stages after TEVAR, extracts dynamic change features and inputs them into the prognostic prediction model, updates the risk prediction results, and triggers an early warning when the updated risk prediction results exceed a threshold.

[0062] The technical solution of this embodiment uses a deep learning network to automatically segment the aorta, its true lumen, and its false lumen. It extracts multi-dimensional imaging features of spatial angle, morphology, density, texture, and vascular branches within a preset region of interest. It integrates clinical full-dimensional data, hemodynamic parameters, and postoperative follow-up data to form an optimal feature set through feature selection. Based on a survival prognostic analysis algorithm, it constructs a prognostic prediction model and outputs risk prediction results. It also extracts dynamic change features from imaging and clinical data at different follow-up stages after surgery to update the prediction results and trigger early warnings. This solution solves the problems of the traditional 301 subtype classification, such as ambiguous subtype boundaries, high heterogeneity, single-dimensional imaging features, insufficient fusion of multi-source data, and lack of postoperative dynamic monitoring. It can improve the accuracy of risk stratification management of adverse events after TEVAR.

[0063] In one alternative embodiment, the CTA image preprocessing module 101 is specifically used for: The CTA imaging data of the patients with type B aortic dissection were obtained through the medical imaging system interface during the preoperative and postoperative follow-up phases.

[0064] The medical imaging system interface refers to a data interface used to connect to a hospital PACS system or other medical image storage systems to obtain DICOM format image data. For example, patient A's CTA image data is automatically obtained from the hospital PACS system through the medical imaging system interface, which uses the DICOM communication protocol.

[0065] The CTA image data is subjected to voxel value normalization, Gaussian filtering for noise removal, and grayscale histogram equalization correction.

[0066] The corrected CTA image data is automatically located for regions of interest, and then the CTA image data after locating the regions of interest is resampled, voxel spacing is unified, and label is encoded to obtain the preprocessed CTA image data.

[0067] The corrected CTA image data refers to CTA images that have undergone voxel value normalization, Gaussian filtering for noise reduction, and grayscale histogram equalization. For example, patient A's original CTA image data is normalized to map grayscale values ​​to the range of 0-255, then Gaussian filtering is used to remove noise, and finally histogram equalization is used to enhance image contrast, resulting in the corrected CTA image data. The CTA image data with region of interest (ROI) localization refers to image data where the ROI is automatically located and cropped from the corrected CTA image. For example, patient A's corrected CTA image, after using an automatic ROI localization algorithm, identifies the aortic region corresponding to the T4 to T11 vertebrae in the sagittal plane, and crops this region, resulting in the CTA image data with the ROI localized.

[0068] In the above-mentioned optional methods, CTA image data during the preoperative and postoperative follow-up stages are further obtained through the medical imaging system interface. After voxel value normalization, Gaussian filtering noise removal and gray-level histogram equalization correction, automatic region of interest localization, resampling and voxel spacing are implemented to improve the standardization of image data, provide a high-quality data foundation for subsequent aortic segmentation and feature extraction, and reduce the impact of noise interference and data heterogeneity.

[0069] In one alternative embodiment, the fully automated aortic segmentation module 102 is specifically used for: The preprocessed CTA image data is input into a deep learning network built on the nnU-Netv2 framework. Through the adaptive network architecture search mechanism and automatic hyperparameter tuning function of the deep learning network, feature maps are extracted from the whole aorta, the true lumen, and the false lumen in the preprocessed CTA image data.

[0070] The nnU-Netv2 framework refers to an adaptive deep learning medical image segmentation framework that can automatically configure the network architecture and training parameters according to the characteristics of the dataset. For example, when the preprocessed CTA image data of patient A is input into a deep learning network built based on the nnU-Netv2 framework, the framework automatically analyzes the voxel spacing, image size and category distribution of the dataset, and configures the optimal network depth and convolution kernel size accordingly.

[0071] The adaptive network architecture search mechanism refers to the mechanism in the nnU-Netv2 framework that automatically searches for and determines the optimal network structure based on the characteristics of the input data. For example, if the voxel spacing of patient A's CTA image data is 0.5 mm × 0.5 mm × 0.5 mm, the nnU-Netv2 framework automatically selects a U-shaped network structure containing 6 downsampling layers through the adaptive network architecture search mechanism. The automatic hyperparameter tuning function refers to the function in the nnU-Netv2 framework that automatically optimizes training hyperparameters based on the characteristics of the dataset. For example, if patient A's CTA image dataset contains images of 200 patients, the nnU-Netv2 framework sets the batch size to 4, the initial learning rate to 0.01, and the number of training epochs to 1000 through the automatic hyperparameter tuning function.

[0072] In this context, a feature map refers to the intermediate layer output generated by a deep learning network during the feature extraction process. It contains feature representations of the input image at different levels of abstraction. For example, after preprocessing, the CTA image data of patient A is processed through the first convolutional layer of the nnU-Netv2 encoder, generating a 64-channel feature map with the same resolution as the original image. These feature maps preserve the edge and texture information of the aorta.

[0073] The feature map is processed through the residual connections and batch normalization mechanism of the deep learning network, and the processed feature map is restored to the same spatial resolution as the CTA image data through the adaptive upsampling layer of the deep learning network, and the segmentation mask is output.

[0074] Residual connections refer to skip connections across one or more layers in a deep learning network, directly adding the output of the previous layer to the output of the next. For example, the encoder part of patient A's nnU-Netv2 network uses residual connections to directly add the feature map before downsampling to the feature map after upsampling, alleviating the gradient vanishing problem. Batch normalization refers to a technique in deep learning networks that normalizes data for each batch, used to accelerate network convergence and improve model stability. For example, patient A's nnU-Netv2 network applies batch normalization after each convolutional layer, normalizing the mean and variance of the feature map for each batch before scaling and translation. An adaptive upsampling layer refers to a network layer in a deep learning network that restores low-resolution feature maps to high-resolution ones, dynamically adjusting the upsampling method based on the input features. For example, the adaptive upsampling layer of patient A's nnU-Netv2 network progressively upsamples the 16×16×8 resolution feature map output from the last layer of the encoder to the original 512×512×200 resolution, and achieves accurate restoration through a combination of bilinear interpolation and transposed convolution.

[0075] The processed feature map refers to the final output feature map after processing such as residual connections, batch normalization, and upsampling. For example, the feature map output by the adaptive upsampling layer of patient A's nnU-Netv2 network is fused with the features of the corresponding encoder layer through residual connections, and then processed by batch normalization and activation functions to obtain the processed feature map, which has a size of 512×512×200 and 3 channels. Spatial resolution refers to the actual physical space size represented by each voxel in a medical image. For example, the spatial resolution of patient A's original CTA image data is 0.5 mm × 0.5 mm × 0.5 mm, that is, each voxel represents a cube space with a side length of 0.5 mm.

[0076] In the above-mentioned optional methods, the preprocessed CTA image data is further input into a deep learning network built on the nnU-Netv2 framework. With the help of adaptive network architecture search and automatic hyperparameter tuning functions, combined with residual connection and batch normalization processing, the spatial resolution is restored through adaptive upsampling layer, realizing fully automatic and accurate segmentation of the aorta, the true lumen and the false lumen, improving segmentation accuracy, reducing manual intervention, and providing an accurate segmentation mask for subsequent feature extraction.

[0077] In an alternative embodiment, the imaging feature extraction module 103 is specifically used for: Within the preset region of interest, based on the region corresponding to the segmentation mask in the CTA image data, the spatial angle features, morphological features, density and texture features, and vascular branch features are calculated.

[0078] The spatial angle features, morphological features, density and texture features, and vascular branching features are normalized to form the multidimensional imaging feature set.

[0079] In the above-mentioned optional methods, spatial angle, morphology, density and texture and vascular branch features are further calculated based on segmentation mask within a preset region of interest, and normalized to form a multi-dimensional imaging feature set, which depicts the anatomical structure and pathological changes of the aorta, supplementing the limitation of the traditional 301 classification which only focuses on angular features, providing rich imaging biomarkers for prognostic prediction models, and improving the comprehensiveness and accuracy of risk assessment.

[0080] In one alternative approach, the true / false cavity vector angle accumulation module is specifically used for: Based on the relative positions of the true and false cavities in the plane and their positions relative to the center point of the vertebral body, the cumulative value of the vector direction angle of the true and false cavities is automatically calculated to obtain a mathematical representation feature set of the spatial relative positions of the true and false cavities.

[0081] Among them, the mathematical representation feature set of the spatial relative position of the true and false cavities refers to: a high-dimensional feature set reflecting the spatial configuration of the true and false cavities, based on the geometric center point of the true and false cavities, the center point of the vertebral body, and the vectors of the true and false cavities, by calculating the cumulative value of the vector direction angle; for example, the mathematical representation feature set output by the vector angle accumulation module of the true and false cavities of patient A includes 32-dimensional features such as the cumulative value of the direction angle and the rate of change of the vectors of the true and false cavities in three-dimensional space.

[0082] The beneficial effects of adopting the above optional method are as follows: by further accumulating the vector angles of the true and false cavities, the relative spatial positions of the true and false cavities are mathematically quantified to obtain high-dimensional angular features, providing a more refined morphological description for the prognostic prediction model and improving the model's ability to characterize the configuration of the interlayer space.

[0083] In one alternative approach, the follow-up data analysis module is specifically used for: During the model construction phase, using bony landmarks in CTA transverse images as a benchmark, the changes in the overall aortic area at the same level were compared between preoperative CTA images and CTA images from each postoperative follow-up. The changes in aortic volume in the region of interest and the occurrence and development of TAD were obtained by integration, with an aortic volume increase of ≥20% defined as TAD.

[0084] The occurrence and development of TAD refers to whether the patient experiences an increase in aortic volume of ≥20% during the follow-up period after TEVAR surgery, which serves as the outcome label for the prognostic prediction model; for example, if patient A's CTA 6 months after surgery shows an increase in aortic volume of 25% compared to before surgery, it is determined that TAD has occurred.

[0085] The beneficial effects of adopting the above optional approach are: by further using the follow-up data analysis module, the TAD outcome event can be accurately defined during the model building period, providing a reliable gold standard label for the prognostic prediction model and improving the accuracy and reliability of model training.

[0086] In one alternative embodiment, the multi-source data fusion module 104 is specifically used for: The multi-dimensional imaging feature set, the full-dimensional clinical data, the hemodynamic parameters, and the postoperative follow-up data are spliced ​​and fused to construct a fused dataset.

[0087] The fusion dataset refers to a data set formed by combining multi-dimensional imaging feature sets, clinical full-dimensional data, hemodynamic parameters, and postoperative follow-up data. For example, patient A's multi-dimensional imaging feature set contains 128 features, clinical full-dimensional data contains 15 features, hemodynamic parameters contain 8 features, and postoperative follow-up data contains 4 outcome variables. After fusion, a dataset containing 155 features is formed.

[0088] A feature selection algorithm combining LASSO regression was used to reduce the dimensionality of the fused dataset, remove redundant features and retain the core features related to adverse events after TEVAR, thus forming the optimal feature set.

[0089] LASSO regression refers to a linear regression method that uses L1 regularization to achieve feature selection and parameter estimation. For example, in a patient A's fusion dataset containing 155 features, LASSO regression is used for feature selection, with the regularization parameter λ set to 0.01. Ultimately, 12 features with non-zero coefficients are retained as core features. Feature selection algorithms are mathematical methods that select the subset of features that contribute most to the prediction target from a high-dimensional feature set. For example, the multi-source data fusion module for patient A uses a feature selection algorithm combined with LASSO regression to select the 12 features most relevant to post-TEVAR adverse events from the 155 features in the fusion dataset.

[0090] Feature dimensionality reduction refers to the process of reducing the number of features and lowering the dimensionality of data through mathematical transformations or selection methods. For example, the feature dimensionality reduction process for patient A's fusion dataset reduced 155 features to 12 core features, reducing model complexity and the risk of overfitting. Core features refer to features that are highly correlated with TAD after TEVAR surgery and are retained after feature selection algorithms. For example, the core features retained after feature selection for patient A's fusion dataset include 12 features such as false lumen expansion rate, true lumen compression rate, false lumen thrombus density variance, number of distal ruptures, false lumen blood flow velocity, age, duration of hypertension history, and postoperative false lumen volume change rate.

[0091] In the above-mentioned optional methods, multi-dimensional imaging feature sets, clinical full-dimensional data, hemodynamic parameters and postoperative follow-up data are further spliced ​​and fused. Feature dimensionality reduction is performed by using a feature selection algorithm combined with LASSO regression to remove redundant features and retain the core features related to adverse events after TEVAR, forming the optimal feature set, reducing data dimensionality and noise interference, and improving training efficiency and prediction accuracy.

[0092] In an alternative embodiment, the intelligent prognostic prediction model module 105 is specifically used for: The optimal feature set is randomly divided into a training set and a validation set according to a preset ratio.

[0093] The preset ratio refers to the pre-defined ratio of the number of samples between the training set and the validation set when partitioning the dataset. For example, the intelligent prognosis prediction model module for patient A randomly partitions the optimal feature set into a training set and a validation set according to a preset ratio of 8:2, with 80% of the samples used for model training and 20% used for model validation. Random partitioning refers to the method of allocating samples to different subsets using random sampling during the dataset partitioning process. For example, after random partitioning the optimal feature set of 200 patients, the data from 160 patients are allocated to the training set, and the data from 40 patients are allocated to the validation set.

[0094] The training set refers to the set of sample data used to train the prognostic prediction model. For example, the optimal feature set of 160 patients constitutes the training set, which is used to build the initial prognostic prediction model, allowing the model to learn the relationship between features and adverse events. The validation set refers to the set of sample data used to evaluate the performance of the trained model and optimize hyperparameters, and does not participate in the model training process. For example, the optimal feature set of 40 patients constitutes the validation set, which is used to evaluate the predictive accuracy of the initial prognostic prediction model and for hyperparameter optimization in the grid search method.

[0095] At least one of the following algorithms—random survival forest, survival support vector machine, or Cox proportional hazards model—is used as the survival prognostic analysis algorithm to construct an initial prognostic prediction model on the training set.

[0096] The initial prognostic prediction model refers to the prognostic prediction model initially constructed on the training set without hyperparameter optimization. For example, the initial prognostic prediction model constructed on the training set by the random survival forest algorithm for patient A contains 100 decision trees, each with a maximum depth of 5, and the C-index of this model on the validation set is 0.72.

[0097] The hyperparameters of the initial prognostic prediction model are optimized using cross-validation and grid search to obtain the optimized prognostic prediction model.

[0098] Cross-validation and grid search refer to a hyperparameter optimization method that combines cross-validation and grid search. The optimal hyperparameters are selected by traversing hyperparameter combinations and evaluating their cross-validation performance. For example, the initial prognostic prediction model for patient A was optimized using 5-fold cross-validation and grid search, traversing combinations with 50, 100, and 200 decision trees and maximum depths of 3, 5, and 7. The combination with the highest average C-index in 5-fold cross-validation was ultimately selected. The optimized prognostic prediction model refers to the model whose performance reaches its best after hyperparameter optimization. For example, after optimization using cross-validation and grid search, the initial prognostic prediction model for patient A was retrained with a hyperparameter combination of 200 decision trees and a maximum depth of 5. The resulting optimized prognostic prediction model achieved a C-index of 0.81 on the validation set.

[0099] The validation set is input into the optimized prognostic prediction model, and the TAD risk prediction results after TEVAR are output for patients with type B aortic dissection.

[0100] In the above-mentioned optional methods, the optimal feature set is further randomly divided into a training set and a validation set according to a preset ratio. Random survival forest, survival support vector machine or Cox proportional hazards model are used as survival prognostic analysis algorithms. An initial prognostic prediction model is built on the training set. The hyperparameters are optimized by cross-validation and grid search to improve the model's generalization ability and accurately output the prediction results of adverse event risks after TEVAR.

[0101] In an alternative embodiment, the postoperative dynamic risk monitoring module 106 is specifically used for: The imaging and clinical data of the patients with type B aortic dissection were obtained at different follow-up stages after surgery.

[0102] The imaging data and clinical data are compared and analyzed with the corresponding preoperative baseline data to extract dynamic change features.

[0103] The dynamic change characteristics are input into the optimized prognostic prediction model to obtain the updated risk prediction result.

[0104] An early warning is triggered when the updated risk prediction result exceeds a preset threshold.

[0105] The preset threshold refers to a pre-set critical value for risk prediction results used to trigger an early warning. In this embodiment, the preset threshold is set to 0.4 by default (it can also be adjusted according to the actual situation, and no restriction is set here). When the updated risk prediction result exceeds 0.4, an early warning is automatically triggered.

[0106] In the above-mentioned optional methods, imaging data and clinical data at different follow-up stages after surgery are further obtained and compared with preoperative baseline data to extract dynamic change features. The dynamic change features are then input into the optimized prognostic prediction model to update the risk prediction results. When the results exceed a preset threshold, an early warning is triggered to achieve dynamic monitoring and real-time early warning of postoperative risks, timely capture of changes in the condition, assist in clinical intervention, and reduce the risk of adverse events.

[0107] In another embodiment of the aortic dilation risk prediction system 100 after endovascular repair of type B aortic dissection of the present invention, the aortic dilation risk prediction system 100 after endovascular repair of type B aortic dissection is a modular computer software system, including ten core modules: CTA image preprocessing module, fully automatic aortic segmentation module, imaging feature extraction module, true and false lumen vector angle accumulation module, follow-up data analysis module, multi-source data fusion module, intelligent prognostic prediction model module, postoperative dynamic risk monitoring module, model interpretability module, and cross-population adaptive calibration module. These modules are interconnected, achieving full automation from CTA image input to postoperative dynamic prognostic prediction and clinical decision-making recommendation output.

[0108] The system architecture is as follows: CTA image data is input to the CTA image preprocessing module to obtain preprocessed CTA image data; the preprocessed CTA image data is input to the fully automated aortic segmentation module to output a segmentation mask; the segmentation mask is input to the true and false lumen vector angle accumulation module to obtain a mathematical representation feature set of the spatial relative positions of the true and false lumens; the follow-up data analysis module outputs the occurrence of TAD during the model building period; the multi-dimensional imaging feature set, the mathematical representation feature set of the spatial relative positions of the true and false lumens, clinical full-dimensional data, hemodynamic parameters, and TAD occurrence are jointly input to the multi-source data fusion module, which forms the optimal feature set through splicing, fusion, and feature selection; the optimal feature set is input to the intelligent prognosis prediction model module, based on the... The system utilizes a prognostic analysis algorithm to construct a prognostic prediction model, outputting the predicted risk of TAD after TEVAR in patients with type B aortic dissection. A postoperative dynamic risk monitoring module acquires imaging and clinical data from different follow-up stages, extracts dynamic change features, inputs them into the prognostic prediction model, updates the risk prediction results, and triggers an alert when the updated risk prediction results exceed a preset threshold. A model interpretability module provides a visual interpretation of the risk prediction results, showcasing the contribution of core features. A cross-population adaptive calibration module calibrates the prognostic prediction model parameters based on new population data, improving the model's applicability across different populations. Finally, the system integrates the risk prediction results and dynamic monitoring information to generate and output personalized clinical decision-making suggestions.

[0109] The specific steps of this embodiment include: Step 1: Standardized Acquisition and Structured Storage of Multi-Source Data: CTA imaging data of patients with type B aortic dissection were acquired preoperatively and at various follow-up stages via a medical imaging system interface, in DICOM format. Comprehensive clinical data, including demographic information, underlying medical history, surgical parameters, and laboratory indicators, were acquired through the hospital's electronic medical record system. Hemodynamic parameters, including distal tear location, number of distal tears, and false lumen blood flow velocity, were also acquired. Postoperative follow-up data, including the occurrence of adverse events such as thoracic aortic dilatation, aortic rupture, reoperation, and death, were also acquired. All data were structured and stored in a standardized database. CTA imaging data, comprehensive clinical data, hemodynamic parameters, and postoperative follow-up data were linked and matched using unique patient identifiers.

[0110] Step 2, Adaptive Preprocessing and Fully Automated Precise Segmentation of CTA Images: CTA image data is input into the adaptive preprocessing submodule, which automatically performs voxel value normalization, Gaussian filtering noise removal, and grayscale histogram equalization correction, and automatically locates regions of interest. Following the nnU-Net standard process, image resampling, voxel spacing unification, and label encoding are completed to obtain preprocessed CTA image data. The preprocessed CTA image data is then input into a deep learning network built on the nnU-Netv2 framework. This deep learning network uses an adaptive network architecture search mechanism and automatic hyperparameter tuning to extract features from the entire aorta, the true lumen, and the false lumen in the preprocessed CTA image data, generating feature maps. Residual connections and batch normalization mechanisms are used to process the feature maps, and an adaptive upsampling layer restores the processed feature maps to a spatial resolution consistent with the CTA image data, outputting a segmentation mask. The fully automated aortic segmentation module outputs segmentation masks for each layer.

[0111] Step 3, Extraction and Quantification of Multidimensional Imaging Features: Based on the segmentation mask output in Step 2, within the preset regions of interest corresponding to the T4 to T11 vertebrae, spatial angle features are calculated, including the three-dimensional angle between the centerline of the true lumen and the centerline of the false lumen, the rate of change of angle, and the helical curvature; morphological features are calculated, including the cross-sectional area of ​​the true lumen, the cross-sectional area of ​​the false lumen, the volume of the true lumen, the volume of the false lumen, the rate of change of diameter, the compression rate of the true lumen, the expansion rate of the false lumen, and the irregularity of the thrombus region; density and texture features are calculated, including the mean CT value, the variance of the CT value, the texture features of the gray-level co-occurrence matrix, and the calcification distribution features of the false lumen wall; vascular branch features are calculated, including the number of intercostal artery openings, the diameter of the false lumen supplying branch arteries, the distribution of the false lumen supplying branch arteries, and the distance features between the spinal cord supplying arteries and the false lumen. The extracted spatial angular features, morphological features, density and texture features, and vascular branching features are normalized to eliminate dimensional differences, forming a multi-dimensional imaging feature set, thus transforming the 301 classification qualitative classification into quantitative indicators. Simultaneously, imaging feature sets such as the geometric center points of the aortic true and false lumens, the vertebral body center points, and the true and false lumen vectors are obtained based on segmentation masks.

[0112] Step 4, Accumulation of vector angles of true and false cavities: Based on the relative positions of the true and false cavities in the plane and their positions relative to the center point of the vertebral body, the accumulated values ​​of the vector direction angles of the true and false cavities are automatically calculated to obtain the mathematical representation feature set of the spatial relative positions of the true and false cavities.

[0113] Step 5, Follow-up data analysis and TAD definition: During the model construction period, based on bony landmarks in CTA transverse images, the changes in the overall aortic area at the same level were compared between preoperative CTA images and CTA images from each postoperative follow-up. The changes in aortic volume in the region of interest and the occurrence and development of TAD were obtained by integration. Aortic volume increase of ≥20% was defined as TAD.

[0114] Step 6, Deep Fusion and Feature Selection of Multi-Source Data: The multi-dimensional imaging feature set formed in Step 3, the mathematical representation feature set of the relative spatial positions of true and false cavities output in Step 4, the full-dimensional clinical data, hemodynamic parameters, and the occurrence of TADs output in Step 5 are spliced ​​and fused to construct a fused dataset. A feature selection algorithm combined with LASSO regression is used to reduce the dimensionality of the fused dataset, remove redundant features, and retain the core features related to adverse events after TEVAR to form the optimal feature set.

[0115] Step 7: Training and Construction of the Intelligent Prognostic Prediction Model Integrating Multi-Dimensional Information: The optimal feature set is randomly divided into a training set and a validation set at an 8:2 ratio, and five-fold cross-validation is introduced. At least one of the following algorithms—random survival forest, survival support vector machine, or Cox proportional hazards model—is used as the survival prognostic analysis algorithm to construct an initial prognostic prediction model on the training set. The hyperparameters of the initial prognostic prediction model are optimized using the validation set and grid search method to determine the optimal configuration of the model, resulting in the optimized prognostic prediction model. During the model construction phase, based on the mathematical representation feature set of the relative spatial positions of the true and false lumens and the occurrence of TAD, a TAD risk prediction model for patients with type B aortic dissection in this region or medical center is constructed and output using survival analysis, random forest, and support vector machine algorithms.

[0116] Step 8, Deployment of the Postoperative Dynamic Risk Monitoring and Real-time Early Warning Model: Deploy the trained and optimized prognostic prediction model to the clinical terminal to construct the postoperative dynamic risk monitoring module. The postoperative dynamic risk monitoring module supports the re-input of CTA imaging data, clinical data, and hemodynamic parameters at different follow-up stages after surgery. The follow-up data is compared and analyzed with the preoperative baseline data to extract dynamic change features, including the rate of change in false lumen volume, the rate of change in true lumen diameter, and blood pressure changes. These dynamic change features are input into the optimized prognostic prediction model to achieve dynamic prognostic prediction at 1 month, 3 months, 6 months, 12 months, and long-term postoperatively. When the optimized prognostic prediction model predicts a risk of adverse events such as thoracic aortic dilatation greater than or equal to a preset threshold of 0.4, the system automatically triggers a real-time early warning and displays a warning message on the clinical terminal. During the model's usage period, the system automatically analyzes the imaging data and TAD occurrence at different follow-up stages after TEVAR, extracts dynamic change features, inputs them into the prognostic prediction model, updates the risk prediction results, and triggers an early warning when the updated risk prediction results exceed the threshold.

[0117] Step 8, Prediction Result Output and Clinical Decision Recommendation Generation: The risk prediction results output by the intelligent prognosis prediction model module and the dynamic monitoring results from the postoperative dynamic risk monitoring module are integrated and output to the clinical terminal in the form of a visual report. Simultaneously, personalized clinical decision recommendations are automatically generated based on quantitative risk stratification: For high-risk patients, intraoperative intercostal artery pre-embolization combined with extended stent closure is recommended, along with strict postoperative blood pressure and heart rate control and increased follow-up frequency; for low- and intermediate-risk patients, routine TEVAR surgery is recommended, with a wider range of blood pressure control and reduced follow-up frequency.

[0118] The technical solution of this embodiment has the following beneficial effects at different levels: In terms of technical performance, by fusing multi-dimensional imaging features and four-dimensional multi-source data, the shortcomings of existing technologies in terms of single features and fragmented data are made up for. Combined with the nonlinear modeling capability of survival prognosis analysis algorithm, the model significantly improves the time-dependent AUC of the prediction of adverse events such as thoracic aortic dilatation after TEVAR, thereby improving the prediction accuracy, refinement and generalization.

[0119] At the clinical application level, the dynamic risk monitoring and real-time early warning function breaks through the limitations of existing static prediction. It can realize the dynamic monitoring of adverse events throughout the entire process based on imaging data and clinical data at different follow-up stages after surgery, and provide real-time early warning when the risk exceeds the standard, providing a basis for timely clinical intervention. The system realizes full-process automation from CTA image input to clinical decision-making recommendation output, which greatly shortens the processing time for a single patient and improves clinical work efficiency.

[0120] At the medical decision-making level, the system automatically generates personalized clinical decision recommendations based on quantitative risk stratification. It recommends enhanced intervention for high-risk patients, reducing the risk of surgical complications such as spinal cord ischemia and paraplegia by more than 40%. It also develops differentiated postoperative management plans for patients with different risk levels, ensuring that high-risk patients are closely monitored while reducing unnecessary examinations and treatments for medium- and low-risk patients, improving patients' medication and follow-up compliance, and reducing the waste of medical resources.

[0121] In terms of patient benefits, accurate risk prediction and timely intervention recommendations can effectively reduce the incidence of adverse events such as thoracic aortic dilatation, aortic rupture, and reoperation after TEVAR, and improve patients' long-term survival prognosis. Patients with low to intermediate risk can relax their blood pressure and heart rate control range and reduce the frequency of follow-up visits according to the system's recommendations, thereby improving their postoperative quality of life and solving the problem of strict control and follow-up for all patients under the current technology.

[0122] Figure 2 This diagram illustrates a flowchart of an embodiment of a method for predicting the risk of aortic dilation after endovascular repair of type B aortic dissection provided by the present invention. Figure 2 As shown, it includes the following steps: S1. Obtain CTA image data of patients with type B aortic dissection during the preoperative and postoperative follow-up stages, and perform standardized preprocessing on the CTA image data to obtain preprocessed CTA image data. S2. Based on a deep learning network, the aorta, true lumen, and false lumen in the preprocessed CTA image data are automatically segmented, and a segmentation mask is output. S3. Within the preset region of interest, extract spatial angle features, morphological features, density and texture features, and vascular branch features based on the segmentation mask, and normalize the extracted features to form a multi-dimensional imaging feature set. S4. The multi-dimensional imaging feature set, clinical full-dimensional data, hemodynamic parameters and postoperative follow-up data are fused to construct a fused dataset, and the fused dataset is subjected to feature screening to form the optimal feature set; S5. Based on the optimal feature set, construct a prognostic prediction model using a survival prognostic analysis algorithm, and output the TAD risk prediction results after TEVAR for patients with type B aortic dissection. S6. Extract dynamic change features from the imaging and clinical data of the patients with type B aortic dissection at different follow-up stages after surgery and input them into the prognostic prediction model to update the TEVAR postoperative TAD risk prediction results, and trigger an early warning when the updated risk prediction results exceed the threshold.

[0123] In one alternative approach, S1 specifically includes: The CTA imaging data of the patients with type B aortic dissection were obtained through the medical imaging system interface during the preoperative and postoperative follow-up phases. The CTA image data is subjected to voxel value normalization, Gaussian filtering for noise removal, and grayscale histogram equalization correction. The corrected CTA image data is automatically located for regions of interest, and then the CTA image data after locating the regions of interest is resampled, voxel spacing is unified, and label is encoded to obtain the preprocessed CTA image data.

[0124] In one alternative approach, S2 specifically includes: The preprocessed CTA image data is input into a deep learning network built on the nnU-Netv2 framework. Through the adaptive network architecture search mechanism and automatic hyperparameter tuning function of the deep learning network, feature maps are extracted from the whole aorta, true lumen and false lumen in the preprocessed CTA image data. The feature map is processed through the residual connections and batch normalization mechanism of the deep learning network, and the processed feature map is restored to the same spatial resolution as the CTA image data through the adaptive upsampling layer of the deep learning network, and the segmentation mask is output.

[0125] In one alternative approach, S3 specifically includes: Within the preset region of interest, based on the region corresponding to the segmentation mask in the CTA image data, the spatial angle features, morphological features, density and texture features, and vascular branch features are calculated; The spatial angle features, morphological features, density and texture features, and vascular branching features are normalized to form the multidimensional imaging feature set.

[0126] In one alternative approach, S4 specifically includes: The multi-dimensional imaging feature set, the full-dimensional clinical data, the hemodynamic parameters, and the postoperative follow-up data are spliced ​​and fused to construct a fused dataset. A feature selection algorithm combining LASSO regression was used to reduce the dimensionality of the fused dataset, remove redundant features and retain the core features related to adverse events after TEVAR, thus forming the optimal feature set.

[0127] In one alternative approach, S5 specifically includes: The optimal feature set is randomly divided into a training set and a validation set according to a preset ratio; At least one of the random survival forest algorithm, survival support vector machine algorithm, or Cox proportional hazards model is used as the survival prognostic analysis algorithm to construct an initial prognostic prediction model on the training set. The hyperparameters of the initial prognostic prediction model were optimized using cross-validation and grid search to obtain the optimized prognostic prediction model. The validation set is input into the optimized prognostic prediction model, and the TAD risk prediction results after TEVAR are output for patients with type B aortic dissection.

[0128] In one alternative approach, S6 specifically includes: Acquire imaging and clinical data of the patients with type B aortic dissection at different follow-up stages after surgery; The imaging data and clinical data are compared and analyzed with the corresponding preoperative baseline data to extract dynamic change features; The dynamic change characteristics are input into the optimized prognostic prediction model to obtain the updated risk prediction result; An early warning is triggered when the updated risk prediction result exceeds a preset threshold.

[0129] It should be noted that the beneficial effects of the aortic dilation risk prediction method after endovascular repair of type B aortic dissection provided in the above embodiments are the same as those of the aortic dilation risk prediction system 100 after endovascular repair of type B aortic dissection, and will not be repeated here. Furthermore, the method and system embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, and will not be repeated here.

[0130] The aortic dilation risk prediction system 100 after endovascular repair of type B aortic dissection of the present invention can be a computer program (including program code) running on a computer device. For example, the aortic dilation risk prediction system 100 after endovascular repair of type B aortic dissection of the present invention is an application software that can be used to execute the corresponding steps in the aortic dilation risk prediction method after endovascular repair of type B aortic dissection of the present invention.

[0131] In some embodiments, the aortic dilation risk prediction system 100 after endovascular repair of type B aortic dissection of the present invention can be implemented in a combination of hardware and software. As an example, the aortic dilation risk prediction system 100 after endovascular repair of type B aortic dissection of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the aortic dilation risk prediction method after endovascular repair of type B aortic dissection of the present invention. For example, the processor in the form of a hardware decoding processor can employ one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0132] The modules described in the embodiments of this invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.

[0133] An electronic device according to an embodiment of the present invention 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, it implements any of the above-mentioned methods for predicting the risk of aortic dilation after endovascular repair of type B aortic dissection. That is, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store the computer program; the processor is used to execute the method for predicting the risk of aortic dilation after endovascular repair of type B aortic dissection as shown in any embodiment of the present invention by calling the computer program.

[0134] In one alternative embodiment, an electronic device is provided, such as Figure 3 As shown, Figure 3The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present invention.

[0135] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0136] Bus 4002 may include a path for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The bus 4002 is represented by only one thick line, but this does not mean that there is only one bus or one type of bus.

[0137] The memory 4003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0138] The memory 4003 stores application code (computer program) for executing the present invention, and its execution is controlled by the processor 4001. The processor 4001 executes the application code stored in the memory 4003 to implement the content shown in the foregoing method embodiments.

[0139] Among them, electronic devices can also be terminal devices. A terminal device can be any terminal device that can install applications and access web pages through applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0140] It should be noted that, Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0141] An embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the above-mentioned methods for predicting the risk of aortic dilatation after endovascular repair of type B aortic dissection.

[0142] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0143] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned method for predicting the risk of aortic dilatation after endovascular repair of type B aortic dissection.

[0144] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0145] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0146] The computer-readable storage medium provided in this invention can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0147] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method shown in the above embodiments.

[0148] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0149] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0150] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0151] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A system for predicting the risk of aortic dilation after endovascular repair of type B aortic dissection, characterized in that, include: The CTA image preprocessing module is used to acquire CTA image data of patients with type B aortic dissection during the preoperative and postoperative follow-up stages, and to perform standardized preprocessing on the CTA image data to obtain preprocessed CTA image data. The fully automated aortic segmentation module is used to perform fully automated segmentation of the aorta, true lumen, and false lumen in the preprocessed CTA image data based on a deep learning network, and output a segmentation mask. The imaging feature extraction module is used to extract spatial angle features, morphological features, density and texture features, and vascular branch features within a preset region of interest based on the segmentation mask, and to normalize the extracted features to form a multi-dimensional imaging feature set. The multi-source data fusion module is used to fuse the multi-dimensional imaging feature set, clinical full-dimensional data, hemodynamic parameters and postoperative follow-up data to construct a fused dataset, and to perform feature filtering on the fused dataset to form the optimal feature set; The intelligent prognostic prediction model module is used to construct a prognostic prediction model based on the optimal feature set and through a survival prognostic analysis algorithm, and output the risk prediction results of TAD after TEVAR in patients with type B aortic dissection. The postoperative dynamic risk monitoring module is used to extract dynamic change features from the imaging and clinical data of the patients with type B aortic dissection at different follow-up stages after surgery and input them into the prognostic prediction model to update the TEVAR postoperative TAD risk prediction results and trigger an early warning when the updated risk prediction results exceed a threshold.

2. The aortic dilation risk prediction system after endovascular repair of type B aortic dissection according to claim 1, characterized in that, The CTA image preprocessing module is specifically used for: The CTA imaging data of the patients with type B aortic dissection were obtained through the medical imaging system interface during the preoperative and postoperative follow-up phases. The CTA image data is subjected to voxel value normalization, Gaussian filtering for noise removal, and grayscale histogram equalization correction. The corrected CTA image data is automatically located for regions of interest, and then the CTA image data after locating the regions of interest is resampled, voxel spacing is unified, and label is encoded to obtain the preprocessed CTA image data.

3. The aortic dilation risk prediction system after endovascular repair of type B aortic dissection according to claim 1, characterized in that, The fully automated aortic segmentation module is specifically used for: The preprocessed CTA image data is input into a deep learning network built on the nnU-Netv2 framework. Through the adaptive network architecture search mechanism and automatic hyperparameter tuning function of the deep learning network, feature maps are extracted from the whole aorta, true lumen and false lumen in the preprocessed CTA image data. The feature map is processed through the residual connections and batch normalization mechanism of the deep learning network, and the processed feature map is restored to the same spatial resolution as the CTA image data through the adaptive upsampling layer of the deep learning network, and the segmentation mask is output.

4. The aortic dilation risk prediction system after endovascular repair of type B aortic dissection according to claim 3, characterized in that, The imaging feature extraction module is specifically used for: Within the preset region of interest, based on the region corresponding to the three-dimensional segmentation mask in the CTA image data, the spatial angle features, morphological features, density and texture features, and vascular branch features are calculated; The spatial angle features, morphological features, density and texture features, and vascular branching features are normalized to form the multidimensional imaging feature set.

5. The aortic dilation risk prediction system after endovascular repair of type B aortic dissection according to claim 4, characterized in that, The multi-source data fusion module is specifically used for: The multi-dimensional imaging feature set, the full-dimensional clinical data, the hemodynamic parameters, and the postoperative follow-up data are spliced ​​and fused to construct a fused dataset. A feature selection algorithm combining LASSO regression was used to reduce the dimensionality of the fused dataset, remove redundant features and retain the core features related to adverse events after TEVAR, thus forming the optimal feature set.

6. The aortic dilation risk prediction system after endovascular repair of type B aortic dissection according to claim 5, characterized in that, The intelligent prognostic prediction model module is specifically used for: The optimal feature set is randomly divided into a training set and a validation set according to a preset ratio; At least one of the random survival forest algorithm, survival support vector machine algorithm, or Cox proportional hazards model is used as the survival prognostic analysis algorithm to construct an initial prognostic prediction model on the training set. The hyperparameters of the initial prognostic prediction model were optimized using cross-validation and grid search to obtain the optimized prognostic prediction model. The validation set is input into the optimized prognostic prediction model, and the TAD risk prediction results after TEVAR are output for patients with type B aortic dissection.

7. The aortic dilation risk prediction system after endovascular repair of type B aortic dissection according to claim 6, characterized in that, The postoperative dynamic risk monitoring module is specifically used for: Acquire imaging and clinical data of the patients with type B aortic dissection at different follow-up stages after surgery; The imaging data and clinical data are compared and analyzed with the corresponding preoperative baseline data to extract dynamic change features; The dynamic change characteristics are input into the optimized prognostic prediction model to obtain the updated risk prediction result; An early warning is triggered when the updated risk prediction result exceeds a preset threshold.

8. A method for predicting the risk of aortic dilation after endovascular repair of type B aortic dissection, characterized in that, include: CTA image data of patients with type B aortic dissection were acquired during the preoperative and postoperative follow-up stages, and the CTA image data were standardized and preprocessed to obtain preprocessed CTA image data. Based on a deep learning network, the aorta, true lumen, and false lumen in the preprocessed CTA image data are automatically segmented, and a segmentation mask is output. Within a preset region of interest, spatial angle features, morphological features, density and texture features, and vascular branch features are extracted based on the segmentation mask, and the extracted features are normalized to form a multi-dimensional imaging feature set. The multi-dimensional imaging feature set, clinical full-dimensional data, hemodynamic parameters and postoperative follow-up data are fused to construct a fused dataset, and the fused dataset is then subjected to feature filtering to form the optimal feature set; Based on the optimal feature set, a prognostic prediction model is constructed using a survival prognostic analysis algorithm, and the risk prediction results of TAD after TEVAR in patients with type B aortic dissection are output. Dynamic change features are extracted from imaging and clinical data of patients with type B aortic dissection at different follow-up stages after surgery and input into the prognostic prediction model to update the TEVAR postoperative TAD risk prediction results. An early warning is triggered when the updated risk prediction results exceed the threshold.

9. An electronic device, characterized in that, The electronic device includes a processor coupled to a memory storing at least one computer program, which is loaded and executed by the processor to enable the electronic device to implement the method for predicting the risk of aortic dilatation after endovascular repair of type B aortic dissection as described in claim 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which, when executed by a processor, implements the method for predicting the risk of aortic dilatation after endovascular repair of type B aortic dissection as described in claim 8.

Images