A method and system for predicting remodeling risk after aortic dissection surgery

Abstract

This invention provides a method and system for predicting the risk of aortic remodeling after aortic dissection, comprising the following steps: receiving CTA images of the patient's chest and abdomen at multiple time points, such as preoperative and short or mid-operative postoperatively, and obtaining aortic segmentation results based on the CTA images; obtaining the displacement field of each voxel point between the two time points based on the aortic segmentation results, and obtaining vascular deformation mapping results based on the displacement field; extracting multi-dimensional features from the three-dimensional aortic segmentation model and the vascular deformation mapping results respectively; and predicting the risk of adverse aortic remodeling after surgery based on the multi-dimensional features. The beneficial effects of this invention are: by comprehensively utilizing multi-dimensional, dynamic, and static features, the prediction model constructed by this invention theoretically has higher discrimination and calibration than existing single-index models. Therefore, this invention can detect early, local abnormal deformations that cannot be detected by traditional diameter measurement, significantly advancing the risk identification time point.

Description

Technical Field

[0001] This invention relates to the field of medical image analysis and clinical decision support technology, and in particular to a method and system for predicting the risk of remodeling after aortic dissection. Background Technology

[0002] Currently, clinical assessment of postoperative outcomes and long-term risk prediction in Stanford type B aortic dissection (TBAD) patients undergoing thoracic endovascular aortic repair (TEVAR) relies heavily on periodic manual measurements of the "maximum aortic diameter." Physicians manually select and measure the maximum aortic diameter on sequential computed tomography (CTA) images during follow-up. This method has significant drawbacks:

[0003] 1. Lag and Insensitivity: "Maximum diameter" is a lagging indicator of morphological changes. By the time a significant increase in diameter is observed, irreversible damage to the aortic wall has often already occurred, and the patient has missed the optimal intervention window. 2. Simplicity and One-Sidedness: Relying solely on a single diameter cannot comprehensively reflect the complex three-dimensional morphological changes and local biomechanical state of the aorta. 3. Subjectivity and Large Error: Manual measurements exhibit significant intra- and inter-observer variability, resulting in poor repeatability and affecting the consistency of judgment. 4. Inefficiency: Manual segmentation, measurement, and comparison are time-consuming and labor-intensive, making it difficult to handle large-scale clinical follow-up data.

[0004] The academic community has attempted to introduce more advanced indicators, such as hemodynamic parameters based on computational fluid dynamics (CFD) simulations and morphological features of interstitial geometry. However, these techniques generally suffer from the following drawbacks:

[0005] 1. Limitations of Static Analysis: Most studies rely on simulation analysis based on preoperative imaging data at a single time point, failing to capture the dynamic biomechanical evolution of the aortic wall, intima-lamina, and thrombus over time after TEVAR, a process that is the core mechanism driving remodeling. 2. High Technical Barriers: Analytical processes such as vascular deformation mapping (VDM) and CFD simulation are extremely complex, heavily reliant on time-consuming manual modeling, mesh generation, and computational setup by specialized engineers, making integration into routine clinical workflows impossible. 3. Lack of Integrated Models: Existing predictive models often rely on single-dimensional parameters (such as only anatomical morphology or only hemodynamics), failing to systematically integrate multi-dimensional information such as dynamic deformation, static anatomy, and blood flow environment, resulting in limited predictive efficacy. Summary of the Invention

[0006] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a method and system for predicting the risk of remodeling after aortic dissection, in order to solve the problems of static analysis limitations, high technical barriers and lack of integrated models in the prior art.

[0007] To achieve the above and other related objectives, the present invention provides the following technical solution:

[0008] A method for predicting the risk of remodeling after aortic dissection includes the following steps: receiving CTA images of the patient's chest and abdomen at multiple time points, such as preoperative, short-term postoperative, or mid-term postoperative, and preprocessing the CTA images; inputting the preprocessed CTA images into a trained deep learning segmentation model, obtaining aortic segmentation results through the trained deep learning segmentation model, and generating a corresponding three-dimensional geometric surface model based on the aortic segmentation results; registering the three-dimensional aortic segmentation results and their original image data at two consecutive time points, and obtaining the displacement field of each voxel between the two time points based on the registration results;

[0009] The displacement field of each voxel between two time points is mapped onto the surface of the three-dimensional aortic segmentation model at the baseline time point, and the vascular deformation mapping result is obtained based on the mapping result. Multi-dimensional features are extracted from the three-dimensional aortic segmentation model and the vascular deformation mapping result, wherein the multi-dimensional features include dynamic deformation features, anatomical morphology features and hemodynamic features. The multi-dimensional features are input into a trained risk prediction model, and the risk of adverse aortic remodeling in the patient after surgery is predicted by the trained risk prediction model.

[0010] A system for predicting the risk of remodeling after aortic dissection surgery includes: an image data preprocessing module for receiving CTA images of the patient's chest and abdomen at multiple time points, such as preoperative and short-term or mid-term postoperative times, and preprocessing the CTA images; an automatic aortic segmentation and 3D modeling module for inputting the preprocessed CTA images into a trained deep learning segmentation model, obtaining aortic segmentation results through the trained deep learning segmentation model, and generating a corresponding 3D geometric surface model based on the aortic segmentation results; and a vascular deformation mapping analysis module for registering the 3D aortic segmentation results and their original image data at two consecutive time points, obtaining the displacement field of each voxel between the two time points based on the registration results, and mapping the displacement field of each voxel between the two time points to the surface of the 3D aortic segmentation model at the baseline time point, obtaining the vascular deformation mapping results based on the mapping results.

[0011] The multi-dimensional feature automatic extraction and quantification module is used to extract multi-dimensional features from the three-dimensional aortic segmentation model and the vascular deformation mapping results, respectively. The multi-dimensional features include dynamic deformation features, anatomical morphology features and hemodynamic features. The risk prediction module is used to input the multi-dimensional features into a trained risk prediction model and predict the risk of adverse aortic remodeling in the patient after surgery through the trained risk prediction model.

[0012] An electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the aortic dissection postoperative remodeling risk prediction method as described above.

[0013] In one embodiment of the present invention, before receiving CTA images of the patient's chest and abdomen at multiple time points such as preoperative, short-term postoperative, and mid-term postoperative, and before preprocessing the patient's CTA images, the method further includes: acquiring preoperative, short-term postoperative, and mid-term postoperative CTA images of several patients, wherein the CTA images are CTA images with the aortic structure manually annotated by experts; using the CTA images with the aortic structure annotated to train an initial deep learning segmentation model, and obtaining a trained deep learning segmentation model based on the training results.

[0014] In one embodiment of the present invention, before receiving CTA images of the patient's chest and abdomen at multiple time points such as preoperative, short-term postoperative, and mid-term postoperative, and before preprocessing the patient's CTA images, the method further includes: acquiring preoperative, short-term postoperative, and mid-term postoperative CTA images of several patients, as well as clinical outcome labels indicating whether abdominal aortic segment dilation occurred in the mid-term postoperative period; obtaining a three-dimensional aortic segmentation model and vascular deformation mapping results based on the patient's CTA images, and extracting multi-dimensional features from the three-dimensional aortic segmentation model and vascular deformation mapping results respectively; inputting the multi-dimensional features and the mid-term postoperative clinical outcome labels into an initial risk prediction model, optimizing the model hyperparameters through five-fold cross-validation, and obtaining a trained risk prediction model.

[0015] In one embodiment of the present invention, the registration of the three-dimensional aortic segmentation results and their original image data at two consecutive time points, and the obtaining of the displacement field of each voxel point between the two time points based on the registration results, includes: registering the three-dimensional aortic segmentation results and their original image data at two consecutive time points using a highly regularized multi-metric deformable registration algorithm, wherein the algorithm applies rigid transformation constraints to voxels within the aortic segmentation region during the registration process, while allowing elastic deformation of the soft tissue surrounding the aorta.

[0016] In one embodiment of the present invention, the step of mapping the displacement field of each voxel point between two time points to the surface of the three-dimensional aortic segmentation model at the baseline time point, and obtaining the vascular deformation mapping result based on the mapping result, includes: mapping the displacement field of each voxel point between two time points to the surface of the three-dimensional aortic segmentation model at the baseline time point, and calculating the radial displacement or radial strain of each vertex in the three-dimensional aortic segmentation model along its surface normal direction.

[0017] In one embodiment of the present invention, calculating the radial displacement or radial strain of each vertex in the three-dimensional aortic segmentation model along its surface normal direction includes: calculating the radial displacement of each vertex in the three-dimensional aortic segmentation model along its surface normal direction according to the following formula: ;in, The radial displacement of each vertex along the normal direction; The displacement of each vertex is obtained by interpolation from the displacement field of each voxel point; Each vertex in the 3D aortic segmentation model contains its corresponding normal vector.

[0018] In one embodiment of the present invention, the step of extracting multi-dimensional features from the three-dimensional aortic segmentation model and the vascular deformation mapping results includes: extracting dynamic deformation features from the vascular deformation mapping results, extracting anatomical morphological features from the three-dimensional aortic model segmented by preoperative CTA, and performing automated CFD simulation based on the three-dimensional aortic model segmented by preoperative CTA to extract hemodynamic features.

[0019] As described above, the method and system for predicting the risk of remodeling after aortic dissection according to the present invention have the following beneficial effects:

[0020] 1. Strong early warning capability: This invention can detect early and local abnormal deformations that cannot be detected by traditional diameter measurement, significantly advancing the risk identification time point;

[0021] 2. High prediction accuracy: This invention comprehensively utilizes the characteristics of multiple dimensions, dynamics and statics to construct a prediction model that theoretically has higher discrimination and calibration than existing single-index models;

[0022] 3. Fully automated and highly efficient: This invention enables "one-click" analysis, completing the analysis process that would normally take several days of manual processing in just a few hours, greatly improving the efficiency and accessibility of clinical work;

[0023] 4. Objectification and standardization: This invention avoids the subjective bias of manual measurement and outputs standardized quantitative parameters and reports, which is beneficial for data comparison and research between different centers;

[0024] 5. Clear clinical guidance value: The risk level and visualization report output by this invention can directly help doctors identify high-risk patients, develop more intensive follow-up plans or more proactive intervention strategies, and achieve individualized management. Attached Figure Description

[0025] Figure 1 This is an overall flowchart of the aortic dissection postoperative remodeling risk prediction method in the first embodiment of the present invention;

[0026] Figure 2 This is an overall flowchart of the aortic dissection remodeling risk prediction method according to the first embodiment of the present invention;

[0027] Figure 3 This is an overall schematic diagram of the aortic dissection remodeling risk prediction system according to the second embodiment of the present invention;

[0028] Figure 4 This is a schematic diagram of an electronic device according to the third embodiment of the present invention;

[0029] Figure 5 This is a color cloud map visualizing the growth / deformation of aortic dissection in this invention. Detailed Implementation

[0030] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. It should be noted that, unless otherwise specified, the following embodiments and features described herein can be combined with each other.

[0031] The first embodiment of the present invention relates to a method for predicting the risk of remodeling after aortic dissection, the process of which is as follows: Figure 1 and Figure 2 As shown, the details are as follows:

[0032] Step 101: Receive CTA images of the patient's chest and abdomen at multiple time points, including preoperative, short-term postoperative, and mid-term postoperative periods, and preprocess the patient's CTA images.

[0033] Specifically, before receiving CTA images of the chest and abdomen from multiple time points, such as preoperative, short-term postoperative, and mid-term postoperative, the process includes: firstly, acquiring preoperative, short-term postoperative, and mid-term postoperative CTA images of several patients, wherein the CTA images are manually annotated with the aortic structure by experts; then, using the CTA images with the aortic structure annotated, the initial deep learning segmentation model is trained, and the trained deep learning segmentation model is obtained based on the training results.

[0034] More specifically, before receiving CTA images of the chest and abdomen from multiple time points, such as preoperative, short-term postoperative, and mid-term postoperative, the process includes: firstly, acquiring preoperative, short-term postoperative, and mid-term postoperative CTA images of several patients, as well as clinical outcome labels indicating whether abdominal aortic segment dilation occurred in the mid-term postoperative period; then, obtaining a three-dimensional aortic segmentation model and vascular deformation mapping results based on the patients' CTA images, and extracting multi-dimensional features from the three-dimensional aortic segmentation model and vascular deformation mapping results respectively; finally, inputting the multi-dimensional features and mid-term postoperative clinical outcome labels into the initial risk prediction model, optimizing the model hyperparameters through five-fold cross-validation, and obtaining a trained risk prediction model.

[0035] To elaborate further, this step corresponds to the image data preprocessing module, which receives and standardizes DICOM format image data of chest and abdominal CTA at multiple time points, including preoperative, short-term postoperative (e.g., 1-3 months), and mid-term postoperative (e.g., 6-12 months).

[0036] Step 102: Input the preprocessed CTA image into the trained deep learning segmentation model, obtain the aortic segmentation result through the trained deep learning segmentation model, and generate the corresponding three-dimensional geometric surface model based on the aortic segmentation result.

[0037] Specifically, this step corresponds to the automatic aortic segmentation and 3D modeling module, which has a built-in trained deep learning segmentation model (e.g., using a 3D U-Net++ architecture). This module can automatically identify and segment the aortic lumen, false lumen, intima sheet, and major branch arteries, and generate the corresponding 3D geometric surface model (such as an STL format file) with one click.

[0038] Step 103: Register the three-dimensional aortic segmentation results and their original image data at two consecutive time points, and obtain the displacement field of each voxel between the two time points based on the registration results.

[0039] Specifically, a highly regularized multi-metric deformable registration algorithm is used to register the three-dimensional aortic segmentation results and their original image data at two consecutive time points.

[0040] More specifically, this step corresponds to the highly regularized registration unit in the vascular deformation mapping analysis module. The highly regularized registration unit registers the 3D aortic segmentation results and their original image data at two consecutive time points. It employs a "highly regularized multi-metric deformable registration algorithm," which applies rigid transformation constraints to voxels within the aortic segmentation region during registration, while allowing elastic deformation of the soft tissue surrounding the aorta, thereby accurately calculating the displacement field of each voxel between the two time points. The registration process is as follows:

[0041] The basic form of its loss function is: ,in, As a function measuring the similarity between moving and stationary images, it determines the overall alignment direction; the rigid motion penalty term is: ,in, Indicates the aortic region. Represents the L2 norm. , It is the Laplace operator, used to measure the second derivative of the anomalous; the smoothing constraint term for the surrounding tissue is: This function can suppress physiological deformation;

[0042] The above constraints allow deformation of the surrounding tissue, but require the deformation field to remain smooth in space to avoid abrupt discontinuities. and These are regularization weight parameters that control different constraint strengths; It is usually set to a large value (such as 10-100) to ensure that the rigid constraint inside the aorta is strong enough. Typically set to a medium value (e.g., 0.1-1.0), it controls the smoothness of deformation of surrounding tissue, enabling it to capture growth signals at different resolution levels; at low resolution levels, the algorithm primarily aligns the overall structure, while at medium to high resolution levels, it adjusts the regularization weights simultaneously ( and The algorithm can capture subtle local deformations; this progressive strategy ensures that millimeter-level growth signals are not prematurely "flattened"; in the process, implicit alignment of the centerline is achieved, at which point the overall shape has been corrected and the relationship between the baseline and the follow is stable.

[0043] The displacement field in the first stage is: ,according to ,in, It is the displacement vector of each voxel; it can be further expressed as: Subsequently, a uniformly regularized B-spline variable registration based on a multi-resolution, multi-scale standard of mutual information is used, with the loss function being: ,in, , It is a fixed image. It is a moving image, the grid spacing is generally set to 10mm, and the bending energy penalty term is used. Set to 100;

[0044] After further alignment of the local structures, the actual growth during deformation can be captured; at this point, This can be further expressed as: The displacement field after deformable registration can ultimately be expressed as: ;because ,therefore .

[0045] Step 104: Map the displacement field of each voxel between the two time points to the surface of the three-dimensional aortic segmentation model at the baseline time point, and obtain the vascular deformation mapping result based on the mapping result.

[0046] Specifically, the displacement field of each voxel between two time points is first mapped onto the surface of the three-dimensional aortic segmentation model at the baseline time point, and then the radial displacement or radial strain of each vertex in the three-dimensional aortic segmentation model along its surface normal direction is calculated.

[0047] More specifically, this step corresponds to the deformation and visualization unit in the vascular deformation mapping analysis module. The deformation and visualization unit exports the segmented aorta as a geometric model, with each vertex... Each contains its corresponding normal vector. The calculated displacement field Map the aortic 3D model surface to the baseline time point and calculate the radial displacement (unit: mm) or radial strain of each vertex along its surface normal direction, where the baseline time point in this implementation is the patient's previous follow-up time point;

[0048] It is important to note that the voxel displacement field is defined on a 3D voxel matrix, while surface vertices are typically not located at voxel centers; therefore, trilinear interpolation (or B-spline interpolation) is required to extract the voxel displacement field. Interpolation is used to obtain the displacement of each vertex, represented as: The mapped vertex coordinates are: The vertex displacement along the normal direction can be expressed as: The results are displayed as color cloud maps overlaid on the 3D model, visually showing the growth / deformation areas. Please refer to [link / reference needed] for details. Figure 5 .

[0049] Step 105: Extract multi-dimensional features from the three-dimensional aortic segmentation model and the vascular deformation mapping results, respectively.

[0050] Specifically, dynamic deformation features are extracted from the vascular deformation mapping results, anatomical morphological features are extracted from the three-dimensional aortic model segmented by preoperative CTA, and automated CFD simulation is performed based on the three-dimensional aortic model segmented by preoperative CTA to extract hemodynamic features.

[0051] More specifically, this step corresponds to the multi-dimensional feature automatic extraction and quantification module, which automatically extracts three types of features from the above process: 1) Dynamic deformation features: extracted from VDM results, such as the 90th percentile, mean, and standard deviation (deformation heterogeneity) of the radial displacement of the target aortic segment (such as the distal descending aorta); 2) Anatomical morphology features: automatically measured from the three-dimensional model, including: number of ruptures, location (Criado partition), maximum rupture area, initial volume and length of the false lumen, dissection spiral angle, and true lumen compression rate; 3) Hemodynamic features (optional enhancement module): based on the preoperative CTA segmentation model, automated CFD simulation is performed to extract parameters such as flow rate at the main rupture, average wall shear force in the false lumen, and the proportion of low shear force area.

[0052] Step 106: Input the multi-dimensional features into the trained risk prediction model, and use the trained risk prediction model to predict the risk of adverse aortic remodeling in patients after surgery.

[0053] Specifically, this step corresponds to the risk prediction module, which consists of: 1) Prediction model: using machine learning algorithms (such as XGBoost or random forest), trained on a historical dataset containing the above-mentioned multidimensional features and known clinical outcomes (whether adverse remodeling occurred), the model ultimately outputs a continuous risk probability value (0-1) or a discrete risk level (such as low, medium, or high risk); 2) Report generator: automatically generating structured graphic reports, including: patient information, a list of key measurement parameters, a multi-timepoint VDM deformation comparison cloud map, risk probability and level, and comparative analysis suggestions with previous population data.

[0054] It should be noted that the entire method follows a pipeline of "data input → automatic processing → feature fusion → prediction output," with the specific steps as follows: 1. Import the patient's sequential CTA images into the system; 2. The system automatically calls the AI ​​segmentation module to segment the aortic structure in the images at all time points and generate a 3D model; 3. The system automatically performs VDM analysis on the model and images at consecutive postoperative time points, generating deformation atlases and quantifying deformation features; 4. The system simultaneously quantifies anatomical morphological features from the preoperative model and can selectively initiate CFD simulation to obtain hemodynamic features; 5. Input all extracted feature vectors into the pre-trained risk prediction model; 6. The model calculates the risk value, and the report generation module integrates all intermediate results and the final risk to generate a complete PDF or webpage format report.

[0055] It should also be noted that this invention is designed based on the following principles:

[0056] 1. Biomechanical Dynamic Evolution Principle: Postoperative remodeling of aortic dissection is a dynamic process. Early, subtle changes in biomechanical properties (manifested as local abnormal deformation) precede macroscopic morphological changes (diameter increase). Therefore, capturing early dynamic deformation signals (VDM) is key to early warning. 2. Multi-parameter Integrated Prediction Principle: A single indicator is insufficient to comprehensively describe complex pathophysiological processes. This invention integrates three complementary types of information: "dynamic deformation" (reflecting the current mechanical state), "static anatomy" (reflecting initial conditions), and "blood flow environment" (reflecting driving forces). The constructed prediction model is more consistent with the disease pathogenesis mechanism and theoretically has higher prediction accuracy. 3. Artificial Intelligence-Enabled Clinical Process Principle: By encapsulating expert-level manual complex operations (segmentation, registration, modeling) into automatically executable AI models and algorithm processes, advanced biomechanical analysis techniques can be transformed from laboratory tools into routine clinical tools, solving their "usability" bottleneck.

[0057] In practical applications, an example is given: Taking the prediction of false lumen dilation of the abdominal aorta segment (defined as a diameter increase ≥5mm) one year after TEVAR surgery as an example, the detailed fabrication process is as follows:

[0058] Step 1: 1) Data preparation and model training phase: Collect imaging data from a retrospective patient cohort (e.g., n=200 cases), including: CTA images of each patient before surgery, 3 months after surgery, and 1 year after surgery, as well as clinical outcome labels for whether abdominal aortic segment dilatation occurred 1 year after surgery.

[0059] 2) Training the AI ​​segmentation model: Using approximately 150 data points, the aortic structure was manually annotated by experts as the gold standard. A 3D U-Net++ network was used, with CTA images as input and segmentation masks as output, for training until the Dice similarity coefficient reached above 0.95 on the independent validation set.

[0060] 3) Construct and train the risk prediction model: a) Use this system to automatically process the 150 training data: automatic segmentation, VDM analysis (comparing 3 months postoperatively with preoperative data), and extraction of all three-dimensional anatomical features; b) For each patient, form a set of feature vectors, such as: [VDM_radial displacement_P90 value, number of ruptures, maximum rupture area, initial volume of false lumen, ...]; c) Use these feature vectors and 1-year clinical outcome labels as the training set, input them into the XGBoost algorithm for training, optimize the model hyperparameters through five-fold cross-validation, and obtain the final prediction model M;

[0061] The process of using these feature vectors and 1-year clinical outcome labels as a training set, inputting them into the XGBoost algorithm for training, and optimizing the model hyperparameters through five-fold cross-validation to obtain the final prediction model M is as follows:

[0062] (1) Label correspondence: The feature vector generated for each patient above is matched one-to-one with the known clinical outcome label of the patient 1 year after surgery (whether abdominal aortic segment dilatation occurred, usually a binary label: 0 represents not occurring, 1 represents occurring). Thus, a training set containing 150 samples is obtained, each sample consisting of a feature vector and a binary label.

[0063] (2) Model Training and Hyperparameter Optimization (Five-fold Cross-validation): Initialize the XGBoost model: Select an XGBoost classifier suitable for binary classification tasks (such as XGBClassifier) ​​and set its basic parameters, such as setting the learning objective to "binary:logistic"; Define the hyperparameter search space: In order to find the best combination of hyperparameters, a parameter grid to be searched needs to be defined; These parameters control the behavior and complexity of the model, such as: max_depth: the maximum depth of the tree, which controls the complexity of the model, such as the trial value [3,5,7]; n_estimators: the number of weak learners (decision trees), such as the trial value [50,100,150]; learning_rate: the learning rate, which controls the contribution of each weak learner to the final model, such as the trial value [0.01,0.1,0.2]; subsample: the proportion of samples used when training each tree, which is used to prevent overfitting, such as the trial value [0.7,0.9]; Other regularization parameters such as reg_alpha and reg_lambda;

[0064] (3) Perform five-fold cross-validation combined with grid search (GridSearchCV): Data partitioning: The training data of 150 patients is randomly and uniformly divided into 5 non-overlapping subsets, called "folds"; Training and validation in cycles: This process will be carried out in 5 rounds; In each round, one fold of data is selected as the validation set to evaluate the performance of the model under the current hyperparameter combination, and the remaining 4 folds of data are merged as the training set to train the model;

[0065] (4) Generate the final risk prediction model: After determining the optimal combination of hyperparameters, proceed to the final model generation stage;

[0066] (5) Training the final model M: Using the optimal hyperparameter combination obtained in the previous step, reconfigure the XGBoost classifier; then, use the entire 150 training sets (instead of 4 / 5 of them) as training data to train the model completely; the model M trained in this way can make the most of all available training data information, and it is the final risk prediction model that can be used for clinical prediction.

[0067] Step Two: Application Stage for Predicting New Patients: 1) Obtain preoperative and 3-month postoperative CTA images of a new patient, "Zhang San"; 2) Input the images into the system, and the AI ​​segmentation module automatically generates 3D models of the aorta at the two time points; 3) The VDM analysis module automatically registers the two models, calculates the aortic deformation field at 3 months postoperatively relative to the preoperative value, and quantifies the deformation characteristics of the abdominal aortic segment (e.g., radial displacement P90 = 2.1 mm); 4) The feature extraction module automatically measures anatomical features such as the number of ruptures = 2 and the maximum rupture area = 85 mm² from the preoperative model; 5) Input the feature vector [2.1, 2, 85, ...] into the trained prediction model M; 6) Model M outputs a risk probability of 0.76 (high risk), and the report generation module automatically integrates the deformation cloud map, measurement data, and risk conclusions to generate a report.

[0068] In practical use, the method of using this invention is as follows:

[0069] 1. Installation and Deployment: Deploy this system software on the auxiliary diagnostic server of the hospital's Picture Archiving and Communication System (PACS), or install it on a workstation that meets the configuration requirements; 2. Data Import: Clinicians or technicians can automatically retrieve or manually upload the DICOM data of the target patient's sequence CTA images from the PACS system using the patient ID in the software interface;

[0070] 3. Start Analysis: Select the analysis task (e.g., "Prediction of Remodeling Risk 1 Year Post-TEVAR") in the software interface and click the "Start Analysis" button; the system will automatically execute all background processing steps; 4. View Report: After the analysis is completed (usually within a few hours), the system will issue a notification; users can view the generated PDF report in the software interface or a specified folder; the first page of the report contains a risk summary and recommendations, and subsequent pages contain a detailed list of parameters and a visualization of VDM deformation;

[0071] 5. Clinical Decision-Making: Based on the high-risk indication in the report, physicians may consider shortening the next CTA follow-up interval for the patient (e.g., from 1 year to 6 months) or initiating more aggressive drug therapy; for low-risk patients, the follow-up interval may be appropriately extended to reduce unnecessary radiation exposure and medical costs. The report can be archived as part of the electronic medical record.

[0072] The second embodiment of the present invention relates to a risk prediction system for remodeling after aortic dissection surgery. Please refer to [link to relevant documentation]. Figure 3 ,include:

[0073] The image data preprocessing module is used to receive CTA images of the patient's chest and abdomen at multiple time points, such as before surgery and short or mid-term after surgery, and to preprocess the patient's CTA images.

[0074] The automatic aortic segmentation and 3D modeling module is used to input the preprocessed CTA image into the trained deep learning segmentation model, obtain the aortic segmentation result through the trained deep learning segmentation model, and generate the corresponding 3D geometric surface model based on the aortic segmentation result.

[0075] The vascular deformation mapping analysis module is used to register the three-dimensional aortic segmentation results and their original image data at two consecutive time points, and obtain the displacement field of each voxel between the two time points based on the registration results; it is also used to map the displacement field of each voxel between the two time points to the surface of the three-dimensional aortic segmentation model at the baseline time point, and obtain the vascular deformation mapping results based on the mapping results.

[0076] The multi-dimensional feature automatic extraction and quantization module is used to extract multi-dimensional features from the three-dimensional aortic segmentation model and the vascular deformation mapping results, respectively. These multi-dimensional features include dynamic deformation features, anatomical morphology features, and hemodynamic features.

[0077] The risk prediction module is used to input multi-dimensional features into a trained risk prediction model, and then use the trained risk prediction model to predict the risk of adverse aortic remodeling in patients after surgery.

[0078] It is not difficult to see that this embodiment is a system implementation corresponding to the first embodiment, and this embodiment can be implemented in conjunction with the first embodiment. The relevant technical details mentioned in the first embodiment are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the first embodiment.

[0079] It is worth mentioning that all modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this invention, this embodiment does not introduce units that are not closely related to solving the technical problem proposed by this invention; however, this does not mean that other units are absent from this embodiment.

[0080] The third embodiment of the present invention relates to an electronic device; please refer to [link / reference]. Figure 4 ,include:

[0081] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described method for predicting the risk of remodeling after aortic dissection.

[0082] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0083] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.

[0084] The fourth embodiment of the present invention relates to a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method embodiments.

[0085] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0086] In summary, to overcome the shortcomings of the prior art, this invention aims to: 1. Achieve end-to-end automated processing from CTA images to risk reports, significantly reducing the barrier to entry; 2. Non-invasively and accurately quantify early local biomechanical deformation of the aorta postoperatively using VDM technology, thereby shifting the risk warning window forward; 3. Construct a comprehensive and highly accurate multi-parameter risk prediction model by integrating dynamic deformation features, static anatomical features, and hemodynamic features through an artificial intelligence model; 4. Output intuitive and structured automated reports to directly serve clinical decision-making.

[0087] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. All equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this invention should still be covered by the claims of this invention.

Claims

1. A method for predicting the risk of remodeling after aortic dissection surgery, characterized in that, Includes the following steps: Receive chest and abdominal CTA images of the patient at multiple time points, such as before surgery and short or mid-term after surgery, and preprocess the patient's CTA images; The preprocessed CTA image is input into the trained deep learning segmentation model, the aortic segmentation result is obtained through the trained deep learning segmentation model, and the corresponding three-dimensional geometric surface model is generated based on the aortic segmentation result. The three-dimensional aortic segmentation results and their original image data at two consecutive time points are registered, and the displacement field of each voxel between the two time points is obtained based on the registration results. The displacement field of each voxel between two time points is mapped onto the surface of the three-dimensional aortic segmentation model at the baseline time point, and the vascular deformation mapping result is obtained based on the mapping result. Multi-dimensional features are extracted from the three-dimensional aortic segmentation model and the vascular deformation mapping results, wherein the multi-dimensional features include dynamic deformation features, anatomical morphology features and hemodynamic features; The multidimensional features are input into a trained risk prediction model, which then predicts the patient's risk of postoperative aortic remodeling.

2. The method for predicting the risk of remodeling after aortic dissection according to claim 1, characterized in that: Before receiving chest and abdominal CTA images from the patient at multiple time points, such as preoperative, short-term postoperative, or mid-term postoperative, and before preprocessing the patient's CTA images, the procedure further includes: Acquire preoperative, short-term postoperative, and mid-term postoperative CTA images of several patients, wherein the CTA images are CTA images with the aortic structure manually annotated by experts; The initial deep learning segmentation model is trained using CTA images with the aortic structure annotated, and the trained deep learning segmentation model is obtained based on the training results.

3. The method for predicting the risk of remodeling after aortic dissection according to claim 1, characterized in that: Before receiving chest and abdominal CTA images from the patient at multiple time points, such as preoperative, short-term postoperative, or mid-term postoperative, and before preprocessing the patient's CTA images, the procedure further includes: Obtain preoperative, short-term postoperative, and mid-term postoperative CTA images of several patients, as well as clinical outcome labels for whether abdominal aortic segment dilatation occurred in the mid-term postoperative period; A three-dimensional aortic segmentation model and vascular deformation mapping results are obtained from the patient's CTA images, and multi-dimensional features are extracted from the three-dimensional aortic segmentation model and vascular deformation mapping results respectively. The multidimensional features and postoperative mid-term clinical outcome labels are input into the initial risk prediction model. The model hyperparameters are optimized through five-fold cross-validation to obtain a trained risk prediction model.

4. The method for predicting the risk of remodeling after aortic dissection according to claim 1, characterized in that: The process of registering the three-dimensional aortic segmentation results and their original image data at two consecutive time points, and obtaining the displacement field of each voxel between the two time points based on the registration results, includes: A highly regularized multi-metric deformable registration algorithm is used to register the three-dimensional aortic segmentation results and their original image data at two consecutive time points. The algorithm applies rigid transformation constraints to voxels within the aortic segmentation region during the registration process, while allowing elastic deformation of the soft tissue surrounding the aorta.

5. The method for predicting the risk of remodeling after aortic dissection according to claim 1, characterized in that: The process of mapping the displacement field of each voxel between two time points to the surface of the three-dimensional aortic segmentation model at the baseline time point, and obtaining the vascular deformation mapping result based on the mapping result, includes: The displacement field of each voxel between two time points is mapped onto the surface of the 3D aortic segmentation model at the baseline time point, and the radial displacement or radial strain of each vertex in the 3D aortic segmentation model along its surface normal direction is calculated.

6. The method for predicting the risk of remodeling after aortic dissection according to claim 5, characterized in that: The calculation of the radial displacement or radial strain of each vertex in the three-dimensional aortic segmentation model along its surface normal direction includes: The radial displacement of each vertex in the 3D aortic segmentation model along its surface normal direction is calculated using the following formula: ； in, The radial displacement of each vertex along the normal direction; The displacement of each vertex is obtained by interpolation from the displacement field of each voxel point; Each vertex in the 3D aortic segmentation model contains its corresponding normal vector.

7. The method for predicting the risk of remodeling after aortic dissection according to claim 1, characterized in that: The extraction of multi-dimensional features from the three-dimensional aortic segmentation model and the vascular deformation mapping results includes: Dynamic deformation features are extracted from the vascular deformation mapping results, anatomical morphological features are extracted from the three-dimensional aortic model segmented by preoperative CTA, and automated CFD simulation is performed based on the three-dimensional aortic model segmented by preoperative CTA to extract hemodynamic features.

8. A system for predicting the risk of remodeling after aortic dissection, characterized in that: include: The image data preprocessing module is used to receive CTA images of the patient's chest and abdomen at multiple time points, such as before surgery and short or mid-term after surgery, and to preprocess the patient's CTA images. The automatic aortic segmentation and 3D modeling module is used to input the preprocessed CTA image into a trained deep learning segmentation model, obtain the aortic segmentation result through the trained deep learning segmentation model, and generate the corresponding 3D geometric surface model based on the aortic segmentation result. The vascular deformation mapping analysis module is used to register the three-dimensional aortic segmentation results and their original image data at two consecutive time points, and obtain the displacement field of each voxel between the two time points based on the registration results; it is also used to map the displacement field of each voxel between the two time points to the surface of the three-dimensional aortic segmentation model at the baseline time point, and obtain the vascular deformation mapping results based on the mapping results. The multi-dimensional feature automatic extraction and quantization module is used to extract multi-dimensional features from the three-dimensional aortic segmentation model and the vascular deformation mapping results, respectively. The multi-dimensional features include dynamic deformation features, anatomical morphology features and hemodynamic features. The risk prediction module is used to input the multi-dimensional features into a trained risk prediction model, and to predict the risk of the patient developing adverse aortic remodeling after surgery through the trained risk prediction model.

9. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the aortic dissection postoperative remodeling risk prediction method as described in any one of claims 1 to 7.

Images