AI-driven left atrium CT three-dimensional analysis and indentation area automatic evaluation method and system

Abstract

The invention provides an AI-driven left atrium CT three-dimensional analysis and indentation area automatic evaluation method and system. The method comprises the following steps: acquiring three-dimensional medical image data of the heart of a patient; performing structure segmentation on the left atrium and the adjacent organs thereof based on a deep learning model; automatically identifying an indentation area by using a spatial distance calculation strategy; extracting, simplifying and smoothing the indentation contour, and calculating the perimeter and area of the indentation contour; finally, the form and distribution of the indentation area are visually displayed in a two-dimensional superposition and three-dimensional reconstruction mode. According to the method, automatic identification, quantitative analysis and multi-dimensional visualization of the indentation area are achieved, the efficiency and precision of atrial fibrillation preoperative planning are improved, and the method has good clinical application value.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing technology, specifically a deep learning-based medical image analysis method, specifically a technique for organ segmentation, measurement, and risk assessment in CT images before atrial fibrillation surgery. This invention provides a method for automated analysis of the heart and its surrounding critical anatomical structures, assisting clinicians in developing surgical plans and conducting risk assessments. Background Art

[0002] Atrial fibrillation is one of the most common sustained cardiac arrhythmias, with high morbidity and mortality, and has become a major public health issue worldwide. Catheter ablation, one of the main treatments for atrial fibrillation, can cause damage to adjacent critical anatomical structures, such as the esophagus and coronary sinus, leading to serious complications. Therefore, accurate preoperative assessment of the spatial relationship between the left atrium and its surrounding anatomical structures is crucial for improving surgical safety.

[0003] Currently, preoperative evaluation mainly relies on doctors manually identifying and measuring organs on CT images. This method has the following problems:

[0004] 1. Manual analysis is time-consuming and labor-intensive, making it difficult to meet clinical needs;

[0005] 2. The analysis results are easily affected by the operator's experience and subjective judgment, and accuracy and consistency are difficult to guarantee;

[0006] 3. It is difficult to perform accurate quantitative analysis of complex three-dimensional anatomical structures;

[0007] 4. The lack of standardized processes and objective evaluation indicators makes it difficult to achieve comparability between clinical practices.

[0008] In recent years, the development of artificial intelligence (AI), particularly the breakthroughs achieved by deep learning in medical image segmentation, has provided new technical means for automated analysis of the heart and its surrounding structures. However, existing methods primarily focus on the segmentation of individual organs, lack comprehensive analysis of multiple related anatomical structures, and do not specifically address the automated identification and quantitative assessment of indentation areas.

[0009] Therefore, there is an urgent need for a method that can automatically, accurately and efficiently analyze preoperative CT images of patients with atrial fibrillation to achieve the positioning, quantification and visualization of the spatial indentation between the left atrial wall and adjacent organs, providing clinicians with a more comprehensive basis for preoperative planning. Summary of the Invention

[0010] The present invention combines artificial intelligence technology with image processing algorithms to provide an AI-driven method for left atrial CT three-dimensional analysis and automatic assessment of indentation areas, including:

[0011] Step 1: Acquire three-dimensional medical image data of the patient's heart;

[0012] Step 2: Extract left atrial structures from the 3D medical image data based on a deep learning model or image segmentation algorithm, including the left atrium, left atrial wall, esophagus, coronary sinus, aorta, left atrial appendage, and pulmonary veins.

[0013] Step 3: Utilizing the spatial relationship between the left atrial wall and adjacent organs, the indentation area is automatically located and point sets are extracted based on the anatomical structure label map and spatial distance calculation method;

[0014] Step 4: Perform geometric contour extraction and quantitative analysis on the identified left atrial indentation area;

[0015] Step 5: Visual display of left atrial multiple indentation results.

[0016] Furthermore, in step 1, the three-dimensional medical image data may be any medical image data containing complete left heart structural information of a human body, including CT images or MRI images.

[0017] Furthermore, in step 2, medical software is used to manually annotate structures, or an artificial intelligence deep learning algorithm is used to automatically segment the main anatomical structures of the left heart region. The deep learning algorithms include STU-Net-B and SwinUNETR. The segmented left heart regions are the left atrium, left atrial wall, coronary sinus, aorta and esophagus.

[0018] Furthermore, the step 3 includes the following steps:

[0019] Step 3.1: Based on the anatomical structure segmentation results obtained by the deep learning model, a 3D label map of each anatomical structure is extracted. The label map is a 3D voxel grid data structure, in which each voxel position has a corresponding label value and a 3D spatial coordinate index. Different label values ​​are assigned to the left atrial wall, aorta, coronary sinus, and esophagus. To improve the accuracy and stability of subsequent calculations, the segmentation results are preprocessed, including removing isolated noise points and filling small holes to ensure the boundary continuity and structural integrity of the labeled area.

[0020] Step 3.2: Construct a spatial distance calculation strategy based on label masks. Using the label mask of the left atrial wall as the reference region, perform voxel-level spatial distance calculations with the target regions represented by the label masks of the aorta, coronary sinus, and esophagus. For each pair of voxels between the reference region and the target region, use the fast Euclidean distance transform algorithm to calculate their shortest spatial distance. The specific calculation method is as follows:

[0021]

[0022] Where p1(x1, y1, z1) and p2(x2, y2, z2) are the spatial coordinates of any two voxels in any two label maps. By setting a distance threshold, preferably 2 mm, all voxel pairs with a distance less than the threshold are screened as the point set of the potential indentation area. To improve the efficiency of calculation and screening, all label voxels are spatially indexed based on the KD tree data structure to accelerate the search process of the nearest neighbor voxels.

[0023] Step 3.3: Perform contour extraction and geometric optimization on the obtained indentation area point set. First, project the 3D point set onto a preset 2D plane to obtain the projected 2D point set of the indentation area:

[0024]

[0025] Among them, (x i ,y i ) represents the coordinates of the indentation point on the two-dimensional plane;

[0026] Based on this point set, the Alpha-shape algorithm is applied to extract the boundary contour of the indentation area, and the contour is defined as follows:

[0027] Γ α =Alpha-shape(P indentation ,α)

[0028] Among them, α is a parameter that controls the complexity of the boundary shape, and its value affects the degree of contour shrinkage and the effect of local feature preservation;

[0029] After obtaining the preliminary contour, the boundary contour is further geometrically optimized: first, the Douglas-Peucker algorithm is used to simplify the contour, removing redundant vertices while maintaining the main geometric features of the contour; then, the simplified contour is smoothed using a B-spline curve to improve the continuity and smoothness of the contour, providing an optimized boundary input for subsequent quantitative analysis.

[0030] Furthermore, in step 4, the combined analysis and calculation of the indentation area includes the following steps:

[0031] Step 4.1: Based on the extracted indentation contour data, calculate the boundary perimeter of the indentation area. The contour is usually composed of several straight segments and curved segments. To calculate the perimeter, first sum the lengths of each segment. For straight segments, their lengths can be calculated directly based on their endpoint coordinates. For curved segments, piecewise linear approximation is used to approximate them into multiple short line segments to improve the accuracy of the calculation. The total perimeter of the indentation area is obtained by summing the lengths of all contour segments. The key to this process is to accurately approximate the curved segments to ensure the accuracy of the overall perimeter.

[0032] Step 4.2: Since the contour of the indentation area is a closed curve in three-dimensional space, the area calculation method of three-dimensional polygonal patches can be used for processing. Specifically, the spatial surface enclosed by the contour is triangulated into multiple triangular patches, the area of ​​each triangular patch is calculated separately, and the areas are summed to obtain the total area of ​​the entire indentation area.

[0033] Furthermore, in step 5, the original CT image is first registered and superimposed with the structural mask generated by automatic segmentation to accurately display the boundaries of each organ. Subsequently, the mask is reconstructed in three dimensions using 3D reconstruction technology, so that the indentation area is clearly presented in a structured form in three-dimensional space. Finally, the indentation area, CT image, and reconstructed mask are comprehensively superimposed to intuitively display the spatial location, morphology, and characteristics of the indentation and its relationship with the surrounding anatomical structures, comprehensively assisting in the assessment of the distribution and risk characteristics of the indentation area from multiple perspectives.

[0034] The present invention also provides an AI-driven left atrial CT three-dimensional analysis and indentation area automatic assessment system for implementing the above method, the system comprising:

[0035] A medical image acquisition module, configured to acquire three-dimensional medical image data including a left atrium region;

[0036] a structure segmentation module, configured to perform anatomical structure recognition on the image data and extract the left atrium, left atrial wall, esophagus, aorta, and coronary sinus, wherein the structure segmentation is achieved by an artificial intelligence deep learning model or a medical image processing algorithm;

[0037] The indentation localization module is used to spatially locate and extract point sets of the indentation area between the left atrial wall and adjacent organs based on the label map and spatial distance calculation strategy;

[0038] a geometric calculation module, configured to perform contour extraction, perimeter calculation, and area calculation on the indentation area, wherein the contour extraction includes Alpha-shape contour construction, Douglas-Peucker algorithm simplification processing, and B-spline curve smoothing processing; and the area calculation includes dividing the projected surface of the indentation area into multiple triangular patches and calculating the area of ​​each patch;

[0039] The visualization display module is used to fuse the indentation area with the CT image and the structural mask to present the spatial position, geometric shape and relative relationship of the indentation with the anatomical structure, which is used to assist clinical preoperative evaluation.

[0040] Technical Effects

[0041] The present invention realizes the precise analysis of CT images and the automated evaluation of indentation areas through an AI-driven preoperative CT three-dimensional analysis and automatic evaluation method of indentation areas for atrial fibrillation. This method can accurately identify the heart and surrounding structures by automatically segmenting, generating masks, and performing three-dimensional reconstruction, and on this basis, achieve precise positioning and morphological analysis of the indentation area. Compared with traditional manual analysis, this method greatly improves the accuracy and consistency of the evaluation, avoids the interference of human factors, and significantly shortens the analysis time. In addition, the visual display of the indentation area not only enhances the understanding of the indentation morphology and distribution, but also provides an intuitive reference for clinicians, improves the accuracy of preoperative evaluation of atrial fibrillation, and provides strong support for the formulation of personalized treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flowchart of the AI-driven left atrial CT three-dimensional analysis and automatic assessment of indentation areas.

[0043] Figure 2 Schematic diagram of the segmentation of key anatomical structures such as the left atrial wall, esophagus, and aorta described in step 2;

[0044] Figure 3 Schematic diagram of the spatial positioning process of the indentation area described in step 3;

[0045] Figure 4 A schematic diagram of the geometric measurement results of the indentation area described in step 4;

[0046] Figure 5 This is a visualization of the indentation area under the 3D fusion view described in step 5. DETAILED DESCRIPTION

[0047] The present invention will be further described below in conjunction with specific examples. It should be understood that the examples are only intended to illustrate the present invention and are not intended to limit the scope of protection of the present invention. In addition, it should be understood that after reading the disclosure of the present invention, those skilled in the art may make various changes or modifications to the present invention, and these equivalent forms also fall within the scope of protection defined by the present invention.

[0048] This paper proposes an AI-driven automatic assessment method for the indentation area of ​​the left atrium and adjacent organs based on medical three-dimensional imaging data. Figure 1 As shown, the following steps are included:

[0049] Step 1: Acquire three-dimensional medical image data of the patient's heart. This step is mainly used to obtain preoperative cardiac CT image data of patients with atrial fibrillation. Specifically, during the acquisition process, it is necessary to ensure that the image data completely covers the left atrium and its adjacent key anatomical structures, including but not limited to the aorta, coronary sinus, esophagus and other important organs that may be indented with the left atrium. Furthermore, the acquired medical image data must undergo standardized preprocessing, including image quality optimization and spatial registration, to ensure the accuracy of subsequent indentation analysis. The preprocessing includes image noise suppression, contrast enhancement, voxel size unification, and spatial registration operations with standard anatomical templates. The preprocessed image data should meet the spatial resolution and clarity standards required for preoperative planning, and provide stable input for subsequent anatomical structure segmentation and automatic identification of indentation areas.

[0050] Step 2: Extract the anatomical structures related to the left atrium from the 3D medical image data. This step is aimed at the needs of preoperative indentation assessment for atrial fibrillation surgery. It focuses on extracting the left atrium and its adjacent key anatomical structures that may form indentations, focusing on the spatial position relationship of the left atrial wall, aorta, coronary sinus and esophagus. The segmentation results are as follows: Figure 2 shown.

[0051] In this step, image segmentation can be achieved through manual annotation or automatic segmentation based on artificial intelligence. For manual annotation, doctors can use medical image processing software such as 3D Slicer or MIMICS to perform three-dimensional annotation of key structures. For automatic segmentation, deep learning-based medical image segmentation network models, such as STU-Net-B (Scalable and Transferable Medical Image Segmentation Models Empowered by Large-Scale Supervised Pre-training) or SwinUNETR (Swin UNEt TRansformers), can be used to perform structural segmentation on CT images.

[0052] The structure segmentation results generated by any of the above methods are converted into a structure label map, which represents the spatial distribution of each anatomical structure at the voxel level. To ensure the accuracy of the label map, the segmentation results are further processed by morphological processing, including removing isolated voxels and filling small holes, to improve the continuity of label boundaries and anatomical consistency.

[0053] Step 3: Spatial positioning of the indentation area. This step is based on the segmentation results obtained in step 2 to achieve accurate positioning of the indentation area between the anatomical structures of the left atrium and adjacent organs. The recognition results are as follows: Figure 3 As shown, it specifically includes the following sub-steps:

[0054] Step 3.1: Extract the anatomical structure label map. Based on the segmentation results of the deep learning model, labels are extracted for key anatomical structures such as the left atrial wall, aorta, coronary sinus, and esophagus. Specifically, a unique label value is assigned to each anatomical structure, and a standardized three-dimensional spatial coordinate index system is constructed to support subsequent distance calculations and spatial registration operations. Furthermore, to ensure the structural continuity and spatial integrity of the label map, the extracted label map is morphologically optimized, including removing isolated noise points (such as isolated small areas) formed during the segmentation process, small holes in the label map, and repairing the discontinuity of the structure boundary.

[0055] Assume the original label graph set is:

[0056]

[0057] The optimized label graph is represented as:

[0058]

[0059] L i represents the label region of the i-th anatomical structure, represents the operation of removing noise from the label graph, Filling the holes in the label area. It is an optimized label map with complete structure and continuous boundaries after removing noise and filling holes.

[0060] Step 3.2: Construct a spatial distance calculation strategy based on the label map. Specifically, based on the label mask area of ​​the left atrial wall, the shortest distance between it and the label mask areas of the aorta, coronary sinus, and esophagus is calculated to identify areas of possible anatomical compression or contact. Furthermore, the fast Euclidean distance transform algorithm is used to calculate the minimum three-dimensional distance between any voxel point pair between the two label areas, thereby realizing spatial distance calculation. The specific calculation formula is as follows:

[0061]

[0062] Among them, p1(x1, y1, z1) and p2(x2, y2, z2) represent the spatial coordinates of any two voxels in the left atrial wall and the target organ label map, respectively, and d euclidean Represents the spatial distance between two points.

[0063] Furthermore, based on clinical experience and intraoperative anatomical tolerance for atrial fibrillation, a distance threshold (preferably 2 mm) was set. All voxel pairs with distances less than this threshold were selected as the point set for potential indentation areas. This distance calculation strategy fully reflects the proximity relationship between anatomical structures and can serve as the basic data support for subsequent indentation area extraction and contour recognition.

[0064] Step 3.3: Indentation contour extraction and optimization. This step extracts the contour and optimizes the geometry of the indentation area point set obtained in step 3.2. It specifically includes the following three stages:

[0065] First, the three-dimensional indentation point set is projected onto the preset two-dimensional observation plane to obtain the two-dimensional point set of the indentation area, which is recorded as:

[0066]

[0067] Among them, (x i ,y i ) is the point set of the indentation area projected onto the two-dimensional plane.

[0068] Apply Alpha-shape algorithm to the point set P indentation Construct the contour and extract the initial boundary contour of the point set:

[0069]

[0070] Among them, α is the key parameter of the Alpha-shape algorithm, which is used to control the complexity of the shape. The smaller α is, the tighter the contour is, which is suitable for preserving local structural features.

[0071] Then, the Douglas-Peucker algorithm is used to calibrate the initial contour. Perform simplification to remove redundant vertices while preserving the overall geometry:

[0072]

[0073] Among them, ∈ is the simplification error tolerance, which is used to control the maximum contour deviation allowed during the simplification process, C simp Represents the simplified outline.

[0074] Finally, the simplified contour is smoothed using B-Spline to improve the continuity and smoothness of the contour and generate the final optimized contour curve.

[0075]

[0076] Where k is the order of the B-spline curve, C final Indicates P indentation The final optimized contour obtained after the point set is processed by the Alpha-shape, Douglas Peucker, and B-Spline algorithms can be directly used for subsequent geometric quantitative analysis such as perimeter and area, and supports three-dimensional space reconstruction and visual rendering.

[0077] Step 4: Quantitative analysis and calculation of the indentation area. In this step, based on the completion of the spatial positioning and contour extraction of the indentation area, its geometric features are systematically quantified, and the key parameters that can be used to evaluate the indentation degree are output. The results are as follows: Figure 4 As shown, it specifically includes the following sub-steps:

[0078] Step 4.1: Calculate the perimeter of the indentation area. This step is used to accurately calculate the perimeter of the indentation area contour. Specifically, the indentation area contour is segmented, with the straight line segment and the curved line segment processed separately. For straight line segments, the distance formula between two points in space is directly used to calculate the length; for curved lines, a piecewise linear approximation method is used to approximate them into several short straight line segments, and the lengths are calculated and accumulated for each segment. Assume that the contour consists of n points arranged in sequence, each of which is:

[0079] p i =(x i ,y i ,z i ), i=1,2,...,n

[0080] The calculation formula for the total perimeter L is:

[0081]

[0082] Among them, ||p i+1 -p i || is the spatial distance of the i-th segment. This parameter is used to measure the overall length of the indentation area boundary and provides an important basis for indentation range assessment.

[0083] Step 4.2: Calculate the Indentation Area. This step calculates the spatial area of ​​the indentation, which is used to quantify the severity of the indentation. Specifically, given the complexity of the indentation area in three-dimensional space, a three-dimensional polygonal patch area calculation method is used. Furthermore, the indentation area is divided into multiple triangular patches through triangulation. The area of ​​each patch is calculated and summed to obtain the total area of ​​the indentation area.

[0084] Suppose the contour consists of n points arranged in sequence, each point is:

[0085] p i =(x i ,y i ,z i ), i=1,2,...,n

[0086] The center point of the contour is:

[0087]

[0088] By p center With two consecutive contour points p i 、p i+1 A triangle can be formed, then the three vertices of the i-th triangle are:

[0089] Then the area of ​​the triangle is:

[0090]

[0091] The total area A of the indentation region is the sum of the areas of n-1 triangles:

[0092]

[0093] This area parameter, as one of the core indicators of indentation quantification results, can be used to support clinical analysis of indentation depth and adjacent tissue risk.

[0094] Step 5: Visual display of indentation analysis results. Figure 5 This step uses a 2D and 3D fusion method to intuitively present the recognition results of the indentation area, facilitating preoperative evaluation and multi-angle observation by doctors. It mainly includes the following two sub-steps:

[0095] Step 5.1: 2D Section Display. In this substep, the structure mask generated by automatic segmentation is precisely registered with the original CT image and displayed superimposed on standard anatomical sections (e.g., coronal, sagittal, and axial planes). Different color markings are used for each structure boundary and indentation area, clearly demonstrating the spatial relationship between the boundaries of each anatomical structure and the indentation area.

[0096] Step 5.2: 3D Reconstruction and Display. This step provides a 3D spatial display of the indentation analysis results. Specifically, 3D reconstruction is performed based on a structural mask, using semi-transparent rendering to present cardiac structures. The spatial distribution characteristics of the indentation area are highlighted or solid. Interactive features such as multi-angle observation and distance measurement are also provided to assist clinicians in assessing the morphology and potential risks of the indentation area from multiple perspectives.

[0097] In order to realize the superposition display of various data in a unified coordinate system, a rigid transformation matrix is ​​used to perform coordinate mapping on the CT image, structure mask, and indentation curve. The conversion formula is as follows:

[0098] Where source∈{CT,mask,curve}

[0099] Among them, T source is the rigid transformation matrix, defined as:

[0100]

[0101] R is the rotation matrix, t is the translation vector, (x source ,y source ,z source ) is the coordinate of the original data. Through the above transformation, the CT image, mask, and space curve can be accurately superimposed and visualized in the same coordinate system.

Claims

1. An AI-driven left atrial CT three-dimensional analysis and automatic assessment method of indentation area, characterized by: include: Step 1: Acquire three-dimensional medical image data of the patient's heart; Step 2: Extract left atrial structures from the 3D medical image data based on a deep learning model or image segmentation algorithm, including the left atrium, left atrial wall, esophagus, coronary sinus, aorta, left atrial appendage, and pulmonary veins. Step 3: Utilizing the spatial relationship between the left atrial wall and adjacent organs, the indentation area is automatically located and point sets are extracted based on the anatomical structure label map and spatial distance calculation method; Step 4: Perform geometric contour extraction and quantitative analysis on the identified left atrial indentation area; Step 5: Visual display of left atrial multiple indentation results.

2. According to the method for left atrial CT three-dimensional analysis and automatic assessment of indentation areas according to claim 1, in step 1, the three-dimensional medical image data can be any medical image data containing complete left heart structural information of the human body, including CT images or MRI images.

3. The method for left atrial CT three-dimensional analysis and automatic indentation area assessment according to claim 2, wherein in step 2, medical software is used to manually annotate structures, or an artificial intelligence deep learning algorithm is used to automatically segment the main anatomical structures of the left heart region, wherein the deep learning algorithm includes STU-Net-B and SwinUNETR. The segmented left heart regions are, in order, the left atrium, left atrial wall, coronary sinus, aorta, and esophagus.

4. The method for left atrial CT three-dimensional analysis and automatic assessment of indentation areas according to claim 3, wherein step 3 comprises the following steps: Step 3.1: Based on the anatomical structure segmentation results obtained by the deep learning model, a 3D label map of each anatomical structure is extracted. The label map is a 3D voxel grid data structure, in which each voxel position has a corresponding label value and a 3D spatial coordinate index. Different label values ​​are assigned to the left atrial wall, aorta, coronary sinus, and esophagus. To improve the accuracy and stability of subsequent calculations, the segmentation results are preprocessed, including removing isolated noise points and filling small holes to ensure the boundary continuity and structural integrity of the labeled area. Step 3.2: Construct a spatial distance calculation strategy based on label masks. Using the label mask of the left atrial wall as the reference region, perform voxel-level spatial distance calculations with the target regions represented by the label masks of the aorta, coronary sinus, and esophagus. For each pair of voxels between the reference region and the target region, use the fast Euclidean distance transform algorithm to calculate their shortest spatial distance. The specific calculation method is as follows: Where p1(x1, y1, z1) and p2(x2, y2, z2) are the spatial coordinates of any two voxels in any two label maps. By setting a distance threshold, preferably 2 mm, all voxel pairs with a distance less than the threshold are screened as the point set of the potential indentation area. To improve the efficiency of calculation and screening, all label voxels are spatially indexed based on the KD tree data structure to accelerate the search process of the nearest neighbor voxels. Step 3.3: Perform contour extraction and geometric optimization on the obtained indentation area point set. First, project the 3D point set onto a preset 2D plane to obtain the projected 2D point set of the indentation area: Among them, (x i ,y i ) represents the coordinates of the indentation point on the two-dimensional plane; Based on this point set, the Alpha-shape algorithm is applied to extract the boundary contour of the indentation area, and the contour is defined as follows: C α =Alpha-shape(P indentation ,a) Among them, α is a parameter that controls the complexity of the boundary shape, and its value affects the degree of contour shrinkage and the effect of local feature preservation; After obtaining the preliminary contour, the boundary contour is further geometrically optimized: first, the Douglas-Peucker algorithm is used to simplify the contour, removing redundant vertices while maintaining the main geometric features of the contour; then, the simplified contour is smoothed using a B-spline curve to improve the continuity and smoothness of the contour, providing an optimized boundary input for subsequent quantitative analysis.

5. The method for left atrial CT three-dimensional analysis and automatic assessment of indentation area according to claim 4, wherein step 4 comprises the following steps: Step 4.1: Based on the extracted indentation contour data, calculate the boundary perimeter of the indentation area. The contour is usually composed of several straight line segments and curved line segments. To calculate the perimeter, first sum the lengths of each line segment. For a straight line segment, its length can be directly calculated based on its endpoint coordinates. For curved segments, a piecewise linear approximation method is used to approximate them into multiple short line segments to improve the accuracy of the calculation. The total perimeter of the indentation area is obtained by accumulating the lengths of all contour segments. The key to this process is to accurately approximate the curved portion to ensure the accuracy of the overall perimeter. Step 4.2: Since the contour of the indentation area is a closed curve in three-dimensional space, the area calculation method of three-dimensional polygonal patches can be used for processing. Specifically, the spatial surface enclosed by the contour is triangulated into multiple triangular patches, the area of ​​each triangular patch is calculated separately, and the areas are summed to obtain the total area of ​​the entire indentation area.

6. According to the method for three-dimensional analysis and automatic assessment of indentation areas of left atrial CT according to claim 5, in step 5, the original CT image is first registered and superimposed with the structural mask generated by automatic segmentation to accurately display the boundaries of each organ; then, the mask is three-dimensionally reconstructed using three-dimensional reconstruction technology, so that the indentation area is clearly presented in a structured form in three-dimensional space; finally, the indentation area, CT image and reconstructed mask are comprehensively superimposed to intuitively present the spatial position, morphology and characteristics of the indentation and its relationship with the surrounding anatomical structures, thereby comprehensively assisting in the assessment of the distribution and risk characteristics of the indentation area from multiple perspectives.

7. An AI-driven left atrial CT three-dimensional analysis and indentation area automatic assessment system for implementing the method according to any one of claims 1 to 6, the system comprising: A medical image acquisition module, configured to acquire three-dimensional medical image data including a left atrium region; a structure segmentation module, configured to perform anatomical structure recognition on the image data and extract the left atrium, left atrial wall, esophagus, aorta, and coronary sinus, wherein the structure segmentation is achieved by an artificial intelligence deep learning model or a medical image processing algorithm; The indentation localization module is used to spatially locate and extract point sets of the indentation area between the left atrial wall and adjacent organs based on the label map and spatial distance calculation strategy; a geometric calculation module, configured to perform contour extraction, perimeter calculation, and area calculation on the indentation area, wherein the contour extraction includes Alpha-shape contour construction, Douglas-Peucker algorithm simplification processing, and B-spline curve smoothing processing; and the area calculation includes dividing the projected surface of the indentation area into multiple triangular patches and calculating the area of ​​each patch; The visualization display module is used to fuse the indentation area with the CT image and the structural mask to present the spatial position, geometric shape and relative relationship of the indentation with the anatomical structure, which is used to assist clinical preoperative evaluation.

Images