A method and system for three-dimensional automatic analysis of tricuspid tri-leaflet coaptation status and quantification of coaptation defects based on artificial intelligence

Abstract

This invention discloses a method and system for three-dimensional automatic analysis of tricuspid valve leaflet occlusion status and quantitative analysis of occlusion defects based on artificial intelligence. The method acquires three-dimensional medical image data and extracts target temporal data showing clear tricuspid valve closure and occlusion. After preprocessing, a deep learning segmentation model is used to independently segment the anterior, posterior, and septal leaflets, automatically extracting the leaflet junction points. Distance fields are calculated for the three groups of adjacent leaflets, and the occlusion contact area and occlusion defect area are divided. The defect area and defect rate are further calculated and their positions mapped, generating three-dimensional visualization results and a structured report. The system includes an image data acquisition and preprocessing module, a leaflet independent segmentation module, a junction point extraction module, an occlusion region extraction module, and an occlusion defect quantitative analysis module. This invention can provide a quantitative reference for tricuspid regurgitation assessment and related interventional treatment planning.

Description

Technical Field

[0001] This invention relates to the fields of medical image processing and artificial intelligence technology, and in particular to a method and system based on an artificial intelligence segmentation model for independent segmentation of the tricuspid valve leaflets, automatic three-dimensional analysis of occlusion status, and quantitative analysis of occlusion defects. Technical Background

[0002] The tricuspid valve is located between the right atrium and right ventricle and consists of three leaflets: the anterior leaflet, the posterior leaflet, and the septal leaflet. These leaflets form three occlusal lines (anterior-posterior occlusal line, anterior-septal occlusal line, and posterior-septal occlusal line), resulting in a relatively complex spatial structure. Under normal physiological conditions, when the heart contracts and closes, the leaflets occlude to prevent blood from flowing back from the right ventricle to the right atrium. When the leaflets fail to occlude adequately, forming a coaptation gap, functional tricuspid regurgitation (FTR) may occur. Therefore, tricuspid occlusive defects are the direct morphological basis of functional tricuspid regurgitation. The extent and spatial location of the defects provide a basis for the clamping position planning of transcatheter edge-to-edge repair (TEER-TV) and the treatment decision of transcatheter tricuspid valve replacement (TTVR).

[0003] In clinical practice, echocardiography can be used to roughly assess the severity of tricuspid regurgitation, but due to the limitations of two-dimensional planar imaging, it cannot quantitatively analyze the area and precise location of tricuspid occlusal defects in three-dimensional space. While cardiac CT imaging can provide high-resolution three-dimensional image data, the weak visualization of the tricuspid valve on CT images and the difficulty in identifying the tricuspid leaflet structure pose significant challenges to its three-dimensional segmentation and occlusal analysis. Currently, there is no mature method for independent segmentation of the three leaflets of the tricuspid valve under CT conditions, and the automatic extraction of the tricuspid occlusal line and the automatic identification and quantification of occlusal defect areas are completely lacking.

[0004] Therefore, there is an urgent need to develop an artificial intelligence-based tricuspid valve image analysis technology to achieve independent segmentation of the three leaflets of the tricuspid valve, three-dimensional analysis of the occlusion status, and quantitative assessment of occlusion defects, thereby improving analysis efficiency, consistency, and objectivity.

[0005] This invention proposes an artificial intelligence-based method and system for three-dimensional automatic analysis of tricuspid valve occlusion status and quantitative analysis of occlusion defects. By combining a deep learning segmentation model with spatial distance field analysis and surface integral algorithms, the system can automatically segment the tricuspid valve occlusions independently in CT images, automatically define three occlusion lines, identify occlusion defect regions, and output key indicators such as the area of ​​the occlusion defect and its position in the tricuspid intersecting coordinate system. This method not only fills the technical gap in automatic quantitative analysis of tricuspid valve occlusion defects under CT conditions but also provides clinicians with objective and accurate diagnostic evidence, assisting in the precise assessment of tricuspid regurgitation and the planning of interventional procedures. Summary of the Invention

[0006] This invention provides an artificial intelligence-based method for three-dimensional automatic analysis of tricuspid valve occlusion status and quantitative analysis of occlusion defects, comprising the following steps: Step 1: Acquire and preprocess 3D medical imaging data. Use 3D medical imaging equipment to acquire cardiac images of the patient, extract the end-systolic phase data of the heart with the clearest tricuspid valve closure and occlusion, and perform preprocessing to improve data quality. Step 2: Independent segmentation of the three lobes of the tricuspid lobe. Based on a deep learning segmentation model, the tricuspid lobe region is automatically segmented to obtain three independent curved surface structures: the anterior lobe, the posterior lobe, and the septum. The segmentation results are optimized through prior morphological constraints and morphological post-processing, and the anterior lobe surface is output. , rear leaf surface and leaf curved surface ; Step 3: Automatic extraction of trilobal junctions. Based on the trilobal segmentation results from Step 2, three junctions are automatically extracted at the intersection of the segmentation boundaries of adjacent lobes, namely the anterior-septal junction. Front-rear intersection point and the junction point of the back-separation ; Step 4: Automatic extraction of the mating region. For each of the three pairs of adjacent leaflets, the distance field between the two leaflet surfaces is calculated, and the corresponding regions are divided into mating contact areas and mating defect areas based on the distance threshold. Step 5: Quantitative analysis of occlusive defects, calculate the surface area of ​​the corresponding occlusive defect region for each adjacent leaflet pair, and map the defect region to the trilobal spatial coordinate system with the valve annulus as the reference, outputting the defect area and defect location; Step 6: 3D visualization and result output. Generate a 3D rendering of the tricuspid valve trilobal surface, annotate the occlusal contact area and occlusal defect area in the form of a heat map, and output a structured result report.

[0007] Furthermore, in step 1, the three-dimensional medical imaging device includes cardiac CT, cardiac MRI, or three-dimensional echocardiography equipment, and the preprocessing includes noise removal, grayscale normalization, and multi-phase registration of the image to improve image quality and enhance consistency between different phases.

[0008] Furthermore, in step 2, the independent segmentation of the three lobes employs a three-dimensional segmentation network that integrates prior morphological constraints. The annular plane serves as the geometric prior constraint on the spatial distribution range of each lobe. The three lobes are distinguished using the morphological prior that the anterior lobe has a relatively large area, the septum is relatively narrow, and the posterior lobe is located between the two. Let the segmentation areas of the three lobes be respectively... An area ratio regularization term is introduced into the loss function of the segmentation network:

[0009] in , The regularization weight coefficients are used to ensure that the segmentation results satisfy the following conditions: Anatomical priors; The total loss function of the segmentation network is defined as:

[0010] in For standard segmentation loss, This is the area ratio regularization term. For topological regularization terms, and To balance the weights; After segmentation, a surface reconstruction algorithm is used to convert the trifoliate voxel mask into a triangular mesh surface. , and .

[0011] Further, in step 3, intersection detection is performed on the boundary voxels of the trifoliate segmentation result. Specifically, in the boundary voxel set of any two-leaf segmentation mask, the Euclidean distance between each pair is calculated, and the voxel with the smallest distance that is also adjacent to the boundary of the third leaf is selected as the intersection point; thus, three intersection points are obtained. They are located at the junctions of the anterior leaf and the septum, the anterior leaf and the posterior leaf, and the posterior leaf and the septum, respectively.

[0012] Furthermore, in step 4, three pairs of adjacent blades are... , , Calculate each leaf surface separately each vertex To adjacent leaf surfaces Shortest Euclidean distance:

[0013] in, curved surface A set of vertices, with a set distance threshold. Distance less than The area is defined as the contact zone. The distance is not less than The area is defined as the conjoint defect area. :

[0014]

[0015] Furthermore, in step 5, for each adjacent leaflet pair corresponding to the defect area... The area of ​​the defect is calculated using the triangular mesh facet accumulation addition method; let the set of faces within the defect area be... For any triangular facet Its area is:

[0016] The corresponding defect area of ​​the adjacent leaflet pair is:

[0017] The total area of ​​the apical defect is the sum of the corresponding defect areas of the three adjacent leaflet pairs:

[0018] in, , , These represent the defect areas corresponding to adjacent leaflet pairs in the anterior-posterior, anterior-septum, and posterior-septum directions, respectively.

[0019] Similarly, the total area of ​​the contact zone The contact area patches are accumulated using the same method.

[0020] Further, in step 5, the defective regions are mapped to a cylindrical coordinate system constructed with the center of the lobular annulus as the origin and the trilobal junction as the reference direction. The area-weighted centroid of each defective region is defined as:

[0021] in, Triangular facet The geometric center; Based on the azimuth position of the area-weighted centroid in the trilobal coordinate system, the corresponding defect location label is output, including front-back boundary defect, front-separation boundary defect, and back-separation boundary defect.

[0022] Furthermore, in step 5, the misalignment rate is defined based on the ratio of the total defect area to the total misalignment area:

[0023] This invention also provides an artificial intelligence-based three-dimensional automatic analysis system for tricuspid valve leaflet occlusion status and quantitative analysis of occlusion defects, comprising: The image data acquisition and preprocessing module is used to acquire and preprocess three-dimensional medical image data, extract end-systolic phase data of the heart, and perform noise removal, grayscale normalization and multi-phase registration to improve data quality. The trilobate independent segmentation module is used to independently segment the anterior, posterior and septal lobes of the tricuspid lobe based on a 3D deep learning segmentation model that incorporates prior morphological constraints. It combines area ratio constraints, topological continuity constraints and morphological post-processing to optimize the segmentation results and output trilobate independent triangular mesh surfaces. The boundary point extraction module is used to automatically extract three trifoliate boundary points based on the boundary voxel intersection detection of the trifoliate segmentation results; The mating region extraction module is used to calculate the Euclidean distance field between the curved surfaces of the two leaflets for three pairs of adjacent leaflets, and automatically divide the mating contact area and mating defect area based on a preset distance threshold. The quantification module for occlusive defects is used to calculate the surface area of ​​the corresponding occlusive defect region for each adjacent leaflet pair using the triangular mesh surface accumulation addition method. It also constructs a cylindrical coordinate system based on the valve annulus and the trilobal junction point, maps the centroid of the defect region to the coordinate system, and outputs the defect location label, defect area, and defect rate index. The 3D visualization and report output module is used to generate a 3D rendering of the tricuspid valve leaflet surface, annotate the occlusal contact area and defect area in the form of a heat map, and automatically generate a structured result report containing the defect area, total defect area, defect rate and defect location labels for each adjacent leaflet pair.

[0024] Technical effect: This invention combines artificial intelligence with spatial geometric analysis technology to achieve automated, precise three-dimensional analysis of the tricuspid valve's three-leaflet occlusion state and quantitative analysis of occlusion defects. Its technical effects are as follows: 1. High efficiency: The fully automated process reduces the time required for traditional manual analysis to just a few seconds, significantly improving diagnostic efficiency and making it suitable for large-scale clinical screening scenarios.

[0025] 2. High precision: Based on a deep learning segmentation model that integrates prior morphological constraints, it achieves independent segmentation of three lobes; based on distance field analysis and surface integral algorithm, it achieves accurate quantification of the area of ​​the combined defect, with a precision significantly better than the traditional coarse ultrasound assessment method.

[0026] 3. Consistency: It eliminates the influence of human factors, ensuring a high degree of consistency and reproducibility of analytical results, and is suitable for multicenter studies and clinical applications.

[0027] 4. Adaptability to complex anatomy: Through prior morphological constraints on the segmentation network, it adapts to the low contrast, asymmetrical shape, and non-planar annulus characteristics of the tricuspid valve with three leaflets, making it suitable for personalized analysis of different patients.

[0028] 5. Clinical applicability: The output information on the area and location of the occlusal defect directly serves the planning of the TEER-TV clamping point and the decision-making of TTVR replacement, improving the accuracy and safety of interventional surgery planning.

[0029] In summary, this invention fills the technical gap in the automatic quantitative analysis of tricuspid valve tricuspid occlusal defects under CT conditions, significantly improves the diagnostic accuracy and treatment planning level of tricuspid regurgitation, provides reliable support for clinical decision-making, and has important application value. Attached Figure Description Figure 1 This is a flowchart illustrating the three-dimensional automatic analysis of tricuspid valve occlusion state and quantitative method for occlusion defects based on artificial intelligence, according to the present invention.

[0030] Figure 2 This is a schematic diagram of the tricuspid valve trilobite independent segmentation result described in step 2 of the present invention.

[0031] Figure 3 This is a schematic diagram of the automatic extraction result of the trifoliate junction point described in step 3 of the present invention.

[0032] Figure 4 This is a schematic diagram of the calculation of the distance field of the mating region and the division of the mating contact area and the defect area as described in step 4 of the present invention.

[0033] Figure 5 This is a schematic diagram of the position mapping result of the misaligned defect area in the trilobed coordinate system described in step 5 of the present invention. Detailed Implementation

[0034] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that the embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. Furthermore, it should be understood that after reading the disclosure of this invention, those skilled in the art can make various modifications or alterations to the invention, and these equivalent forms also fall within the scope of protection defined by this invention.

[0035] This invention combines artificial intelligence technology and image processing algorithms to provide an AI-based method and system for three-dimensional automatic analysis of tricuspid valve tricuspid occlusion status and quantitative analysis of occlusion defects. The method's flow is as follows: Figure 1 As shown, the specific steps include: Step 1: CT Image Acquisition and Preprocessing Retrospective electrocardiogram-gated scans of the patient's heart were performed using high-precision cardiac CT imaging equipment to acquire three-dimensional image data covering the entire cardiac cycle. To ensure the accuracy of tricuspid valve occlusion assessment, end-systolic image data was selected as the target analysis data from the three-dimensional medical imaging data. Since the tricuspid valve is closed at this phase, the occlusion between the three leaflets is clearest, thus facilitating subsequent three-dimensional analysis and quantification of occlusion defects.

[0036] To improve the accuracy of subsequent segmentation, modeling, and calculation, the target analysis data is preprocessed, including: Noise Removal: An edge-preserving filtering algorithm is used to reduce random noise introduced during imaging, while retaining important anatomical boundary information.

[0037] Gray-level normalization: Maps image gray-level values ​​to a uniform numerical range [0,1] to eliminate differences caused by different devices or scanning parameters and ensure the consistency of model input.

[0038] Multi-phase registration: For multi-temporal CT image data, non-rigid registration technology is used to align images of different temporal phases, so that the images of each temporal phase maintain spatial consistency in anatomical position, which facilitates subsequent temporal selection and comparative analysis.

[0039] Step 2: Independent splitting of the tricuspid valve's three leaflets This step addresses the low contrast, morphological asymmetry, and non-planar distribution of the tricuspid valve's three leaflets in CT images. It designs a 3D deep learning segmentation network incorporating prior morphological constraints to automatically segment the tricuspid valve region into three independent structures: the anterior leaflet, posterior leaflet, and septal leaflet. The segmentation results are shown below. Figure 2 As shown. Specifically, it includes the following sub-steps: Step 2.1: Design of Prior Morphological Constraints Based on the anatomical features of the tricuspid valve with three leaflets, the following prior morphological constraints are designed: (1) Valve annulus plane constraint: Using the tricuspid valve annulus fitting plane as the geometric prior, the spatial growth range of each leaflet is limited to the ventricular side region below the valve annulus plane to prevent the segmentation results from exceeding the reasonable anatomical space. The valve annulus plane is obtained by least-squares fitting of the valve annulus point set, and its normal vector is denoted as... The center point of the annulus Defined as:

[0040] in, For the annulus One sampling point.

[0041] (2) Trilobular morphology prior: Utilizing the anatomical features of the tricuspid valve—the anterior leaflet having the largest area, the septum being the narrowest and closest to the interventricular septum, and the posterior leaflet having an area between the two—an area ratio constraint is introduced into the loss function of the segmentation network. Let the segmentation areas of the anterior, posterior, and septal leaves be respectively... Introducing the area ratio regularization term, it can be expressed as:

[0042] in, , The regularization weight coefficients are used to ensure that the segmentation results satisfy the following conditions: Anatomical prior relationships.

[0043] (3) Spatial continuity constraint: Add a topological regularization term to the loss function. This ensures that each leaf segmentation result is a simply connected region, avoids fragmented segmentation, and improves the anatomical rationality of the segmentation results.

[0044] Step 2.2: Segmentation Network Architecture and Training A 3D deep learning segmentation network is used to automatically segment the preprocessed target image. Preferably, a hybrid network architecture based on the fusion of a 3D vision Transformer and a state space model can be used as the segmentation backbone network. This network takes 3D voxel data as input and captures local texture details and global contextual information simultaneously through hierarchical feature extraction, making it suitable for segmenting targets with low contrast and subtle morphological differences, such as the tricuspid lobe with three lobes. The network output is a four-channel probability map, corresponding to the anterior lobe, posterior lobe, septum, and background, respectively.

[0045] The total loss function of the network is defined as:

[0046] in, The standard segmentation loss term is a weighted combination of Dice loss and cross-entropy loss. This is the area ratio regularization term. For topological regularization terms, and These are the balancing weighting coefficients.

[0047] Step 2.3: Post-segmentation processing Morphological operations and connected component analysis were performed on the segmentation results of each lobe to remove noise and artifacts. Specifically, this included: (1) Perform morphological closing operations on each petal segmentation mask to fill small holes.

[0048] (2) Extract the maximum connected component for each leaf to remove discrete noise points.

[0049] (3) Smooth the segmentation boundary to obtain a smooth leaflet surface.

[0050] The final result is a trifoliate, independent split surface, denoted as the front leaf surface. , rear leaf surface and leaf curved surface .

[0051] Step 2.4: Surface Reconstruction The segmentation masks for the anterior, posterior, and septal leaflets of the tricuspid valve obtained in step 2.3 are used for surface reconstruction to generate corresponding 3D surface models. Preferably, the Marching Cubes surface reconstruction algorithm is used to convert the voxel masks of each leaflet into corresponding triangular mesh surfaces. After reconstruction, the triangular mesh surfaces corresponding to the anterior, posterior, and septal leaflets are denoted as follows: , and Each triangular mesh surface can be represented as:

[0052] in, For vertex set For a collection of dough pieces, further:

[0053] in, Represents the total number of vertices. This represents the total number of triangular faces.

[0054] In some implementations, each triangular facet This represents a triangle composed of the indices of its three vertices.

[0055] Through the above surface reconstruction process, the discrete voxel segmentation results can be converted into a continuous three-dimensional surface representation, which facilitates subsequent extraction of trilobal junction points, calculation of distance fields between adjacent blades, and quantitative analysis of the area of ​​the mating contact area and the mating defect area.

[0056] Step 3: Automatic extraction of trifoliate junctions This step, based on the independent segmentation results of the anterior, posterior, and septal lobes of the tricuspid valve obtained in step 2, automatically extracts three tricuspid junction points, providing a reference benchmark for subsequent extraction of the synaptic region and construction of the tricuspid spatial coordinate system. The extraction results are as follows: Figure 3 As shown. Specifically, it includes the following sub-steps: Step 3.1: Boundary Voxel Set Extraction Extract the segmentation mask boundary voxel sets for the anterior leaf, posterior leaf, and septum leaf, respectively. For any leaflet... Segmentation mask Its boundary voxel set can be defined as:

[0057] in, voxels The neighborhood of the first leaf is preferably a 26-neighborhood. This yields the front leaf boundary set. Later leaf boundary set and leaf boundary set .

[0058] Step 3.2: Detection of candidate regions at the junction of two leaves For each pair of adjacent leaflets Calculate the boundary voxel set and The pointwise Euclidean distance between them. When the two boundary voxels When the following formula is satisfied:

[0059] Then voxel pairs These are labeled as candidate voxel pairs for the boundary. This is the adjacency threshold. Based on this, the set of midpoints of all candidate voxel pairs is taken as the candidate boundary region between the corresponding two lobes. .

[0060] Step 3.3: Determining the intersection point of the three leaves The boundary candidate region obtained in step 3.2 In the process, voxels that are simultaneously adjacent to the boundary of the third leaf are searched. Specifically, for candidate boundary regions between the anterior and posterior leaves... Search for each candidate point to the set of voxels at the leaf boundary. The point with the smallest distance is denoted as the boundary between the front and back leaves. For the candidate boundary region between the anterior leaf and the septum. Search for the set of voxels from each candidate point to the posterior leaf boundary. The point with the smallest distance is denoted as the junction of the anterior leaf and the septum. For the candidate boundary region between the posterior leaf and the septum. Search for the set of voxels from each candidate point to the front leaf boundary. The point with the smallest distance is designated as the boundary between the posterior leaf and the septum. .

[0061] In some implementations, if the minimum distance is greater than a preset threshold Then, an approximate method can be used to determine the location of the boundary point. For example, take... The intersection of the centroids of the three candidate regions is used as the approximate intersection point.

[0062] Following the method described above, the following three boundary points were obtained: anterior leaf-septal leaf junction : The point located in the region at the junction of the anterior leaf and the septum, and closest to the boundary of the posterior leaf.

[0063] anterior-posterior leaf boundary : The point located in the region where the anterior and posterior leaves meet, and which is closest to the boundary of the septum.

[0064] posterior leaf-septal leaf junction : The point located in the region at the junction of the posterior leaf and the septum, and closest to the boundary of the anterior leaf.

[0065] Through the above processing, the junction of the three leaflets of the tricuspid valve can be automatically extracted, providing a foundation for subsequent analysis of the occlusal region, identification of defect locations, and construction of a tricuspid reference coordinate system.

[0066] Step 4: Automatic extraction of the matching region This step, based on the tricuspid valve triangular mesh surface obtained in step 2, automatically identifies the contact area and defect area between adjacent leaflets through leaflet distance field calculation and threshold determination. The results are as follows: Figure 4 As shown. Specifically, it includes the following sub-steps: Step 4.1: Define the pair of synapses Based on the tricuspid valve anatomy, the three synechiae are defined as the pairs of adjacent leaflets corresponding to each other: (1) Front-back alignment line: front leaf surface With the back leaf surface The combination between them.

[0067] (2) Front-septum mating line: front leaf surface With leaf curved surface The combination between them.

[0068] (3) Rear-separation line: rear leaf surface With leaf curved surface The combination between them.

[0069] Step 4.2: Calculate the interleaf distance field For any pair of adjacent leaves ,in Calculate the surface respectively each vertex To the curved surface The shortest distance is defined as:

[0070] in express Norm, i.e., the Euclidean length of a vector, is similarly used to calculate the norm of a surface. to each vertex shortest distance .

[0071] To obtain a symmetric distance metric, for each vertex... Define symmetric distance:

[0072] in, for On curved surfaces The nearest neighbor on the network, curved surface The set of vertices.

[0073] Step 4.3: Delineation of the mating contact area and the mating defect area Set distance threshold (Unit: mm) Based on the symmetrical distance field, the curved surface region is divided into the contact region and the defect region: In some implementations, when using symmetrical distances for partitioning, for curved surfaces... The set of vertices on a given surface can be defined as follows:

[0074]

[0075] Among them, the distance is less than The area is defined as the contact zone. The distance is not less than The area is defined as the conjoint defect area. .

[0076] In some implementations, the threshold Based on clinical experience and leaflet thickness, the value is preferably set in the range of 2mm to 3mm.

[0077] Step 4.4: Extraction of the boundary of the incoherent region To accurately identify the boundaries of the missing regions, boundary patches are extracted from the region division results obtained in step 4.3. For triangular meshes... Any triangular facet on If the classification labels of the three vertices corresponding to the face are not completely consistent, then at least some of them belong to the category of facet. And at least part of it belongs to If so, the patch is determined to be a boundary patch.

[0078] Based on this, the non-shared edges of all boundary patches are extracted, sorted, and connected to obtain the closed boundary curve of the conjoint defect region.

[0079] Step 5: Quantitative analysis of synaptic defects This step, based on the conjoined defect area obtained in step 4, calculates the defect area and determines the defect location. The results are as follows: Figure 5 As shown. Specifically, it includes the following sub-steps: Step 5.1: Calculation of the area of ​​the missing parts For each adjacent leaflet pair, the corresponding defect area The surface area of ​​the missing region is calculated using the addition of triangular mesh facets. Let the set of facets within the missing region be... That is, all three vertices of the face belong to The set of facets. For any triangular facet... Its area is calculated using the cross product of vectors as follows:

[0080] in express The norm, i.e., the Euclidean length of a vector, gives the area of ​​the defect corresponding to the adjacent leaflet pair:

[0081] Similarly, for the set of facets in the contact area The areas are summed up using the same method to obtain the corresponding contact area of ​​the adjacent leaflet pairs. :

[0082] The total area of ​​the occlusal defect and the total contact area are the sum of the areas corresponding to the three adjacent leaflet pairs:

[0083]

[0084] Step 5.2: Construct a trilobal coordinate system To accurately describe the defect location, a cylindrical coordinate system is constructed based on the lobe annulus and the trilobal junction point obtained in step 3. The specific method is as follows: Step 5.2.1: Using the valve annulus center defined in Step 2.1 As the origin of the coordinate system.

[0085] Step 5.2.2: Using the normal vector of the lobe ring plane The vertical axis direction is the z-axis direction.

[0086] Step 5.2.3: Foreleaf-septum junction The projection direction onto the annular plane is the angular reference zero point. Specifically, the reference direction is defined as follows: for:

[0087] in express Norm, also known as the Euclidean length of a vector. Relative to the origin The unit projection direction on the annular plane.

[0088] Define azimuth angle For the annular plane relative to Angular coordinates.

[0089] Step 5.2.4: Calculate the azimuth angles of the three intersection points on the annular plane. The annular plane is divided into three sectors according to the azimuth angle of the intersection point: Front-separation boundary sector: angle range .

[0090] Front-rear boundary sector: angle range .

[0091] Rear-separation boundary sector: angular range .

[0092] Step 5.3: Mapping the defect location Map the centroids of each missing region to a trifoliate coordinate system. For the missing region... Its area-weighted centroid is defined as:

[0093] in, Triangular facet The geometric center.

[0094] center of mass Project onto the lobe-ring plane and calculate its relative to the reference direction. azimuth Output the defect location label according to its sector: like If so, it is marked as "anterior-septal junction defect".

[0095] like If so, it is marked as "anterior-posterior junction defect".

[0096] like If so, it is marked as "posterior-septal junction defect".

[0097] Step 5.4: Defect Severity Grading The misalignment defect rate is defined as the ratio of the defect area to the total misalignment area.

[0098] in, and These are the total mating defect area and the total mating contact area calculated in step 5.1, respectively. The larger the value, the more severe the defect.

[0099] Step 6: 3D visualization output This step, based on all the aforementioned analysis results, generates intuitive 3D visualizations and structured numerical reports. Specifically, it includes the following sub-steps: Step 6.1: 3D rendering of the trilobal surface The trilobed triangular mesh surface obtained in step 2 , , 3D rendering is performed using preset color differentiation, with the front leaf, back leaf, and lateral leaf each marked with different colors to visually display the spatial shape and relative positional relationship of the three leaves. Simultaneously, the junction points of the three leaves obtained in step 3 are... , , Displayed as highlighted points in the 3D rendering.

[0100] Step 6.2: Annotate the heatmap of the combined state Based on the occlusal contact area and occlusal defect area defined in step 4, a heat map is used to annotate the trilobal surface; for example, the occlusal contact area is marked in green, indicating normal occlusal of the trilobes; the occlusal defect area is marked in red, indicating missing occlusal of the trilobes. The color depth corresponds to the interlobular symmetry distance. The red color is directly proportional to the distance; the greater the distance, the deeper the red, to visually reflect the severity of the defect.

[0101] Step 6.3: Output of Structured Numerical Report Automatically generate structured numerical reports containing the following core metrics: (1) The defect area corresponding to each adjacent leaflet pair: The unit is mm. 2 .

[0102] (2) The contact area of ​​each adjacent leaflet pair: The unit is mm. 2 .

[0103] (3) Total area of ​​the missing parts: The unit is mm. 2 .

[0104] (4) Collateral defect rate: , expressed as a percentage.

[0105] (5) Defect location label: anterior-posterior junction defect, anterior-separation junction defect, posterior-separation junction defect.

[0106] (6) Area-weighted centroid of each defect region Azimuth in a trifoliate coordinate system and radial distance.

[0107] Through the above visualization and structured output, an intuitive display of the tricuspid valve's three-leaf occlusion status and occlusion defects can be achieved.

[0108] This invention also provides an artificial intelligence-based three-dimensional automatic analysis system for tricuspid valve leaflet occlusion status and quantitative analysis of occlusion defects, comprising: The image data acquisition and preprocessing module is used to acquire and preprocess three-dimensional medical image data, including cardiac CT, cardiac MRI or three-dimensional echocardiography data, extract end-systolic phase data, and perform noise removal, grayscale normalization and multi-phase registration to improve data quality and image consistency. The trilobate independent segmentation module is used to independently segment the anterior, posterior and septal lobes of the tricuspid lobe based on a 3D deep learning segmentation model that incorporates prior morphological constraints. It combines area ratio constraints, topological continuity constraints and morphological post-processing to optimize the segmentation results and uses a surface reconstruction algorithm to convert the segmentation mask into a trilobate independent triangular mesh surface. The boundary point extraction module is used for boundary voxel intersection detection based on the trilobal segmentation results. It searches for intersection voxels that are simultaneously adjacent to the third leaf in the boundary voxel set of adjacent leaves and automatically extracts three trilobal boundary points to provide a benchmark for subsequent coordinate system construction. The mating region extraction module is used to calculate the symmetrical Euclidean distance field between the surfaces of two adjacent blades for three pairs of adjacent blades, automatically divide the mating contact area and the mating defect area based on the distance threshold, and extract the closed boundary curve of the defect area. The module for quantifying closure defects is used to calculate the surface area of ​​the defect area and contact area of ​​each adjacent leaflet pair using the triangular mesh surface accumulation addition method. A cylindrical coordinate system is constructed based on the valve annulus center and the trilobal junction point. The weighted centroid of the defect area area is mapped to the coordinate system, and the defect location label, the defect area of ​​each adjacent leaflet pair, the total defect area, and the defect rate index are output. The 3D visualization and report output module is used to generate a 3D rendering of the trilobed surface and annotations of the trilobed junction points. It annotates the mating contact area and defect area in the form of a heat map, and automatically generates a structured numerical report containing the defect area, contact area, total defect area, defect rate, defect location label, and defect centroid coordinates of each adjacent lobe pair.

[0109] This system can automatically execute the entire process from input of cardiac CT images to quantitative analysis and visualization output of occlusal defects without manual intervention, improving the accuracy and stability of tricuspid valve occlusal status assessment and providing reliable imaging support for TEER-TV clamping point planning, TTVR replacement decision-making, and accurate assessment of tricuspid regurgitation.

Claims

1. An artificial intelligence-based three-dimensional automatic analysis of tricuspid tri-leaflet coaptation status and coaptation defect quantification method, characterized in that, The steps include: Step 1: Acquire and preprocess three-dimensional medical image data. Use a three-dimensional medical imaging device to acquire cardiac images of the patient, extract target temporal data with clear tricuspid valve closure and occlusion, preferably end-systolic phase data, and preprocess it. Step 2: Tricuspid valve three-leaflet independent segmentation, based on a deep learning segmentation model, automatically segmenting the tricuspid valve region to obtain three independent curved surface structures of the anterior leaflet, the posterior leaflet and the septal leaflet, and optimizing the segmentation results through prior morphological constraints and morphological post-processing, and outputting the anterior leaflet curved surface , the posterior leaflet curved surface and the septal leaflet curved surface ; Step 3: Automatic extraction of trifoliate junction points, based on the trifoliate segmentation result of Step 2, three junction points are automatically extracted at the intersection of the segmentation boundaries of adjacent leaflets, which are the anterior-septal junction point , the anterior-posterior junction point , and the posterior-septal junction point ; Step 4: Automatic extraction of the mating region. For each of the three pairs of adjacent leaflets, the distance field between the two leaflet surfaces is calculated, and the corresponding regions are divided into mating contact areas and mating defect areas based on the distance threshold. Step 5: Quantitative analysis of occlusive defects, calculate the surface area of ​​the corresponding occlusive defect region for each adjacent leaflet pair, and map the defect region to the trilobal spatial coordinate system with the valve annulus as the reference, outputting the defect area and defect location; Step 6: 3D visualization and result output. Generate a 3D rendering of the tricuspid valve trilobal surface, annotate the occlusal contact area and occlusal defect area in the form of a heat map, and output a structured result report.

2. The method according to claim 1, characterized in that, In step 1, the three-dimensional medical imaging equipment includes cardiac CT, cardiac MRI or three-dimensional echocardiography equipment, and the preprocessing includes noise removal, grayscale normalization and multi-phase registration of the image.

3. The method according to claim 1, characterized in that, In step 2, the independent segmentation of the three lobes adopts a three-dimensional segmentation network that integrates prior morphological constraints. The lobular plane is used as a geometric prior constraint to constrain the spatial distribution range of each lobe, and the three lobes are distinguished based on the morphological prior that the anterior lobe has a relatively large area, the septum has a relatively narrow area, and the posterior lobe is located between the two.

4. The method according to claim 3, characterized in that, An area ratio regularization term is introduced into the loss function of the three-dimensional segmentation network to ensure that the segmentation results satisfy the anatomical prior relationship that the segmentation area of ​​the front leaf is greater than that of the back leaf and that the segmentation area of ​​the back leaf is greater than that of the septum leaf.

5. The method according to claim 4, characterized in that, Let the areas of the front leaf, back leaf, and septum be respectively... Then the area ratio regularization term satisfy: ； in , These are the regularization weight coefficients; The total loss function of the segmentation network is defined as: ； in, For standard segmentation loss, For topological regularization terms, and To balance the weights; After segmentation, a surface reconstruction algorithm is used to convert the trifoliate voxel mask into a triangular mesh surface. , and .

6. The method according to claim 1, characterized in that, In step 3, intersection detection is performed on the boundary voxels of the trifoliate segmentation result. Specifically, in the boundary voxel set of any two-leaf segmentation mask, the Euclidean distance between each pair is calculated, and the voxel with the smallest distance that is also adjacent to the third leaf boundary is selected as the intersection point; thus, three intersection points are obtained. They are located at the junctions of the anterior leaf and the septum, the anterior leaf and the posterior leaf, and the posterior leaf and the septum, respectively.

7. The method according to claim 5, characterized in that, In step 4, three pairs of adjacent blades , , Calculate each leaf surface separately each vertex To adjacent leaf surfaces Shortest Euclidean distance: ； in, express Norm, which is the Euclidean length of a vector. curved surface A set of vertices; setting a distance threshold. If the distance is less than The area is defined as the contact zone. Distance greater than or equal to The area is defined as the conjoint defect area. They respectively satisfy: ； 。 8. The method according to claim 7, characterized in that, In step 5, for each adjacent leaflet pair corresponding to the defect area The area of ​​the defect is calculated using the triangular mesh facet accumulation addition method; let the set of faces within the defect area be... For any triangular facet Its area is: ； The defect area corresponding to the adjacent leaflet pair is: ； Similarly, for the set of facets in the contact area The areas are summed up using the same method to obtain the corresponding contact area of ​​the adjacent leaflet pairs. ; The total area of ​​the apical defect is the sum of the defect areas corresponding to the three adjacent leaflet pairs: ； in, , and These represent the defect areas corresponding to adjacent leaflet pairs in the anterior-posterior, anterior-septum, and posterior-septum sections, respectively. Similarly, the total area of ​​the contact zone The contact area patches are accumulated using the same method to obtain the following: ； in, These represent the contact areas of adjacent leaflet pairs in the anterior-posterior, anterior-septum, and posterior-septum sections, respectively.

9. The method according to claim 8, characterized in that, In step 5, the defective region is mapped to a cylindrical coordinate system constructed with the center of the lobular annulus as the origin and the trilobal junction as the reference direction; the area-weighted centroid of each defective region is defined as: ； in, Triangular facet The geometric center; Based on the azimuth position of the area-weighted centroid in the trilobal spatial coordinate system, the corresponding defect location label is output. The defect location label includes anterior-posterior boundary defect, anterior-separation boundary defect, and posterior-separation boundary defect.

10. The method according to claim 9, characterized in that, In step 5, the misalignment rate is defined based on the ratio of the total misalignment defect area to the total misalignment region area. 。 11. A three-dimensional automatic analysis system for tricuspid valve leaflet occlusion status and quantitative analysis of occlusion defects based on artificial intelligence, characterized in that, include: The image data acquisition and preprocessing module is used to acquire and preprocess three-dimensional medical image data, extract target phase data with clear tricuspid valve closure and occlusion, preferably end-systolic phase data, and perform noise removal, grayscale normalization and multi-phase registration to improve data quality. The trilobate independent segmentation module is used to independently segment the anterior, posterior and septal lobes of the tricuspid lobe based on a 3D deep learning segmentation model that incorporates prior morphological constraints. It combines area ratio constraints, topological continuity constraints and morphological post-processing to optimize the segmentation results and output trilobate independent triangular mesh surfaces. The boundary point extraction module is used to automatically extract three trifoliate boundary points based on the boundary voxel intersection detection of the trifoliate segmentation results; The mating region extraction module is used to calculate the Euclidean distance field between the curved surfaces of the two leaflets for three pairs of adjacent leaflets, and automatically divide the mating contact area and mating defect area based on a preset distance threshold. The quantification module for occlusal defects is used to calculate the surface area of ​​the corresponding occlusal defect region for each adjacent leaflet pair using the triangular mesh surface accumulation addition method, and to construct a cylindrical coordinate system based on the leaflet annulus and the trilobal junction point, mapping the centroid of the defect region to the coordinate system, and outputting the defect location label, defect area and defect rate index. The 3D visualization and report output module is used to generate a 3D rendering of the tricuspid valve leaflet surface, annotate the occlusal contact area and occlusal defect area in the form of a heat map, and automatically generate a structured result report containing the defect area corresponding to each adjacent leaflet pair, the total occlusal defect area, the defect rate, and defect location labels.

Images