A medical image-based automated analysis and visualization system for post-tavi

Abstract

The application discloses a kind of TAVI postoperative automatic analysis and visualization system based on medical image, including data import, automatic analysis, valve leaflet function evaluation, valve stent geometry evaluation, interactive alignment and report generation module.System imports time series medical image data covering at least one cardiac cycle, segments aorta, coronary artery, left ventricle, valve stent and calcified structure, extracts stent central axis and establishes stent local coordinate system, generates transverse, longitudinal, bottom, coronary artery opening and walsh sinus reconstruction plane.Based on multi-plane reconstruction display, the system evaluates HALT, SFD and PFD, and calculates perimeter, area, long-short diameter, AED and PED and other parameters at stent inflow end, valve leaflet nadir and commissure height plane.Interactive alignment module maps valve structure to standard observation plane and superimposes simulated valve model, and report generation module integrates results to output structured report.

Description

Technical Field

[0001] This invention relates to the field of medical image processing and postoperative monitoring technology for cardiovascular diseases, specifically to an automated analysis and visualization system based on medical images for prosthetic-associated subclinical thrombootic events (PASTE) after transcatheter aortic valve implantation / replacement (TAVI / TAVR). The core of this invention focuses on low-attenuation leaflet thickening (HALT) in PASTE and integrates standardized measurement and report generation functions for sinus filling defect (SFD), prosthesis filling defect (PFD), and reduced leaflet motion (RELM). Technical Background

[0002] Transcatheter aortic valve replacement (TAVI / TAVR) has expanded from high-risk elderly patients to younger, low-surgical-risk patients, and the durability and long-term prognosis of bioprosthetic valves have gradually become a focus of clinical attention. Subclinical leaflet thrombosis (SLT) is one of the important potential causes of bioprosthetic valve failure: it can restrict leaflet movement, reduce the effective orifice area (EOA), and increase the mean transaortic gradient (MAG), which can eventually progress to clinical thrombosis, valvular failure, and even major adverse cardiovascular events such as stroke and readmission.

[0003] Current follow-up assessments primarily rely on manual image interpretation, requiring the simultaneous evaluation of four typical imaging features: low-attenuation leaflet thickening (HALT), sinus filling defect (SFD), prosthetic filling defect (PFD), and restricted leaflet movement (RELM). The existing manual assessment process suffers from problems such as being time-consuming, highly subjective, and lacking repeatability, thus becoming a bottleneck in clinical follow-up.

[0004] Low-attenuation leaflet thickening (HALT) refers to the thickening of artificial valve leaflets observed by cardiac CTA. It appears as low density, suggesting thrombosis on or inside the leaflet surface. The low-attenuation area needs to be measured in a diastolic transverse view and a score of 0–3 should be given.

[0005] Sinus filling defect (SFD) refers to a low-density filling defect in the aortic sinus. It requires locating three aortic sinuses on short-axis / long-axis MPR, scoring them according to a grade of 0–2 based on the number of affected sinuses, and noting whether non-coronary sinuses are involved (65.2% of cases occur here). Flow artifacts often cause blurred boundaries, leading to significant errors in manual localization.

[0006] Prosthetic filling defects (PFDs) refer to low-density filling defects within the prosthetic valve stent structure, indicating intrastent thrombosis or fibrosis. PFD assessment requires identifying low-attenuation defects within the prosthetic metal frame and differentiating between thrombosis and fibrosis. The 5-year incidence is as high as 73.8%, and small defects with ring-shaped or eccentric distributions are easily masked by metal artifacts; moreover, current methods do not combine PFD with HALT / SFD / RELM, making it difficult to complete the overall risk stratification of PASTE.

[0007] Current PACS and post-processing workstations lack a one-stop module for "prosthesis-related subclinical thrombotic events (PASTE)". On the one hand, they cannot automatically locate the standard MPR plane and cannot simultaneously extract HALT, SFD, PFD, and RELM parameters; on the other hand, they cannot generate reports separately according to the formats of core lab studies and clinical follow-up studies. Physicians often need to switch repeatedly between multiple software programs, which is inefficient, error-prone, and makes it difficult to intuitively display the panoramic process of PASTE evolution over time.

[0008] With the surge in TAVI patients and the extension of follow-up periods to over 5 years, manual assessment can no longer meet the clinical needs for "precision, efficiency, and comprehensive coverage." Developing an intelligent system that can automatically locate the measurement plane (allowing physicians to make fine adjustments), simultaneously extract the four major parameters according to standards, and generate structured reports has become an urgent priority to overcome the bottleneck of post-TAVI PASTE monitoring and improve long-term patient outcomes. Summary of the Invention

[0009] This invention combines artificial intelligence technology and image processing algorithms to provide an automated analysis and visualization system for TAVI surgery based on medical images, comprising the following modules: Module 1: Data Import Module, used to import medical imaging data covering at least one cardiac cycle; Module 2: Automated analysis module, used to automatically extract the aortic valve stent region based on the medical image data, establish a local coordinate system for the stent, and generate one or more reconstruction planes in the local coordinate system for the stent; Module 3: Leaflet function assessment module, used to automatically analyze leaflet function or stent filling based on the reconstruction plane, and obtain one or more postoperative assessment results; Module 4: Valve stent geometry assessment module, used to calculate one or more geometric parameters related to the aortic stent morphology on the reconstruction plane; Module 5: Interactive alignment module, used to spatially align the valve structure with one or more standard observation planes based on the reconstruction plane and the stent local coordinate system, so as to realize the visualization and positional correspondence of the valve leaflet or stent internal structure; Module 6: Postoperative report generation module, used to generate a visual interface and structured postoperative analysis report based on the evaluation results and the geometric parameters.

[0010] Furthermore, in module 1, the data import module includes: Module 1.1: Data Acquisition and Input Time-series medical imaging data of patients after TAVI surgery were acquired. This time-series medical imaging data includes CTA imaging data and conventional CT imaging data, and meets the following conditions: Time dimension: Covering at least one complete cardiac motion cycle, including clearly defined image frames corresponding to the systolic phase (left ventricular ejection phase, aortic valve fully open) and diastolic phase (left ventricular filling phase, aortic valve fully closed), with an interval of ≤50ms between adjacent image frames; Spatial dimensions: slice thickness ≤1mm, matrix size ≥512×512, ensuring clear display of aortic stents, calcified structures and valve leaflet details; Additional information input: Input metadata corresponding to the imaging data, including aortic stent type, follow-up time, and assessment status.

[0011] Phase markers for module 1.2: Users mark the corresponding image frames for systole and diastole using the system. The specific process is as follows: After loading the time-series image sequence, the system simultaneously displays axial, coronal, and sagittal images to allow users to identify the phases. The characteristics of the systolic image frame are: the aortic valve leaflets are fully open, the left ventricular cavity volume is at its minimum, and the ascending aorta is relatively well-filled with contrast agent; the characteristics of the diastolic image frame are: the aortic valve leaflets are fully closed, the left ventricular cavity volume is at its maximum, and the valve annulus plane is relatively clearly visible. Users mark the confirmed systolic / diastolic image frames, and the marking information includes at least the frame number, the cardiac cycle phase (%), and the marking timestamp. The marked data is automatically associated with the subsequent fully automated analysis module.

[0012] Furthermore, in module 2, the fully automated analysis module includes the following processing: segmenting the medical images corresponding to the systolic and diastolic phases imported from module 1 to obtain segmentation results for the target structures. The segmentation process is implemented using traditional image segmentation algorithms, or automatically implemented using deep learning algorithms, such as nnU-Net network structures. Through this segmentation process, mask information for the aorta, coronary arteries, left ventricle, aortic stent, and calcification structures is obtained.

[0013] Based on the mask information, the aortic stent region is extracted using traditional image processing algorithms, and principal component analysis (PCA) or least squares fitting is performed on the aortic stent region to obtain the stent's central axis. This central axis is defined as the major axis reference axis of the aortic stent (along the aortic blood flow direction). A local coordinate system for the stent is established based on this major axis reference axis, and one or more reconstruction planes are generated within this local coordinate system. These reconstruction planes include a transverse plane, a longitudinal plane, a top plane, a bottom plane, a coronary ostium plane, and a Warburg sinus (SOV) plane.

[0014] Furthermore, in module 3, multiplanar reconstruction (MPR) is performed on one or more reconstruction planes generated by module 2 to display the reconstruction and to assess leaflet function. This assessment mainly includes low-attenuation leaflet thickening (HALT) assessment, intrasinusoidal low-density filling defect (SFD) assessment, and prosthesis filling defect (PFD) assessment. The specific process is as follows: Module 3.1 HALT Assessment: The user selects images from at least two phases as the analysis object. Based on the bottom plane of the stent obtained in Module 2, the system translates it along the normal direction of the plane according to the preset offset parameters corresponding to different stent models to obtain the analysis plane for HALT observation.

[0015] Let the equation of the bottom plane of the support be:

[0016] The equation for the moving plane is:

[0017] Where t is the distance traveled. Users can switch between different phase images and fine-tune the plane vertically to observe the presence of HALT. The system records the HALT level and marks its location.

[0018] Module 3.2 SFD Evaluation: The system guides the display interface to the Warburg sinus (SOV) plane obtained in Module 2 and performs SFD discrimination based on the SOV plane. The SOV plane is input to the trained multi-class discrimination model, which classifies it using GoogLeNet, ResNet, or other convolutional neural network structures; the output of the fully connected layer is used to calculate the probability using the softmax function to obtain the probability vector. , representing the probability of SFD versus non-SFD:

[0019] in, , represent the probabilities of SFD existing and SFD not existing, respectively. , These are the model parameters. The system can combine the model's predictions with the user's manual judgment to record the location of the affected leaflet.

[0020] Module 3.3 PFD Assessment: The system positions the analysis interface to the reconstruction plane corresponding to the inflow end of the aortic stent and simultaneously displays MPR images in the long and short axis directions to intuitively determine the presence of PFD; the system provides a manual measurement tool for users to measure the thickness of suspected PFD areas and record the measurement results.

[0021] Furthermore, in module 4, the valve stent geometry evaluation module is used to perform morphological analysis of the aortic stent; the module performs measurements at the stent inflow plane, the plane corresponding to the lowest point of the leaflet (Nadir plane), and the plane corresponding to the commissure height (commissure plane). The inflow plane, Nadir, and commissure planes are reconstructed planes at fixed offset positions relative to the stent bottom plane.

[0022] The module's measurement parameters include geometric features such as the perimeter, area, longest diameter, shortest diameter, average perimeter diameter, and average area diameter after aortic stent implantation. To obtain these geometric parameters, the system resamples or normalizes the three-dimensional medical image and extracts a two-dimensional cross-section on the target reconstruction plane passing through the stent. Subsequently, the perimeter and area are calculated based on the contour point set of the two-dimensional cross-section, and the maximum convex hull is extracted from the point set. The specific steps are as follows: Module 4.1 Two-Dimensional Section Extraction: For the aortic stent region, the equation of the target plane passing through the stent structure is calculated, so that the target plane intersects with the three-dimensional image volume data to form a two-dimensional section to be extracted. For any pixel on the two-dimensional section, its horizontal and vertical coordinates in the two-dimensional coordinate system are used as input. Based on the geometric mapping relationship between the target plane and the three-dimensional volume coordinate system, the corresponding depth value of the pixel in the depth direction of the volume data is calculated. If the depth value is not an integer, the volume data is interpolated. Then, the two-dimensional section image of the target plane passing through the three-dimensional segmented image is calculated.

[0023] Module 4.2: Convex Hull Point Set Acquisition: Obtain the coordinates of all non-zero pixels in a 2D image and construct the point set using the Graham scan algorithm. P = {p1, p2, ..., p n}, where p = (x , y ), i ∈ [1, n] The convex hull calculation module is as follows: A starting point s is selected, where the starting point is the point with the smallest x-coordinate and the smallest y-coordinate in the point set, i.e. The coordinates are To ensure the starting point of the convex hull is located at the lower left corner of the point set; using the starting point s as the reference origin, calculate the polar angle of each of the other points relative to s. The polar angle and Sort all points in ascending order of polar angle to obtain the sorted point set: p sort = {p (1) , p (2) , ..., p (m)}, Where, p (1) Let p be the starting point s. (2) to p (m) Sort by polar angle from smallest to largest, m≤n; The core logic of filtering convex hull vertices using a stack structure is to determine the direction of three consecutive points: left-turning points (points with a cross product greater than 0) are retained; right-turning points (points with a cross product less than or equal to 0) are removed. The specific process is as follows: Initialize stack H: sort the first 3 points, p (1) ,p (2) ,p (3) Push onto the stack (if m < 3, directly return psort as the convex hull). Traverse the remaining points p (k) k≥4: Let the two points located at the top H of the stack be top1 = H[-1] (top of the stack) and top2 = H[-2] (the next element after the top of the stack). Calculate the vector:

[0024]

[0025]

[0026] in, The directed line segment from top1 to top2 Let p(k) be the directed line segment from p(k) to top1, where cross is the cross product of vectors. If cross ≤ 0, pop the top of the stack (top1) and repeat the process until cross > 0; if cross > 0, then pop the current p. (k) Push H onto the stack. (Straight p) sort End of p sort All points in the middle.

[0027] Module 4.3 Calculation of Planar Related Parameters: Assuming the number of points in stack H is n, calculate its position in a 1*1*1 mm plane. 3 The formula for calculating the perimeter L at the given resolution is as follows: +dis( )

[0028] in Let S be the distance between two points on H. The area S is calculated using the shoelace method; for each point on H, the area S is...

[0029] The relevant radial parameters are derived from the perimeter and area: ,

[0030] PED is the perimeter-derived diameter, and AED is the area-derived diameter. L represents the perimeter, and S represents the area. Calculation of major and minor diameter parameters:

[0031]

[0032] q is the point set corresponding to H, where Let these be the coordinates of the midpoint. Let H be the two principal axes of the set of points in plane H, and let H be a 2x2 matrix. ].

[0033] ,

[0034] ,

[0035]

[0036]

[0037] In the principal direction coordinate system, find the first principal direction ( The leftmost and rightmost points in the direction) and in the second main direction ( direction) the lowest and uppermost points ; ,in For the major axis, For the minor axis, The average diameter is 1.

[0038] Module 5: Handover Alignment Module: During HALT observation, the system needs to perform evaluation in the coronal plane. Since the actual plane where the implanted bioprosthetic valve is located is usually not completely consistent with the coronal observation plane, the system simulates the position of the valve in the coronal plane through image processing. Specifically, the system obtains the maximum convex hull of the two-dimensional plane of the stent according to the two-dimensional cross-section extraction method described in Module 4, and calculates the center point of the convex hull. The simulated bioprosthetic valve is covered at the center point, and rotation of the valve model is supported.

[0039] Module 6: Report Generation Module: The report generation module is configured to automatically call the built-in template engine after modules 1 to 4 have completed image acquisition, stent area identification, postoperative evaluation result generation and geometric parameter extraction, to integrate multiple analysis results and generate a structured postoperative analysis report.

[0040] The report includes the following: (1) Basic information: patient identification, gender, age, and scan date; (2) Scanning protocol information: Scanning protocol parameters including CT equipment model, tube voltage, tube current, slice thickness, and reconstruction core; (3) Stent information: The brand, model, nominal diameter, length, implantation site and orientation information of the stent obtained from the parsing in module 1; (4) Assessment results and imaging evidence: HALT score table output by module 2 and its corresponding axial, sagittal and coronal maximum density projection screenshots; SFD score table output by module 3 and screenshots of stent inlet, outlet and maximum deformation angle; PFD score table output by module 4 and screenshots of maximum displacement of valve frame. (5) Geometric parameters: Minimum area, maximum area, average area, perimeter, ellipticity index, AED, PED, major diameter, minor diameter, major-minor diameter ratio and symmetry index of the stent cavity calculated by module 4.

[0041] In addition, the report includes individualized anticoagulation regimen recommendations based on the built-in decision tree rule base, which includes recommendations on anticoagulant drug type, dosage, and duration of treatment, as well as recommendations on follow-up intervals based on a risk stratification model.

[0042] The report can be exported in PDF and / or HTML formats. The PDF version can use an archive-compatible format (e.g., PDF / A standard) and embed TrueType fonts, employing lossless JPEG2000 image compression and a 128-bit AES encrypted digital signature to ensure long-term readability and tamper-proof performance. The HTML version uses a responsive Bootstrap 5 framework, SVG vector graphics, DataTables dynamic tables, and MathJax 3 formula rendering technology to support cross-platform browsing and real-time retrieval. The report's homepage automatically generates a QR code and SHA-256 checksum for quick access and integrity verification on mobile devices. Through these methods, a closed-loop document output of clinical data, imaging evidence, quantitative indicators, and follow-up guidance is achieved simultaneously with the completion of post-transcatheter aortic valve surgery follow-up assessment.

[0043] Technical effect: This invention integrates artificial intelligence and image processing technologies to construct an automated analysis and visualization system after TAVI surgery, achieving the following technical effects: 1. Full-process automation The entire process, from importing time-series image data to generating structured reports, is fully automated, reducing manual intervention and improving analysis efficiency. 2. Multi-dimensional evaluation and dynamic functional analysis: By using systolic / diastolic markers and multiphase analysis, abnormalities in valve dynamic function (such as HALT, SFD, PFD) can be accurately identified. 3. Geometric morphology quantification: Standardize the measurement of key planes such as the inflow plane of the stent and the lowest point of the leaflet, and their morphological parameters (circumference, area, major and minor diameters, etc.).

[0044] 4. Synergistic analysis of calcification and scaffold: By combining deep learning segmentation and PCA fitting, stent axis extraction and calcification localization can be achieved, supporting coronary artery ostium risk assessment.

[0045] 5. Structured output: Automatically generate structured reports that include image evidence and quantitative indicators. Attached image description: Figure 1 This is a flowchart illustrating the overall system flow of the present invention. Figure 2 Flowchart for the automated analysis module; Figure 3 Flowchart for the leaflet function assessment module; Figure 4 Flowchart for the valve stent geometry evaluation module; Figure 5 This is a schematic diagram of the scaffold segmentation and related reconstruction plan; Figure 6 This is a schematic diagram for evaluating leaflet thickening in low-attenuation HALT. Figure 7 Schematic diagram for assessing low-density filling defects in the SFD sinus; Figure 8 A schematic diagram for assessing filling defects in PFD prostheses; Figure 9 A schematic diagram illustrating the calculation of plane parameters at the inflow end; Figure 10 This is a diagram illustrating the reporting indicators. Detailed Implementation

[0046] 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, after reading the disclosure of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope of protection defined by this invention.

[0047] like Figure 1 As shown, this invention proposes an automated analysis and visualization system for post-TAVI surgery based on medical images, comprising the following modules: Module 1: Data Import Module. This module reads image sequences through the "Data Import Module" to realize the input of relevant data information and phase marking.

[0048] Module 1.1: DICOM metadata parsing. The system reads the image sequence through the "Data Import Module" and performs DICOM Tag parsing in the background, automatically extracting fields such as: Patient ID (unique patient identifier), StudyDate (examination date), SeriesDescription (sequence description, used to distinguish different scan phases), PixelSpacing (pixel spacing, used for in-plane size calibration), and SliceThickness (slice thickness, used for axial size calibration).

[0049] Module 1.2: Stent Model Parametric Input. The system prompts the user to enter or select the model of the implanted stent from a preset list (e.g., Medtronic Evolut, Edwards SAPIEN 3, Boston Scientific Acurate, etc.). This information is crucial, as it will serve as the default setting for the subsequent annular plane displacement t (e.g., different stent designs may correspond to different leaflet function assessment plane heights) and as baseline data for follow-up comparisons.

[0050] Module 1.3: Users can mark systolic and diastolic image frames through the system. After loading the 4D image sequence, the system simultaneously displays axial, coronal, and sagittal three-plane images for user identification.

[0051] Systolic frame: The aortic valve leaflets are fully open and the left ventricular cavity volume is at its minimum.

[0052] Diastolic frame: The aortic valve leaflets are completely closed and the annular plane is clearly visible.

[0053] The tagging information includes at least the frame number, cardiac cycle phase (%), and tagging timestamp, and is automatically linked to the subsequent fully automated analysis module.

[0054] Module 2: Fully automated analysis module, which processes images to achieve 3D medical image recognition and key plane parameter localization, such as... Figure 2 As shown.

[0055] Module 2.1 Segmentation Network: The nnunetv2 pre-trained model was used, with a training set of 123 cases of post-TAVI CTA. Five structures were manually labeled: aorta, stent frame, calcification, left coronary artery, right coronary artery, and left ventricle. The output was a probability map with a threshold of 0.5, yielding binary masks for each structure. The Dice coefficients (validation set n=30) were: aorta 0.96, stent frame 0.94, left coronary artery 0.92, right coronary artery 0.93, and left ventricle 0.92, meeting clinically acceptable accuracy.

[0056] Module 2.2 Planar Localization and Extraction: Based on the aortic stent segmentation results obtained in Module 2.1, a local coordinate system is established using Principal Component Analysis (PCA). Using the center point of the stent point cloud as the starting point of the annular plane, and based on the lowest point of the stent, the left and right coronary artery ostia, and the highest point of the stent, the system moves along the normal vector with a step size of 1 mm, generating 60 cross-sections within a range of 0-60 mm for subsequent HALT, PFD, and geometric parameter measurements. Some of these planes are shown below. Figure 5 As shown.

[0057] For fitting the aortic stent, the stent segmentation results are fitted using the PCA algorithm to obtain the long axis, which represents the direction of aortic blood flow. The calculation method is as follows: Calculate the coordinates of the center point of the support: Calculate all points on the valve Define a decentralized data matrix Its size is , where n is the number of support point cloud points.

[0058] ] Calculate the covariance matrix and eigenvectors:

[0059] in, for The matrix formed by these elements is a (3, n) matrix. C is the covariance matrix. For finding eigenvalues ​​and eigenvectors of a matrix:

[0060] The eigenvalues ​​are: For eigenvalues, These are the eigenvectors.

[0061]

[0062] Select That is, the direction with the greatest difference above the aortic stent. The corresponding normalized vector, taken as the major axis of the support, i.e., the normal vector of the plane to be calculated later, is denoted as [a, b, c]. T Then the equation of the plane passing through the center point is:

[0063] Where a, b, and c are the normalized normal vectors obtained. , This is the center point of the support point cloud structure.

[0064] Module 2.3 Calculating the Coronary Plane: Calculating the point set P of the aorta and left coronary artery using image processing. L The center point is denoted as (x l ,y l ,z l ), and the point set P of the artery and the left coronary artery. R The center point is denoted as (x r ,y r ,z r ), calculate the coordinates of the center points of the left and right coronary arteries (x mid ,y mid ,z mid ), then after (x mid ,y mid ,z midThe equation of the plane at point is:

[0065] Furthermore, the top plane and bottom plane of the support can be calculated in this way.

[0066] Module 2.4 calculates the inflow plane, Nadir (leaf lowest point plane), and commitment (joint height / joint column plane) as positional parameters that are fixed relative to the bottom plane. The calculation method is as follows:

[0067] Where (a,b,c,d) are the coefficients of the bottom plane equation of the lobe ring, and t is the displacement.

[0068] Module 3: Leaflet Function Assessment Module, module flowchart as follows Figure 3 As shown; the system loads the segmented view by default, along with the corresponding transverse, sagittal, and coronal views of the CT image, as shown. Figure 6-8 As shown; Module 3.1 uses the annular bottom plane obtained from Module 2 and translates it upwards by a fixed distance t according to the stent model parameters preset in Module 1 to obtain the analysis plane for HALT observation. Users can switch between diastolic and systolic periods to observe images of the same location. The plane can be finely adjusted vertically using the system's built-in MPR tool. The presence of a halt can be determined using the major and minor axis planes. Users can observe the location and severity of the halt, such as... Figure 6 As shown.

[0069] Module 3.2 SFD Evaluation: The system jumps to the SOV plane, that is, by calculating the plane, the plane with the largest area passing through the bottom of the aorta is obtained. Deep learning-assisted classification is used, loading the ResNet50 classification model, and outputting p. sfd If p sfd A value >0.5 indicates the presence of SFD. The system combines the model prediction with the user's manual judgment, records the affected sinus (left / right / none) and its number (grade 0-2), and allows for manual correction. Figure 7 As shown.

[0070] Module 3.3 PFD Assessment: The system locates the outflow plane. Users can use the system's built-in MPR tool to fine-tune the plane vertically. Users can determine the presence of PFD by observing the information of the three axes, and can also use the measurement tool to record the thickness.

[0071] Module 4: Valve Stent Geometry Assessment; This module performs morphological analysis of aortic stents, accurately calculates the geometric parameters of key sections, and calculates the area, perimeter, area-derived diameter, and perimeter-derived diameter of three points on the aortic stent: the inflow plane, the nadir (leaflet lowest point plane), and the commisure (commissure height / joint column plane). This allows for the assessment of the geometric morphology of the aortic stent after implantation. Figure 9 As shown.

[0072] Module 4.1 For the aortic stent region, calculate the equation of the plane passing through the stent structure, so that the plane intersects with the 3D image volume data to form a 2D cross section to be extracted. For any pixel on the 2D cross section, use its horizontal and vertical coordinates in the 2D coordinate system as input, and calculate the corresponding depth value of the pixel in the depth direction of the volume data according to the geometric relationship of the spatial plane; if the depth value is not an integer, interpolation is performed; then calculate the 2D cross section image of the plane passing through the 3D segmented image. The specific method is as follows: Given the plane equation: ax + by + cz = 0, given x and y, find the value of z, record the hu value at (x, y, z), iterate through all x and y values, and then proceed to the two-dimensional plane, where x ≤ y≤ , z≤ .

[0073] Module 4.2 Convex Hull Extraction: To calculate specific parameters, it is necessary to calculate the area, perimeter, area-derived diameter, and perimeter-derived diameter of the intersection between the plane and the support. The support point cloud is extracted, and the plane intersecting the point cloud is calculated. The calculation method in the 2D image is as follows: The plane intersects with the point cloud. The coordinates of all non-zero pixels in the 2D image are obtained, and the Graham scan algorithm is used to construct the point set. P = {p1, p2, ..., p n}, where p = (x , y ), i ∈ [1, n] The convex hull calculation module is as follows: A starting point s is selected, where the starting point is the point with the smallest x-coordinate and the smallest y-coordinate in the point set, i.e. The coordinates are To ensure the starting point of the convex hull is located at the lower left corner of the point set; using the starting point s as the reference origin, calculate the polar angle of each of the other points relative to s. The polar angle and Sort all points in ascending order of polar angle to obtain the sorted point set: psort = {p (1) , p (2) , ..., p (m)}, Where, p (1) Let p be the starting point s. (2) to p (m) Sort by polar angle from smallest to largest, m≤n; The core logic of filtering convex hull vertices using a stack structure is to determine the direction of three consecutive points (left turn means calculating the cross product > 0 and keeping them; right turn means calculating the cross product ≤ 0 and discarding them). The specific process is as follows: Initialize stack H: Push the first 3 sorted points, p(1), p(2), p(3), onto the stack (if m < 3, directly return psort as the convex hull). Traverse the remaining points p (k) k≥4: Let the two points located at the top H of the stack be top1 = H[-1] (top of the stack) and top2 = H[-2] (the next element after the top of the stack). Calculate the vector:

[0074]

[0075]

[0076] in, The directed line segment from top1 to top2 Let p(k) be the directed line segment from p(k) to top1, where cross is the cross product of vectors. If cross ≤ 0, pop the top of the stack (top1) and repeat the process until cross > 0; if cross > 0, then pop the current p. (k) Push H onto the stack. (Straight p) sort End of p sort All points in the middle.

[0077] Module 4.3 obtains the convex hull of the corresponding plane. By standardizing the points on the convex hull, its coordinates in 1×1×1 mm are solved. 3 The coordinates below are approximately the points where the plane passes through the outermost ring of the support. Through standardized mapping of these coordinates, the corresponding cross-sectional parameters can be obtained, including: perimeter L and area S, respectively calculated using the formulas:

[0078]

[0079] Area equivalent diameter circumference equivalent diameter The major and minor paths are obtained through eigenvalue decomposition of the covariance matrix.

[0080] The relevant radial parameters are derived from the perimeter and area: ,

[0081] PED is the perimeter-derived diameter, and AED is the area-derived diameter. L represents the perimeter, and S represents the area. Calculation of major and minor diameter parameters:

[0082]

[0083] q is the point set corresponding to H, where Let these be the coordinates of the midpoint. Let H be the two principal axes of the set of points in plane H, and let H be a 2x2 matrix. ].

[0084] ,

[0085] ,

[0086]

[0087]

[0088] In the principal direction coordinate system, find the first principal direction ( The leftmost and rightmost points in the direction) and in the second main direction ( direction) the lowest and uppermost points ; ,in For the major axis, For the minor axis, The average diameter is 1.

[0089] The relevant parameters of the three planes—inflow plane, Nadir (leaf lowest point plane), and commisure (joining height / joining column plane)—are calculated sequentially using the above method, and the corresponding cross-sectional parameters are displayed by locating the plane.

[0090] Module 5, the junction alignment module, generates a virtual three-leaflet model (120° symmetry) with the stent center as the origin on the coronal reference plane. The user rotates the model by a slider θ∈[-180°, 180°] to simulate the bioprosthetic valve's position within the patient's body. An angle measurement tool calculates the angles between the bioprosthetic valve and the junctions of the left and right coronary arteries. The convex hull is obtained through calculation module 4.3, and then the center point c(x) of the convex hull is determined.c y c The diameter of the simulated valve is obtained by combining the diameter parameters from module 4 with the diameter parameters from module 4. The equation for rotation is as follows:

[0091]

[0092]

[0093] Where x and y are points on the original simulated valve, To simulate valve rotation The coordinate position after the angle, where θ is the angle in radians.

[0094] Module 6 Postoperative Report Generation: This module is configured to automatically call the built-in template engine after all analyses are completed in Modules 1 through 5, integrating multiple analysis results and writing them into a unified structured postoperative analysis report, such as... Figure 10 As shown. The system generates structured PDF and HTML reports with one click, including: patient information, scan parameters, stent type; HALT, SFD, PFD scoring tables and key screenshots; geometric parameter tables (area, perimeter, AED, PED, major and minor axes); follow-up recommendations (anticoagulation regimen, re-examination interval), and takes safety measures for the report.

[0095] The PDF version adheres to the PDF / A-2b archiving standard, ensuring font compatibility through embedded TrueType fonts. It employs JPEG2000 lossless compression technology to preserve image details and combines 128-bit AES encryption and digital signature mechanisms to achieve long-term readability and tamper-proof verification. The HTML version is built on the responsive Bootstrap 5 framework, integrating SVG vector graphics to ensure cross-device display accuracy. It uses the DataTables component for dynamic data table interaction and the MathJax3 engine for accurate formula rendering, adapting to multi-terminal browsing and real-time data retrieval. The report's first page automatically generates a unique QR code and SHA-256 checksum. The QR code allows for quick access on mobile devices, while the SHA-256 checksum uses a hash algorithm to encrypt the report data, verifying data integrity. This solution integrates clinical data, imaging evidence, quantitative indicators, and follow-up guidance through these technical means, forming a closed-loop document output system that ensures the standardization, security, and ease of use of postoperative follow-up assessments.

Claims

1. An automated analysis and visualization system for post-TAVI surgery based on medical images, characterized in that, Includes a data import module for importing medical imaging data covering at least one cardiac cycle; An automated analysis module is used to automatically extract the aortic valve stent region based on the medical imaging data, establish a local coordinate system for the stent, and generate one or more reconstruction planes in the local coordinate system for the stent; a leaflet function assessment module is used to automatically analyze the leaflet function and / or in-stent filling status based on the one or more reconstruction planes to obtain one or more postoperative assessment results; a valve stent geometry assessment module is used to calculate one or more geometric parameters related to the morphology of the aortic valve stent on the one or more reconstruction planes. An interactive alignment module is used to spatially align the valve structure with one or more standard observation planes based on the one or more reconstruction planes and the local coordinate system of the stent, so as to realize the visualization and positional correspondence of the valve leaflet or stent structure; and a postoperative report generation module is used to generate a visualization interface and a structured postoperative analysis report based on the one or more postoperative evaluation results and the one or more geometric parameters.

2. The system according to claim 1, characterized in that, The data import module includes a data acquisition and input unit and a phase marking unit. The data acquisition and input unit is used to acquire time-series medical image data of patients after TAVI and input metadata corresponding to the time-series medical image data. The time-series medical image data includes CTA image data and / or conventional CT image data. The metadata includes aortic valve stent type, follow-up time and / or evaluation status. The phase marking unit is used to mark the image frames corresponding to the systolic and diastolic phases when the system simultaneously displays axial, coronal and sagittal three-plane images. The marking information includes frame number, cardiac cycle phase and / or marking timestamp.

3. The system according to claim 1, characterized in that, The automated analysis module includes: segmenting the medical images corresponding to systole and diastole imported by the data import module to obtain mask information of the aorta, coronary arteries, left ventricle, aortic valve stent, and calcification structures, wherein the segmentation is performed using a traditional image segmentation algorithm or a deep learning-based segmentation network; extracting the aortic valve stent region based on the mask information, and performing principal component analysis and / or least squares fitting on the aortic valve stent region to obtain the stent central axis and using the stent central axis as the reference axis of the stent's long axis; and establishing a stent local coordinate system based on the stent's long axis reference axis, and generating one or more reconstruction planes in the stent local coordinate system, wherein the reconstruction planes include a transverse plane, a longitudinal plane, a top plane, a bottom plane, a coronary ostium plane, and / or a Valsalva sinus plane.

4. The system according to claim 1, characterized in that, The leaflet function assessment module is configured to perform multi-plane reconstruction display on one or more reconstruction planes generated by the automated analysis module, and to perform leaflet function assessment based on the one or more reconstruction planes. The leaflet function assessment includes HALT assessment, SFD assessment and PFD assessment.

5. The system according to claim 4, characterized in that, The HALT evaluation includes: generating a HALT analysis plane by translating along the normal direction of the plane based on the bottom plane of the stent and according to the preset offset parameters corresponding to the stent model; and observing and annotating the HALT analysis plane based on images of at least two phases to output the HALT evaluation results.

6. The system according to claim 4, characterized in that, The SFD assessment includes: inputting the Valsalva sinus plane into a classification model to output the probability of SFD presence, and combining interactive confirmation to record the location information of the affected sinus.

7. The system according to claim 4, characterized in that, The PFD assessment includes: displaying multi-plane reconstructed images along the major and minor axes on the reconstruction plane corresponding to the stent inflow end, and measuring and recording the thickness of the suspected PFD area to output the PFD assessment results.

8. The system according to claim 1, characterized in that, The valve stent geometry evaluation module is configured to measure the aortic valve stent at the stent inlet plane, the plane corresponding to the lowest point of the leaflet, and the plane corresponding to the fusion height, wherein the stent inlet plane, the plane corresponding to the lowest point of the leaflet, and the plane corresponding to the fusion height are reconstructed planes at fixed offset positions relative to the stent bottom plane.

9. The system according to claim 8, characterized in that, The valve stent geometry evaluation module is configured to output one or more geometric parameters, including perimeter, area, longest diameter, shortest diameter, perimeter-derived diameter, and / or area-derived diameter.

10. The system according to claim 8, characterized in that, The valve stent geometry evaluation module is configured to resample and / or normalize three-dimensional medical image data and extract two-dimensional cross-sectional images on the target reconstruction plane passing through the stent.

11. The system according to claim 10, characterized in that, The valve stent geometry evaluation module is configured to obtain a set of stent contour points based on the two-dimensional cross-sectional image, and to calculate the perimeter and area based on the set of stent contour points, wherein the set of stent contour points can optionally be used to obtain an outer contour point set through convex hull processing.

12. The system according to claim 1, characterized in that, The interactive alignment module is configured to map the valve structure to the coronal standard observation plane based on the reconstruction plane and the stent local coordinate system, and to overlay and display the simulated valve model in the standard observation plane. It also supports rotation and / or translation of the simulated valve model to achieve positional correspondence and visualization.

13. The system according to claim 1, characterized in that, The postoperative report generation module is configured to call a template engine to integrate one or more postoperative assessment results, one or more geometric parameters, and metadata corresponding to medical imaging data to generate a structured postoperative analysis report, which is then output in PDF and / or HTML format.

Images