Acute pancreatitis CT image intelligent segmentation method and system

Abstract

The invention belongs to the technical field of image segmentation, and particularly relates to an intelligent segmentation method and system for an acute pancreatitis CT image, and the method comprises the steps: simulating a normal pancreatic CT image, carrying out the virtual lesion generation, fusing a virtual CT image set with a real CT image, extracting the topological structure features of the pancreatic tissue of the fused image, and carrying out the segmentation of the pancreatic tissue; performing unsupervised clustering on the topological structure image by adopting a topological perception diffusion coacervation method, and distinguishing pancreas and background regions; differentiated mechanical parameters are distributed to all areas of the preliminary segmentation mask, and boundary deformation is corrected; carrying out fine correction on the pixels of the optimized mask by adopting an entropy optimization threshold method; and according to the region segmentation mask, automatically calculating a necrotic volume ratio and an exudation maximum sectional area, and outputting a region segmentation result. According to the method, a virtual image is generated through fluid diffusion, topological guide segmentation and mechanical correction are carried out, accurate partitioning and quantitative grading are achieved, and clinical diagnosis is assisted.

Description

Technical Field

[0001] This invention belongs to the field of image segmentation and analysis technology, specifically to the intelligent segmentation method and system for CT images of acute pancreatitis. Background Technology

[0002] Intelligent segmentation of CT images for acute pancreatitis is a technology that combines medical imaging technology, computer vision, and artificial intelligence algorithms to automatically and precisely segment the abdominal enhanced CT images of patients with acute pancreatitis. Its core objective is to accurately identify pathological areas such as pancreatic parenchyma, edema areas, necrotic areas, peripancreatic exudate, and pseudocysts, providing quantitative basis for disease grading and treatment plan formulation.

[0003] Traditional CT image segmentation methods for acute pancreatitis have several drawbacks: First, they rely on manually labeled samples to train the model, resulting in high labeling costs and weak generalization ability, leading to a significant decrease in segmentation accuracy for rare, severe cases. Second, they depend solely on superficial features such as grayscale and texture, making them susceptible to interference from intestinal gas and blood vessels, and difficult to distinguish between similar grayscale areas such as edema and necrosis. Third, they fail to consider the tissue mechanical deformation caused by pancreatic inflammation, resulting in significant deviations between segmentation boundaries and the actual anatomical morphology. Fourth, they cannot accurately distinguish between pseudocysts and peripancreatic effusion, easily confusing the two types of pathological areas. Finally, the algorithms are complex and computationally intensive, making them difficult to deploy on clinical workstations for real-time segmentation. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, this invention proposes an intelligent segmentation method for CT images of acute pancreatitis. This invention primarily addresses the problem of significant deviations between the segmentation boundaries of traditional intelligent segmentation methods for acute pancreatitis CT images and the actual anatomical morphology.

[0005] The intelligent segmentation method for CT images of acute pancreatitis provided by the present invention includes: S1: generating a virtual lesion image set by simulating normal pancreatic CT images based on physical fluid diffusion, and outputting the virtual CT image set.

[0006] S2: The virtual CT image set is fused with the real CT image set. The topological features of pancreatic tissue in the fused image are extracted by discrete Morse theory, and the pancreatic topological image is output.

[0007] S3: Unsupervised clustering of the topological image is performed using topologically-aware diffusion-agglomeration to distinguish the pancreas from the background region. Labeled samples are introduced to construct a semi-supervised loss function, optimizing the cluster centers, further subdividing edematous, viable, and necrotic tissues, and outputting a preliminary segmentation mask.

[0008] S4: Construct a dynamic mechanical finite element analysis model, assign differentiated mechanical parameters to each region of the initial segmentation mask based on the measured data of elastic modulus and Poisson's ratio of pancreatic tissue at different disease stages, simulate tissue mechanical properties, correct boundary deformation, and output an optimized mask.

[0009] S5: The entropy value optimization threshold method is used to refine the pixels of the optimized mask, determine the boundary between pseudocysts and peripancreatic effusion, and output the region segmentation mask.

[0010] S6: Automatically calculate the proportion of necrotic volume and the maximum cross-sectional area of ​​exudation based on the region segmentation mask, and output the region segmentation results.

[0011] According to the intelligent segmentation method for CT images of acute pancreatitis provided by the present invention, the specific steps for outputting the virtual CT image set in step S1 are as follows:

[0012] S11: Obtain normal pancreatic CT images and corresponding anatomical structure annotations, extract basic features such as CT value distribution and contour morphology of pancreatic parenchyma, and output normal pancreatic image feature set.

[0013] S12: Based on the pathological evolution of acute pancreatitis, fluid diffusion parameters for edema, necrosis, and exudation are set. The normal pancreatic image feature set is used as the initial diffusion field to simulate the fluid diffusion process in the lesion area and output the lesion prototype image.

[0014] S13: Combining the disease course characteristics of mild and severe cases, adjust the time step and range of diffusion simulation, optimize the morphology of the lesion area in the lesion prototype image, make the lesion distribution conform to the clinical pathological characteristics, and output a virtual CT image set.

[0015] According to the intelligent segmentation method for CT images of acute pancreatitis provided by the present invention, the specific steps for outputting the pancreatic topological structure image in step S2 are as follows:

[0016] S21: Perform rigid registration between the virtual CT image set and the real acute pancreatitis CT images, align the anatomical location of the pancreas with the scanning parameters, eliminate image shifts caused by equipment differences, and output normalized CT images.

[0017] S22: Calculate the gradient value of the image gray field based on the normalized CT image, identify the gradient extrema and critical connection lines through discrete Morse theory, capture the connectivity and boundary continuity features of pancreatic tissue, and output the pancreatic topological feature point set.

[0018] S23: Construct the topological connection relationship between feature points based on the pancreatic topological feature point set, restore the spatial topological structure of the pancreatic parenchyma and surrounding tissues, and output the pancreatic topological structure image.

[0019] According to the intelligent segmentation method for CT images of acute pancreatitis provided by the present invention, the specific steps for outputting the pancreatic topological feature point set in step S22 are as follows:

[0020] The Sobel operator is used to calculate the gray-level gradient of normalized CT images to obtain the gradient magnitude and direction of each pixel, generating a gray-level gradient field image.

[0021] Based on the discrete Morse theory, gradient thresholds are set according to the grayscale gradient field image to filter gradient extrema points and output the initial set of extrema points.

[0022] By using the critical connection rule in discrete Morse theory, adjacent extreme points are connected to form critical edges, and extreme points and critical edges belonging only to the pancreatic region are selected, outputting a set of topological feature points.

[0023] According to the intelligent segmentation method for CT images of acute pancreatitis provided by the present invention, the specific steps for outputting the preliminary segmentation mask in step S3 are as follows:

[0024] S31: Calculate the topological similarity and spatial distance between pixels in the pancreatic topological structure image using the topological sensing diffusion agglomeration method, aggregate pixels according to the similarity threshold to complete unsupervised clustering, distinguish the pancreas from the background region, and output the pancreatic region cluster map.

[0025] S32: Extract the topological features and gray-level distribution features of the clusters from the cluster map of the pancreatic region, introduce a small number of labeled samples to construct a semi-supervised loss function, take the feature deviation between the cluster center and the labeled samples as the optimization objective, iteratively update the cluster center parameters, and output the optimized cluster center.

[0026] S33: Perform secondary pixel division on the pancreatic region cluster map carrying topological labels based on the cluster center parameter set, further subdivide edematous, viable, and necrotic tissues according to the matching degree between cluster centers and pixel features, and integrate the division results to generate a preliminary segmentation mask that matches the boundaries and topological structure of each tissue.

[0027] According to the intelligent segmentation method for CT images of acute pancreatitis provided by the present invention, the specific steps for outputting the optimized mask in step S4 are as follows:

[0028] S41: Collect clinically measured data on the elastic modulus and Poisson's ratio of pancreatic tissue at different disease stages, establish a mechanical parameter library covering edema, viable and necrotic areas, and output a set of tissue mechanical parameters.

[0029] S42: Assign corresponding mechanical parameters to different tissue regions of the preliminary segmented mask using the tissue mechanical parameter set, perform mesh and parameter assignment for the dynamic mechanical finite element analysis model, and output the parameterized finite element model.

[0030] S43: Simulate the mechanical deformation of pancreatic tissue under inflammatory conditions based on the parametric finite element model, correct boundary ambiguity and deformation deviation, make the segmentation boundary fit the real anatomical shape, and output an optimized mask.

[0031] According to the intelligent segmentation method for CT images of acute pancreatitis provided by the present invention, the specific steps for outputting the parameterized finite element model in step S42 are as follows:

[0032] Identify the pixel region boundaries of three types of tissues in the preliminary segmentation mask: edematous, viable, and necrotic tissues. Perform mesh generation according to the requirements of finite element analysis, generate a finite element mesh model that matches the tissue region, and output the pancreatic tissue mesh generation map.

[0033] The pancreatic tissue mesh is used to divide the tissue type of each mesh unit labeled with the icon. The elastic modulus and Poisson's ratio of the corresponding tissue are matched with the tissue mechanical parameter set. The parameters are assigned to each finite element mesh unit one by one, and the mechanical parameter mesh model is output.

[0034] The boundary conditions and dynamic analysis time step of the finite element model are defined based on the mechanical parameter mesh model. The mesh and parameter information are integrated to output the parameterized finite element model.

[0035] According to the intelligent segmentation method for CT images of acute pancreatitis provided by the present invention, the specific steps for outputting the region segmentation mask in step S5 are as follows:

[0036] S51: Extract the pixel grayscale features of candidate regions for pseudocysts and peripancreatic effusion in the optimized mask, calculate the image entropy value under different thresholds, determine the optimal segmentation threshold when the entropy value is the maximum, and output the optimal threshold parameter.

[0037] S52: Perform grayscale threshold segmentation on candidate region pixels of the optimized mask according to the optimal threshold parameter, initially distinguish the boundary between pseudocysts and peripancreatic exudate, and output the intermediate segmentation image.

[0038] S53: Correct boundary noise and blurred areas based on the boundary information of the intermediate segmentation image and the original topology of the optimized mask, and output the region segmentation mask.

[0039] According to the intelligent segmentation method for CT images of acute pancreatitis provided by the present invention, the specific steps for outputting the region segmentation results in step S6 are as follows:

[0040] S61: Extract the pixel region of necrotic tissue from the region segmentation mask, calculate the total number of pixels in the region, calculate the actual volume of necrotic tissue and the total volume of pancreas based on the mapping relationship between image pixels and actual size, and output the necrotic tissue volume data.

[0041] S62: Based on the necrotic tissue volume data and the pixel distribution of the peripancreatic exudate region in the region segmentation mask, extract the cross-sectional pixel set of the exudate region along different anatomical sections, calculate the actual area of ​​each cross section, and output the exudate cross-sectional area dataset.

[0042] S63: Based on the exudate cross-sectional area dataset, select the cross-sectional area with the largest value as the maximum exudate cross-sectional area, calculate the ratio of necrotic volume to total pancreatic volume to obtain the necrotic volume percentage, and integrate the two indicators to output a structured region segmentation result.

[0043] This invention also provides an intelligent segmentation system for CT images of acute pancreatitis, comprising:

[0044] The virtual lesion module is used to generate virtual lesions by simulating normal pancreatic CT images based on physical fluid diffusion, and outputs a set of virtual CT images.

[0045] The topology extraction module is used to fuse virtual CT image sets with real CT images. It extracts the topological features of pancreatic tissue from the fused images using discrete Morse theory and outputs a pancreatic topological image.

[0046] The clustering subdivision module is used to perform unsupervised clustering of topological images using topologically-aware diffusion-agglomeration to distinguish pancreatic regions from background regions. Labeled samples are introduced to construct a semi-supervised loss function, optimizing cluster centers, subdividing edematous, viable, and necrotic tissues, and outputting a preliminary segmentation mask.

[0047] The mechanical correction module is used to construct a dynamic mechanical finite element analysis model. Based on the measured data of elastic modulus and Poisson's ratio of pancreatic tissue at different disease stages, it assigns differentiated mechanical parameters to each region of the initial segmentation mask, simulates the mechanical properties of the tissue, corrects boundary deformation, and outputs an optimized mask.

[0048] The pixel optimization module is used to refine the pixels of the optimized mask using the entropy value optimization threshold method, determine the boundary between pseudocysts and peripancreatic effusion, and output a region segmentation mask.

[0049] The quantization output module is used to automatically calculate the proportion of necrotic volume and the maximum cross-sectional area of ​​exudation based on the region segmentation mask, and output the region segmentation results.

[0050] The intelligent segmentation method for CT images of acute pancreatitis provided by this invention has the following beneficial effects:

[0051] 1. This invention employs a physics-based fluid diffusion simulation technique. Using a normal pancreatic image feature set as a benchmark, and combining it with the pathological patterns of edema-necrosis-exudation in acute pancreatitis, differentiated diffusion parameters are set to generate a virtual CT image set covering mild and severe cases. This design effectively alleviates the problem of scarce labeled samples for acute pancreatitis and the difficulty in obtaining severe cases in clinical practice, providing sufficient and diverse data support for algorithm training. Simultaneously, the rigid registration between virtual and real images eliminates differences in equipment and scanning parameters, allowing the algorithm to maintain stable segmentation accuracy under different data sources, significantly improving the model's clinical generalization ability.

[0052] 2. This invention introduces discrete Morse theory to extract the topological features of the pancreas. The constructed topological map can accurately capture the connectivity and boundary continuity between the pancreas and surrounding tissues. Based on this, a topological sensing diffusion-aggregation method combined with a semi-supervised loss function achieves unsupervised subdivision of edematous, viable, and necrotic tissues. A dynamic finite element model corrects for boundary deformation caused by inflammation and edema based on the mechanical parameters of different tissues. An entropy-optimized threshold method further distinguishes the boundaries between pseudocysts and peripancreatic exudate. Multi-module layer-by-layer optimization solves the problem of low accuracy in segmenting fuzzy boundaries in traditional algorithms, ultimately achieving accurate segmentation of four regions: pancreatic parenchyma, necrotic area, exudative area, and pseudocysts, providing reliable imaging evidence for disease assessment.

[0053] 3. This invention calculates key indicators such as the proportion of necrotic volume and the maximum cross-sectional area of ​​exudate, and combines this with the Atlanta grading rule base to generate disease grading conclusions, directly serving clinical diagnostic decisions. Furthermore, it compresses the entire process using homotopy simplification technology, forming a lightweight algorithm package that can run locally on clinical workstations, without relying on high-performance computing clusters. This integrated design of precise segmentation, quantitative analysis, and lightweight deployment bridges the gap between the algorithm's laboratory and clinical applications, meeting both the segmentation accuracy requirements of radiologists and the convenience requirements of clinical scenarios, demonstrating extremely high commercialization value. Attached Figure Description

[0054] The invention will now be further described with reference to the accompanying drawings.

[0055] Figure 1 This is a flowchart illustrating the steps of the intelligent segmentation method for CT images of acute pancreatitis provided in this embodiment of the invention.

[0056] Figure 2 This is a flowchart of the intelligent segmentation method for CT images of acute pancreatitis provided in this embodiment of the invention;

[0057] Figure 3 This is a block diagram of the intelligent segmentation system for CT images of acute pancreatitis provided in an embodiment of the present invention. Detailed Implementation

[0058] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below according to specific embodiments.

[0059] like Figures 1 to 3 As shown in the embodiment of the present invention, the intelligent segmentation method for CT images of acute pancreatitis includes:

[0060] S1: Using a physical fluid diffusion simulation, virtual lesions are generated from normal pancreatic CT images, and a set of virtual CT images covering different stages of mild and severe cases is output.

[0061] S11: Obtain normal pancreatic CT images and corresponding anatomical structure annotations, extract basic features such as CT value distribution and contour morphology of pancreatic parenchyma, and output normal pancreatic image feature set.

[0062] Enhanced abdominal CT images of healthy individuals with no pancreatic disease or history of abdominal surgery were selected. Senior radiologists annotated the pancreatic anatomy, clearly defining the boundaries of the pancreatic head, body, and tail, and their relative positions to surrounding blood vessels and intestines. Based on the annotated areas, quantitative features such as CT value ranges, grayscale histogram distribution, and contour smoothness of the pancreatic parenchyma were extracted. Simultaneously, image scanning parameters were recorded, and these were ultimately integrated to form a standardized set of normal pancreatic image features, providing a benchmark template for subsequent lesion simulation.

[0063] S12: Based on the pathological evolution of acute pancreatitis, fluid diffusion parameters for edema, necrosis, and exudation are set. The normal pancreatic image feature set is used as the initial diffusion field to simulate the fluid diffusion process in the lesion area and output the lesion prototype image.

[0064] Referring to the pathological progression of acute pancreatitis—edema → necrosis → peripancreatic exudation—differentiated fluid diffusion parameters were set: diffusion coefficient 0.8 and viscosity 0.5 for the edema area, diffusion coefficient 1.2 and viscosity 0.3 for the necrosis area, and diffusion coefficient 1.5 and viscosity 0.2 for the exudation area. The normal pancreatic image feature set output from S11 was used as the initial field for diffusion simulation. The infiltration process of the lesion area within the pancreatic parenchyma was simulated using fluid dynamics equations, resulting in diffuse low-density changes in the edema area, focal extremely low-density areas in the necrosis area, and exudation extending into the peripancreatic fat space, ultimately generating a preliminary lesion image that preserves the basic anatomical structure.

[0065] S13: Combining the disease course characteristics of mild and severe cases, adjust the time step and range of diffusion simulation, optimize the morphology of the lesion area in the lesion prototype image, make the lesion distribution conform to the clinical pathological characteristics, and output a virtual CT image set.

[0066] Based on preliminary lesion images, for mild pancreatitis, a short time step and small diffusion range are set to optimize the blurring of the edema area boundaries, making it consistent with the imaging manifestations of mild cases. For severe pancreatitis, a long time step and large diffusion range are set to enhance the irregular morphology of the necrotic area and the flocculent distribution characteristics of the exudate. At the same time, the CT values ​​of the lesion area are calibrated to match real clinical cases, ultimately generating a virtual CT image set covering different stages of mild and severe pancreatitis.

[0067] S2: The virtual dataset is fused with real CT images, and the topological features of pancreatic tissue are extracted using discrete Morse theory to output a pancreatic topological map.

[0068] S21: Perform rigid registration between the virtual CT image set and the real acute pancreatitis CT images, align the anatomical location of the pancreas with the scanning parameters, eliminate image shifts caused by equipment differences, and output normalized CT images.

[0069] Anatomical landmarks of the pancreas are extracted from both virtual and real CT images to establish a correspondence between the landmarks in the two images, and a set of landmark coordinates is output. Based on the coordinate set, the rigid registration matrix is ​​solved using the least squares method. The virtual CT images are then translated and rotated to accurately align the anatomical positions of the pancreas in the two images, and a preliminary registered image is output. The scanning parameters of the preliminary registered image are standardized to eliminate grayscale differences caused by scanning from different devices, and a normalized CT image is output.

[0070] S22: Calculate the gradient value of the image gray field based on the normalized CT image, identify the gradient extrema and critical connection lines through discrete Morse theory, capture the connectivity and boundary continuity features of pancreatic tissue, and output the pancreatic topological feature point set.

[0071] The Sobel operator is used to calculate the gray-level gradient of normalized CT images to obtain the gradient magnitude and direction of each pixel, generating a gray-level gradient field image that highlights the gray-level change boundary between the pancreas and surrounding tissues.

[0072] Based on discrete Morse theory, a gradient threshold is set to filter gradient extrema points and output the initial set of extrema points.

[0073] By using the critical connection rule in discrete Morse theory, adjacent extreme points are connected to form critical edges, and extreme points and critical edges belonging only to the pancreatic region are selected, outputting a set of topological feature points.

[0074] S23: Construct the topological connection relationship between feature points based on the pancreatic topological feature point set, restore the spatial topological structure of the pancreatic parenchyma and surrounding tissues, and output the pancreatic topological structure image.

[0075] Based on a set of pancreatic topological feature points, a Delaunay triangulation algorithm is used to construct a topological connectivity network between feature points, initially reconstructing the spatial morphology of the pancreas and outputting the topological connectivity network. Inappropriate connections (such as abnormal connections crossing the intestine) are removed from the topological connectivity network to optimize the topological structure. The optimized topological connectivity network is then visualized and rendered, using different lines to distinguish the pancreatic parenchymal boundaries from the internal connected structures, reducing background interference, and outputting a clear image of the pancreatic topological structure.

[0076] S3: Guided by the topology map, the topology-aware diffusion-aggregation method is used to perform unsupervised clustering of CT image pixels. Annotated samples are introduced to construct a semi-supervised loss function, optimize the cluster centers, subdivide edematous, viable and necrotic tissues, and output a preliminary segmentation mask (preliminary regional segmentation mask image of pancreatic tissue).

[0077] S31: Calculate the topological similarity and spatial distance between pixels in the pancreatic topological structure image using the topological sensing diffusion agglomeration method, aggregate pixels according to the similarity threshold to complete unsupervised clustering, distinguish the pancreas from the background region, and output the pancreatic region cluster map.

[0078] Extract the topological and grayscale attributes of each pixel in the pancreatic topological structure image, construct a pixel-level feature vector, calculate the topological similarity and Euclidean distance between any two pixels, and output the pixel feature similarity matrix.

[0079] The formula for topological similarity is expressed as:

[0080]

[0081] In the formula, Let be the topological similarity between pixel i and pixel j. , These are the connectivity and boundary distance weights, respectively. , Let be the connectivity coefficients of pixels i and j. The maximum value of the connectivity coefficient. Let be the distance difference from pixels i and j to the pancreatic boundary. This represents the standard deviation of the boundary distance.

[0082] Set a topological similarity threshold and a spatial distance threshold, and use a diffusion-agglomeration algorithm to aggregate pixels with high similarity as the core to form initial clusters and output the initial clustering results.

[0083] Based on the prior knowledge of pancreatic topology, scattered and disconnected background clusters are removed, while clusters that conform to pancreatic morphological characteristics are retained, resulting in a well-defined pancreatic region cluster map.

[0084] S32: Extract the topological features and gray-level distribution features of the clusters from the cluster map of the pancreatic region, introduce a small number of labeled samples to construct a semi-supervised loss function, take the feature deviation between the cluster center and the labeled samples as the optimization objective, iteratively update the cluster center parameters, and output the optimized cluster center.

[0085] The core features of each cluster are extracted from the cluster map of the pancreatic region: topological features include the number of connected branches and boundary smoothness of the cluster, and grayscale features include the mean and variance of the CT value within the cluster. These features are integrated to form a cluster feature set, and the feature quantification results are output.

[0086] A small number of labeled samples are introduced, and the feature vectors of the labeled samples are extracted. A semi-supervised loss function is constructed. The loss term includes the feature deviation between the cluster center and the labeled sample, and the feature consistency of pixels within the cluster. The loss function model is then output.

[0087] The formula for the semi-supervised loss function is as follows:

[0088]

[0089] In the formula, For cross-entropy loss, λ is the regularization loss, and λ is the regularization weight.

[0090]

[0091] In the formula, For cross-entropy loss, To label the number of samples, The true label for sample k belonging to category c. Let k be the predicted probability that sample k belongs to class c.

[0092]

[0093] In the formula, The loss function is the regularization loss, where N is the total number of pixels. Let i be the feature vector of pixel i. Let be the cluster center feature vector of category c.

[0094] With the goal of minimizing the loss function value, the gradient descent method is used to iteratively update the feature parameters of the cluster centers until the loss value converges, and the optimized cluster centers that accurately match the features of the labeled samples are output.

[0095] S33: Perform secondary pixel division on the pancreatic region cluster map carrying topological labels based on the cluster center parameter set, further subdivide edematous, viable, and necrotic tissues according to the matching degree between cluster centers and pixel features, and integrate the division results to generate a preliminary segmentation mask that matches the boundaries and topological structure of each tissue.

[0096] Retrieve the optimized cluster center parameter set, calculate the feature matching degree between each pixel in the pancreatic region cluster map and the three cluster centers of edema, survival, and necrosis, and output the pixel-cluster center matching degree matrix.

[0097] The formula for calculating feature matching degree is expressed as follows:

[0098]

[0099] In the formula, The degree of matching between pixel i and the cluster center of category c. , The dot product of the eigenvectors, , Let be the Euclidean norm of the eigenvectors.

[0100] Set a matching threshold, assign each pixel to the cluster category with the highest matching degree, complete the pixel-level organization type division, and output the pixel classification map of the three types of organizations.

[0101] By combining the boundary features of pancreatic topological structure images, the tissue boundaries of the classification map are smoothed to eliminate isolated noise pixels, integrate the three types of tissue regions, and generate a preliminary segmentation mask carrying topological labels.

[0102] S4: Construct a finite element analysis model of dynamic mechanical properties. Based on clinically measured data of elastic modulus and Poisson's ratio of pancreatic tissue at different disease stages, assign differentiated mechanical parameters to different regions of the initial segmentation mask, simulate the mechanical properties of the pancreas and peripancreatic tissues, correct boundary deformation caused by inflammation and edema, optimize the fuzzy boundary between exudate and adipose tissue, and output an optimized mask with precise boundaries.

[0103] S41: Collect clinically measured data on the elastic modulus and Poisson's ratio of pancreatic tissue at different disease stages, establish a mechanical parameter library covering edema, viable and necrotic areas, and output a set of tissue mechanical parameters.

[0104] Clinical biomechanical data of patients with mild edematous and severe necrotizing acute pancreatitis were screened, and the original measured values ​​of elastic modulus and Poisson's ratio of pancreatic tissues of three types (edematous, viable, and necrotic) were extracted to output a multi-stage biomechanical dataset.

[0105] Outlier removal and data normalization are performed on the original dataset to eliminate errors caused by differences in detection equipment and individual samples, and output a standardized subset of mechanical parameters.

[0106] By combining the pathological grading standards of pancreatic tissue, standardized parameters are mapped one-to-one with edema, viable, and necrotic tissue types to construct a well-defined histomechanical parameter library and output a structured histomechanical parameter set.

[0107] S42: Assign corresponding mechanical parameters to different tissue regions of the preliminary segmented mask using the tissue mechanical parameter set, perform mesh and parameter assignment for the dynamic mechanical finite element analysis model, and output the parameterized finite element model.

[0108] Identify the pixel region boundaries of three types of tissues in the preliminary segmentation mask: edematous, viable, and necrotic tissues. Perform mesh generation according to the requirements of finite element analysis, generate a finite element mesh model that matches the tissue region, and output the pancreatic tissue mesh generation map.

[0109] The pancreatic tissue mesh is used to divide the tissue type of each mesh unit labeled with the icon. The elastic modulus and Poisson's ratio of the corresponding tissue are matched with the tissue mechanical parameter set. The parameters are assigned to each finite element mesh unit one by one, and the mechanical parameter mesh model is output.

[0110] The boundary conditions and dynamic analysis time step of the finite element model are defined based on the mechanical parameter mesh model. The mesh and parameter information are integrated to output the parameterized finite element model.

[0111] S43: Simulate the mechanical deformation of pancreatic tissue under inflammatory conditions based on the parametric finite element model, correct boundary ambiguity and deformation deviation, make the segmentation boundary fit the real anatomical shape, and output an optimized mask.

[0112] Based on the parametric finite element model, a dynamic pressure load matching the course of inflammation is applied, and a mechanical deformation simulation is run to calculate the boundary deformation displacement of pancreatic tissue caused by edema and necrosis, and output the tissue deformation displacement field distribution data.

[0113] The deformation displacement field data is mapped to the boundary pixels of the initial segmentation mask, the blurred boundaries are corrected by displacement, boundary noise caused by algorithm error is removed, and the intermediate segmentation mask after boundary correction is output.

[0114] By comparing the fit between the corrected intermediate segmentation mask and the actual clinical anatomical morphology of the pancreas, the boundary pixel coordinates are finely adjusted to ensure that the boundaries of the edema and necrosis areas are consistent with the clinical pathological features, and finally, an optimized mask with precise boundaries is output.

[0115] S5: The entropy value optimization threshold method is used to refine the pixels of the optimized mask, remove artifact noise, clearly distinguish the boundary between pseudocysts and peripancreatic exudate, and output a region segmentation mask that labels four types of regions: pancreatic parenchyma, necrotic area, exudate area, and pseudocyst.

[0116] S51: Extract the pixel grayscale features of candidate regions for pseudocysts and peripancreatic effusion in the optimized mask, calculate the image entropy value under different thresholds, determine the optimal segmentation threshold when the entropy value is the maximum, and output the optimal threshold parameter.

[0117] Candidate regions for pseudocysts and peripancreatic effusion are located from the optimized mask. The gray values ​​of all pixels within the region are extracted, and the distribution range and frequency of the gray values ​​are statistically analyzed. The gray value distribution histogram of the candidate region is then output.

[0118] The determination of candidate regions for peripancreatic exudate is based on a multi-dimensional logical connection of optimized mask spatial constraints, gray-scale feature threshold screening, anatomical location anchoring, and neighborhood connectivity analysis. The specific steps are as follows:

[0119] The edge contour of the pancreas body is extracted and optimized, and a preset anatomical safety distance is extended outward to form a zone of interest (ROI) around the pancreas. The boundary of this ROI is determined by the anatomical adjacency between the pancreas and surrounding tissues, excluding irrelevant areas far from the pancreas. At the same time, the pixel masking function of the optimized mask is used to remove pixel units within the ROI that have been labeled as pancreatic parenchyma or pseudocysts, thus achieving preliminary spatial limitation of the candidate region.

[0120] Grayscale values ​​of the remaining pixels within the band of interest are sampled. Based on prior knowledge of the grayscale distribution of peripancreatic exudate in clinical images, a grayscale threshold range is set to filter out interfering pixels such as adipose tissue and vascular structures with excessively high grayscale values, and calcifications with excessively low grayscale values, thus selecting a subset of pixels that meet the grayscale characteristics of exudate.

[0121] Connectivity analysis is performed on the filtered subset of pixels to calculate the area, morphological factor and other geometric parameters of each connected region. Isolated noise points with too small an area and artifact regions with irregular shapes are removed, and connected regions with continuous distribution characteristics are retained.

[0122] By merging these connected components that meet the requirements of spatial location, grayscale features, and geometric shape, the candidate region for peripancreatic effusion can be determined. The pixels in this region will be included in the subsequent grayscale value extraction and distribution statistics process. In the above steps, the spatial reference of the optimized mask defines the core range of the candidate region, the grayscale threshold screening achieves the initial purification at the feature level, and the connected component analysis completes the verification of the integrity of the region. The three form a progressive logical dependency relationship to ensure the accuracy and effectiveness of the candidate region.

[0123] Set the traversal range of the grayscale threshold, select several threshold points within the range, calculate the entropy values ​​of the image foreground and background under each threshold, sum them to obtain the total entropy value of the corresponding threshold, and output the multi-threshold total entropy value dataset.

[0124] The formula for calculating the entropy value of the background region is expressed as follows:

[0125]

[0126] In the formula, For grayscale frequency, The sum of grayscale frequencies is denoted by T, where T is the grayscale threshold.

[0127] The formula for calculating the entropy of the foreground region is expressed as:

[0128]

[0129] In the formula, L represents the number of gray levels, and T represents the gray level threshold. For grayscale frequency, It is the sum of grayscale frequencies.

[0130] By comparing the values ​​in the multi-threshold total entropy dataset, the threshold point with the largest total entropy value is selected and determined as the optimal segmentation threshold. The structured optimal threshold parameters are then output.

[0131] S52: Perform grayscale threshold segmentation on candidate region pixels of the optimized mask according to the optimal threshold parameter, initially distinguish the boundary between pseudocysts and peripancreatic exudate, and output the intermediate segmentation image.

[0132] Import the optimal threshold parameter, perform threshold determination on the pixel grayscale values ​​of the optimized mask candidate region, mark pixels with grayscale values ​​higher than the threshold as pseudocyst regions, and mark pixels with grayscale values ​​lower than the threshold as peripancreatic exudate regions, and output the pixel labeling matrix.

[0133] Based on the pixel marker matrix, the preliminary boundary contours of the pseudocyst and peripancreatic effusion are drawn. The contour information is then superimposed onto the candidate regions corresponding to the optimized mask, and the initial segmentation map with boundary markers is output.

[0134] Connectivity analysis is performed on the initial segmentation map to remove isolated noise regions with an area smaller than a set threshold, retaining the target regions that conform to clinical morphological characteristics, and outputting an intermediate segmentation map with preliminary boundary formation.

[0135] S53: Correct boundary noise and blurred areas based on the boundary information of the intermediate segmentation image and the original topology of the optimized mask, and output the region segmentation mask.

[0136] Extract the boundary pixel coordinates of the intermediate segmentation image, combine them with the original topology of the optimized mask, analyze the neighborhood connectivity of the boundary pixels, identify the spiky noise and blurred transition regions on the boundary, and output the boundary abnormal region location data.

[0137] Morphological closing operations are used to smooth the boundary abnormal regions, secondary grayscale determination is performed on the pixels in the blurred transition region, the boundary pixel coordinates are fine-tuned, and the transition segmentation map after boundary correction is output.

[0138] By comparing the anatomical features of clinical pancreatic images, the boundary fit of the transition segmentation map was verified. Local deviation areas were manually fine-tuned and optimized to output a region segmentation mask with accurate boundaries.

[0139] S6: Based on the region segmentation mask, core indicators such as the proportion of necrotic volume and the maximum cross-sectional area of ​​exudation are automatically calculated, and a disease grading conclusion is generated by combining the Atlanta grading expert rule base. At the same time, the entire process is compressed through homotopy simplification technology to form a lightweight segmentation algorithm package that can run locally on clinical workstations. It can directly output four types of region segmentation results by inputting any acute pancreatitis CT image.

[0140] S61: Extract the pixel region of necrotic tissue from the region segmentation mask, calculate the total number of pixels in the region, calculate the actual volume of necrotic tissue and the total volume of pancreas based on the mapping relationship between image pixels and actual size, and output the necrotic tissue volume data.

[0141] Identify and extract the set of pixels marked as necrotic tissue from the region segmentation mask, remove isolated noise pixels with an area smaller than a set threshold within the set, and obtain a clean necrotic tissue pixel region without interference. Output the necrotic tissue pixel region.

[0142] Count the total number of pixels in the necrotic tissue region and the total number of pixels in the entire pancreatic tissue in the region segmentation mask. Complete the pixel count for both types of regions and output the pixel count results for the necrotic tissue and the overall pancreas.

[0143] The pixel-to-actual-size mapping relationship preset by the imaging device is invoked to convert the number of pixels of necrotic tissue and pancreas into corresponding actual volume values. The two volume data are then integrated to output the volume data of necrotic tissue.

[0144] The formula for calculating the actual volume of necrotic tissue is as follows:

[0145]

[0146]

[0147] In the formula, This represents the actual volume of necrotic tissue. α represents the total volume of the pancreas, and α is the image voxel volume coefficient.

[0148] S62: Based on the necrotic tissue volume data and the pixel distribution of the peripancreatic exudate region in the region segmentation mask, extract the cross-sectional pixel set of the exudate region along different anatomical sections, calculate the actual area of ​​each cross section, and output the exudate cross-sectional area dataset.

[0149] The spatial anatomical range of the pancreas is defined based on the volume data of necrotic tissue. The complete pixel area of ​​peripancreatic exudate is located from the region segmentation mask. Equally spaced sections are set along the three standard anatomical directions of coronal, sagittal and transverse planes, and multi-directional anatomical section positioning information is output.

[0150] Based on the anatomical section location information, extract the cross-sectional pixel set of the peripancreatic exudate region on each section, count the number of pixels in each set, form a list of exudate cross-sectional pixel counts for each section, and output the exudate cross-sectional pixel count results.

[0151] By combining the pixel-to-actual-size mapping relationship of the image, the number of pixels of the exudate cross section of each section is converted into the corresponding actual cross-sectional area. The area data of all sections are summarized, and the exudate cross-sectional area dataset is output.

[0152] The formula for the actual cross-sectional area of ​​the exudate is expressed as:

[0153]

[0154] In the formula, Let be the cross-sectional area of ​​the exudate from the i-th section. β represents the number of pixels in the exudate section of the i-th anatomical section, and β is the image pixel area coefficient.

[0155] S63: Based on the exudate cross-sectional area dataset, select the cross-sectional area with the largest value as the maximum exudate cross-sectional area, calculate the ratio of necrotic volume to total pancreatic volume to obtain the necrotic volume percentage, and integrate the two indicators to output a structured region segmentation result.

[0156] By comparing the cross-sectional area values ​​of all sections from the exudate cross-sectional area dataset, the section with the largest value is selected as the maximum exudate cross-sectional area. At the same time, the anatomical section information corresponding to this section is recorded, and the maximum exudate cross-sectional area and the corresponding section data are output.

[0157] Retrieve the actual volume of necrotic tissue and the total volume of the pancreas from the necrotic tissue volume data, calculate the ratio between the two and convert it into a percentage to obtain the necrotic volume percentage data, and output the necrotic volume percentage data.

[0158] The formula for calculating the percentage of necrotic volume is as follows:

[0159]

[0160] In the formula, This represents the actual volume of necrotic tissue. This represents the total volume of the pancreas.

[0161] The system integrates the maximum cross-sectional area of ​​exudation, the corresponding cross-sectional information, and the proportion of necrotic volume, and organizes them in a structured manner according to the standard format of clinical diagnostic reports to output the region segmentation results.

[0162] like Figure 3 As shown, the present invention also provides an intelligent segmentation system for CT images of acute pancreatitis, comprising:

[0163] The virtual lesion module is used to generate virtual lesions by simulating normal pancreatic CT images based on physical fluid diffusion, and outputs a set of virtual CT images.

[0164] The topology extraction module is used to fuse virtual CT image sets with real CT images. It extracts the topological features of pancreatic tissue from the fused images using discrete Morse theory and outputs a pancreatic topological image.

[0165] The clustering subdivision module is used to perform unsupervised clustering of topological images using topologically-aware diffusion-agglomeration to distinguish pancreatic regions from background regions. Labeled samples are introduced to construct a semi-supervised loss function, optimizing cluster centers, subdividing edematous, viable, and necrotic tissues, and outputting a preliminary segmentation mask.

[0166] The mechanical correction module is used to construct a dynamic mechanical finite element analysis model. Based on the measured data of elastic modulus and Poisson's ratio of pancreatic tissue at different disease stages, it assigns differentiated mechanical parameters to each region of the initial segmentation mask, simulates the mechanical properties of the tissue, corrects boundary deformation, and outputs an optimized mask.

[0167] The pixel optimization module is used to refine the pixels of the optimized mask using the entropy value optimization threshold method, determine the boundary between pseudocysts and peripancreatic effusion, and output a region segmentation mask.

[0168] The quantization output module is used to automatically calculate the proportion of necrotic volume and the maximum cross-sectional area of ​​exudation based on the region segmentation mask, and output the region segmentation results.

[0169] In summary, this embodiment provides an intelligent segmentation method and system for CT images of acute pancreatitis. By calculating core indicators such as the proportion of necrotic volume and the maximum cross-sectional area of ​​exudation, and combining this with the Atlanta classification rule base, it generates a disease grading conclusion, directly serving clinical diagnostic decisions. Furthermore, homotopy simplification technology is used to compress the entire process, forming a lightweight algorithm package that can run locally on clinical workstations, without relying on high-performance computing clusters. This integrated design of precise segmentation, quantitative analysis, and lightweight deployment bridges the gap between the algorithm's application in the laboratory and its clinical use, meeting both the segmentation accuracy requirements of radiologists and the convenience requirements of clinical scenarios, demonstrating significant commercialization value.

[0170] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0171] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent segmentation method for CT images of acute pancreatitis, characterized in that, include: S1: Use physical fluid diffusion to simulate normal pancreatic CT images to generate a virtual lesion image set and output the virtual CT image set; S2: Fuse virtual CT images with real CT images, extract the topological features of pancreatic tissue from the fused images using discrete Morse theory, and output a pancreatic topological image. S3: Unsupervised clustering of the topological structure image is performed using the topological sensing diffusion-aggregation method to distinguish the pancreas from the background region; labeled samples are introduced to construct a semi-supervised loss function to optimize the cluster centers, further subdivide edema, viable and necrotic tissues, and output a preliminary segmentation mask; S4: Construct a dynamic mechanical finite element analysis model, assign differentiated mechanical parameters to each region of the initial segmentation mask based on the measured data of elastic modulus and Poisson's ratio of pancreatic tissue at different disease stages, simulate tissue mechanical properties, correct boundary deformation, and output an optimized mask; S5: The pixels of the optimized mask are refined using the entropy value optimization threshold method to determine the boundary between pseudocysts and peripancreatic effusion, and the region segmentation mask is output. S6: Automatically calculate the necrotic volume percentage and the maximum cross-sectional area of ​​exudation based on the region segmentation mask, and output the region segmentation results.

2. The intelligent segmentation method for CT images of acute pancreatitis according to claim 1, characterized in that: In step S1, the specific steps for outputting the virtual CT image set are as follows: S11: Obtain normal pancreatic CT images and corresponding anatomical structure annotations, extract CT value distribution and contour morphology of pancreatic parenchyma, and output normal pancreatic image feature set; S12: Based on the pathological evolution of acute pancreatitis, fluid diffusion parameters for edema, necrosis, and exudation are set. The normal pancreatic image feature set is used as the initial diffusion field to simulate the fluid diffusion process in the lesion area and output the lesion prototype image. S13: Combining the disease course characteristics of mild and severe cases, adjust the time step and range of diffusion simulation, optimize the morphology of the lesion area in the lesion prototype image, so that the lesion distribution conforms to the clinical pathological characteristics, and output a virtual CT image set.

3. The intelligent segmentation method for CT images of acute pancreatitis according to claim 1, characterized in that: In step S2, the specific steps for outputting the pancreatic topological structure image are as follows: S21: Rigidly register the virtual CT image set with the real acute pancreatitis CT images, align the pancreatic anatomical location and scanning parameters, eliminate image shifts caused by equipment differences, and output normalized CT images. S22: Calculate the gradient value of the image gray field based on the normalized CT image, identify the gradient extreme points and critical connection lines through discrete Morse theory, capture the connectivity and boundary continuity features of pancreatic tissue, and output the pancreatic topological feature point set. S23: Construct the topological connection relationship between the feature points based on the pancreatic topological feature point set, restore the spatial topological structure of the pancreatic parenchyma and surrounding tissues, and output the pancreatic topological structure image.

4. The intelligent segmentation method for CT images of acute pancreatitis according to claim 3, characterized in that: In step S22, the specific steps for outputting the pancreatic topological feature point set are as follows: The Sobel operator is used to calculate the gray-level gradient of the normalized CT image to obtain the gradient magnitude and direction of each pixel, and generate a gray-level gradient field image. Based on discrete Morse theory, gradient extreme points are filtered by setting a gradient threshold according to the gray-scale gradient field image, and an initial set of extreme points is output. By using the critical connection rule in discrete Morse theory, adjacent extreme points are connected to form critical edges, and extreme points and critical edges belonging only to the pancreatic region are selected, outputting a set of topological feature points.

5. The intelligent segmentation method for CT images of acute pancreatitis according to claim 1, characterized in that: In step S3, the specific steps for outputting the initial segmentation mask are as follows: S31: Calculate the topological similarity and spatial distance between pixels in the pancreatic topological structure image using the topological sensing diffusion-aggregation method, aggregate pixels according to the similarity threshold to complete unsupervised clustering, distinguish the pancreas from the background region, and output the pancreatic region clustering map; S32: Extract the topological features and gray-level distribution features of the clusters from the cluster map of the pancreatic region, introduce a small number of labeled samples to construct a semi-supervised loss function, take the feature deviation between the cluster center and the labeled sample as the optimization objective, iteratively update the cluster center parameters, and output the optimized cluster center; S33: Perform secondary pixel division on the pancreatic region cluster map carrying topological labels according to the cluster center parameter set, further subdivide edematous, viable, and necrotic tissues according to the matching degree between cluster centers and pixel features, and integrate the division results to generate a preliminary segmentation mask that matches the boundaries and topological structure of each tissue.

6. The intelligent segmentation method for CT images of acute pancreatitis according to claim 1, characterized in that: In step S4, the specific steps for outputting the optimized mask are as follows: S41: Collect clinically measured data on the elastic modulus and Poisson's ratio of pancreatic tissue at different disease stages, establish a mechanical parameter library covering edema, viable and necrotic areas, and output a set of tissue mechanical parameters; S42: Using the set of tissue mechanical parameters, assign corresponding mechanical parameters to different tissue regions of the preliminary segmented mask, perform mesh and parameter assignment for the dynamic mechanical finite element analysis model, and output the parameterized finite element model; S43: Simulate the mechanical deformation of pancreatic tissue under inflammatory conditions based on the parametric finite element model, correct boundary ambiguity and deformation deviation, make the segmentation boundary fit the real anatomical shape, and output an optimized mask.

7. The intelligent segmentation method for CT images of acute pancreatitis according to claim 1, characterized in that: In step S42, the specific steps for outputting the parameterized finite element model are as follows: Identify the pixel region boundaries of three types of tissues in the preliminary segmentation mask: edematous, viable, and necrotic tissues. Perform mesh generation according to the requirements of finite element analysis, generate a finite element mesh model that matches the tissue region, and output the pancreatic tissue mesh generation map. Using the pancreatic tissue mesh to divide the tissue types of each mesh unit marked on the icon, the elastic modulus and Poisson's ratio of the corresponding tissues are matched in the tissue mechanical parameter set, and the parameters are assigned to each finite element mesh unit one by one, and the mechanical parameter mesh model is output. Based on the mechanical parameter mesh model, the boundary conditions and dynamic analysis time step of the finite element model are defined, the mesh and parameter information are integrated, and the parameterized finite element model is output.

8. The intelligent segmentation method for CT images of acute pancreatitis according to claim 1, characterized in that: In step S5, the specific steps for outputting the region segmentation mask are as follows: S51: Extract the pixel grayscale features of candidate regions for pseudocysts and peripancreatic effusion in the optimized mask, calculate the image entropy value under different thresholds, determine the optimal segmentation threshold when the entropy value is the maximum, and output the optimal threshold parameter. S52: Perform grayscale threshold segmentation on the candidate region pixels of the optimized mask according to the optimal threshold parameter, initially distinguish the boundary between pseudocysts and peripancreatic effusion, and output the intermediate segmentation image; S53: Correct boundary noise and blurred areas based on the boundary information of the intermediate segmentation image and the original topology of the optimized mask, and output the region segmentation mask.

9. The intelligent segmentation method for CT images of acute pancreatitis according to claim 1, characterized in that: In step S6, the specific steps for outputting the region segmentation result are as follows: S61: Extract the pixel region of necrotic tissue from the region segmentation mask, calculate the total number of pixels in the region, calculate the actual volume of necrotic tissue and the total volume of pancreas according to the mapping relationship between image pixels and actual size, and output the necrotic tissue volume data. S62: Based on the necrotic tissue volume data and the pixel distribution of the peripancreatic exudate region in the region segmentation mask, extract the cross-sectional pixel set of the exudate region along different anatomical sections, calculate the actual area of ​​each cross section, and output the exudate cross-sectional area dataset. S63: Based on the exudate cross-sectional area dataset, select the cross-sectional area with the largest value as the maximum exudate cross-sectional area, calculate the ratio of necrotic volume to total pancreatic volume to obtain the necrotic volume percentage, and integrate the two indicators to output a structured region segmentation result.

10. An intelligent segmentation system for CT images of acute pancreatitis, which employs the intelligent segmentation method for CT images of acute pancreatitis as described in any one of claims 1 to 9, characterized in that, The segmentation system includes: The virtual lesion module is used to generate virtual lesions by simulating normal pancreatic CT images using physical fluid diffusion, and outputs a set of virtual CT images. The topology extraction module is used to fuse virtual CT image sets with real CT images. It extracts the topological features of pancreatic tissue from the fused images using discrete Morse theory and outputs a pancreatic topological image. The clustering subdivision module is used to perform unsupervised clustering of the topological structure image using the topological sensing diffusion agglomeration method to distinguish the pancreas from the background region; it introduces labeled samples to construct a semi-supervised loss function, optimizes the cluster centers, subdivides edematous, viable, and necrotic tissues, and outputs a preliminary segmentation mask; The mechanical correction module is used to construct a dynamic mechanical finite element analysis model. Based on the measured data of elastic modulus and Poisson's ratio of pancreatic tissue at different disease stages, it assigns differentiated mechanical parameters to each region of the initial segmentation mask, simulates the mechanical properties of the tissue, corrects boundary deformation, and outputs an optimized mask. The pixel optimization module is used to refine the pixels of the optimized mask using the entropy value optimization threshold method, determine the boundary between pseudocysts and peripancreatic effusion, and output a region segmentation mask. The quantization output module is used to automatically calculate the proportion of necrotic volume and the maximum cross-sectional area of ​​exudation based on the region segmentation mask, and output the region segmentation results.

Images