CN-122023316-A - Intelligent segmentation method and system for CT (computed tomography) image of acute pancreatitis

CN122023316ACN 122023316 ACN122023316 ACN 122023316ACN-122023316-A

Abstract

The invention belongs to the technical field of image segmentation, in particular to an intelligent segmentation method and system for CT images of acute pancreatitis, comprising the steps of simulating normal pancreas CT images to generate virtual lesions, fusing a virtual CT image set with a real CT image, extracting topological structure characteristics of pancreatic tissues of the fused images, performing unsupervised clustering on the topological structure images by adopting a topological perception diffusion condensation method, and distinguishing pancreas and background areas; the method comprises the steps of distributing differential mechanical parameters for each region of a preliminary segmentation mask, correcting boundary deformation, adopting an entropy value optimization threshold method to carry out fine correction on pixels of the optimization mask, automatically calculating the necrosis volume ratio and the exudation maximum cross-sectional area according to the region segmentation mask, and outputting a region segmentation result. According to the invention, the virtual image is generated through fluid diffusion, the topology guiding segmentation and the mechanical correction are realized, the accurate partitioning and the quantitative grading are realized, and the clinical diagnosis is assisted.

Inventors

DU ZHAOHUI
WANG RUYI
ZHENG CHUANMING
QIU ZHAOLEI
WANG ZHENJIE

Assignees

蚌埠医科大学第一附属医院(蚌埠医科大学附属肿瘤医院)

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (10)

1. The intelligent segmentation method for the acute pancreatitis CT image is characterized by comprising the following steps of: S1, simulating a normal pancreas CT image by using physical fluid diffusion to generate a virtual lesion image set, and outputting the virtual CT image set; S2, fusing the virtual CT image set with the real CT image, extracting topological structure features of pancreatic tissues of the fused image through a discrete Morse theory, and outputting a pancreatic topological structure image; S3, performing unsupervised clustering on the topological structure image by adopting a topological perception diffusion condensation method, distinguishing pancreas from background areas, introducing a labeling sample to construct a semi-supervised loss function, optimizing a clustering center, subdividing edema, survival and necrosis tissues, and outputting a preliminary segmentation mask; S4, constructing a dynamic mechanical finite element analysis model, distributing differential mechanical parameters for each region of the preliminary segmentation mask according to the elastic modulus and Poisson' S ratio measured data of pancreatic tissues in different courses of disease, simulating mechanical properties of the tissues, correcting boundary deformation, and outputting an optimized mask; s5, performing refined correction on pixels of the optimized mask by adopting an entropy value optimization threshold method, determining and distinguishing boundary between pseudocyst and peripancreatic exudate, and outputting an area segmentation mask; and S6, automatically calculating the necrosis volume ratio and the exudation maximum cross-sectional area according to the region segmentation mask, and outputting a region segmentation result.
2. The intelligent segmentation method for acute pancreatitis CT images according to claim 1, wherein in step S1, the specific steps of outputting the virtual CT image set are as follows: s11, acquiring a normal pancreas CT image and corresponding anatomical structure labels, extracting CT value distribution and outline morphology of a pancreas parenchymal region, and outputting a normal pancreas image feature set; S12, setting fluid diffusion parameters of edema, necrosis and exudation based on a pathological evolution rule of acute pancreatitis, taking a normal pancreas image feature set as a diffusion initial field, simulating a fluid diffusion process of a pathological change region, and outputting a pathological change embryonic image; And S13, combining the disease course characteristics of the mild symptoms and the severe symptoms, adjusting the time step and the range of diffusion simulation, performing morphological optimization on the lesion area of the lesion embryonic image, enabling the lesion distribution to accord with the clinical pathological characteristics, and outputting a virtual CT image set.
3. The intelligent segmentation method for acute pancreatitis CT images according to claim 1, wherein in step S2, the specific steps of outputting pancreatic topological structure images are as follows: s21, performing rigid registration on the virtual CT image set and a real acute pancreatitis CT image, aligning the pancreatic anatomy position with the scanning parameters, eliminating image offset caused by equipment difference, and outputting a normalized CT image; S22, calculating a gradient value of an image gray field according to the normalized CT image, identifying gradient extreme points and critical connecting lines through a discrete Morse theory, capturing connectivity and boundary continuity characteristics of pancreatic tissues, and outputting a pancreatic topological characteristic point set; S23, constructing a topological connection relation among feature points according to the pancreas topological feature point set, reducing a space topological structure of pancreas parenchyma and surrounding tissues, and outputting a pancreas topological structure image.
4. The intelligent segmentation method for acute pancreatitis CT images as set forth in claim 3, wherein in step S22, the specific steps of outputting the pancreatic topological feature point set are as follows: Carrying out gray gradient calculation on the normalized CT image by adopting a Sobel operator to obtain the gradient amplitude and direction of each pixel, and generating a gray gradient field image; based on a discrete Morse theory, setting a gradient threshold value according to the gray gradient field image to screen gradient extreme points, and outputting an initial extreme point set; And connecting adjacent extreme points to form a critical edge according to a critical connecting line judging rule in the discrete Morse theory, screening out the extreme points and the critical edges only belonging to the pancreas region, and outputting a topological characteristic point set.
5. The intelligent segmentation method for acute pancreatitis CT images according to claim 1, wherein in step S3, the specific step of outputting the preliminary segmentation mask is: s31, calculating the topological similarity and the spatial distance between the pancreatic topological structure image pixels by a topological perception diffusion condensation method, aggregating the pixels according to a similarity threshold value to complete unsupervised clustering, distinguishing pancreas and background areas, and outputting a pancreas area clustering graph; S32, extracting topological features and gray distribution features of clusters from the pancreatic region cluster map, introducing a small amount of labeling samples to construct a semi-supervised loss function, taking feature deviation of a cluster center and the labeling samples as an optimization target, iteratively updating cluster center parameters, and outputting an optimized cluster center; And S33, carrying out secondary pixel division on the pancreatic region cluster map carrying the topological mark according to the cluster center parameter set, subdividing edema, survival and necrosis tissues according to the matching degree of the cluster center and the pixel characteristics, and integrating the division result to generate a primary segmentation mask with the matching of each tissue boundary and the topological structure.
6. The intelligent segmentation method for acute pancreatitis CT images according to claim 1, wherein in step S4, the specific steps of outputting an optimized mask are as follows: S41, acquiring elastic modulus and Poisson' S ratio clinical actual measurement data of pancreatic tissues in different courses, establishing a mechanical parameter library covering edema, survival and necrosis areas, and outputting a tissue mechanical parameter set; S42, allocating corresponding mechanical parameters for different tissue areas of the preliminary segmentation mask by using the tissue mechanical parameter set, carrying out grid and parameter assignment of a dynamic mechanical finite element analysis model, and outputting a parameterized finite element model; S43, simulating the mechanical deformation of pancreatic tissues in an inflammatory state according to the parameterized finite element model, correcting the boundary blurring and deformation deviation, enabling the segmentation boundary to be attached to the real anatomical form, and outputting an optimization mask.
7. The intelligent segmentation method for acute pancreatitis CT images according to claim 1, wherein in step S42, the specific steps of outputting the parameterized finite element model are as follows: identifying the pixel region boundaries of edema, survival and necrosis tissues in the preliminary segmentation mask, performing grid division according to the finite element analysis requirement, generating a finite element grid model matched with the tissue region, and outputting a pancreatic tissue grid division map; Matching the elastic modulus and the poisson ratio of corresponding tissues in the mechanical parameter set of the tissues by using the tissue types of each grid cell marked by the pancreatic tissue grid dividing map, assigning the parameters to each finite element grid cell one by one, and outputting a mechanical parameter grid model; And defining boundary conditions and dynamic analysis time steps of the finite element model according to the mechanical parameter grid model, integrating grid and parameter information, and outputting the parameterized finite element model.
8. The intelligent segmentation method for acute pancreatitis CT images according to claim 1, wherein in step S5, the specific steps of outputting the region segmentation mask are as follows: S51, extracting pixel gray features of pseudocyst and pancreatic exudate candidate areas in an optimized mask, calculating image entropy values under different thresholds, determining an optimal segmentation threshold when the entropy value is maximum, and outputting optimal threshold parameters; S52, carrying out gray threshold segmentation on pixels of a candidate region of the optimization mask according to the optimal threshold parameters, primarily distinguishing boundary between pseudocyst and peripancreatic exudate, and outputting an intermediate segmentation map; And S53, correcting boundary noise points and fuzzy areas according to the boundary information of the intermediate segmentation map and the original topological structure of the optimization mask, and outputting the area segmentation mask.
9. The intelligent segmentation method for acute pancreatitis CT images according to claim 1, wherein in step S6, the specific steps of outputting the region segmentation result are as follows: S61, extracting pixel areas of necrotic tissues in the area segmentation mask, calculating the total pixel number of the areas, calculating the actual volume of the necrotic tissues and the total pancreatic volume according to the mapping relation between the image pixels and the actual size, and outputting necrotic tissue volume data; S62, extracting a cross-section pixel set of an effusion area along different anatomical sections according to the necrotic tissue volume data and the pixel distribution of the peripancreatic effusion area in the area segmentation mask, calculating the actual area of each cross section, and outputting an effusion cross-section area data set; And S63, screening out the cross section with the largest value as the exudation maximum cross section according to the exudates cross section area data set, calculating the ratio of the necrosis volume to the total pancreas volume to obtain the necrosis volume ratio, and integrating the two indexes to output a structured region segmentation result.
10. An acute pancreatitis CT image intelligent segmentation system, which adopts the acute pancreatitis CT image intelligent segmentation method as defined in any one of claims 1 to 9, characterized in that the segmentation system comprises: The virtual focus module is used for simulating normal pancreas CT images based on physical fluid diffusion to generate virtual lesions and outputting a virtual CT image set; The topology extraction module is used for fusing the virtual CT image set with the real CT image, extracting the topological structure characteristics of pancreatic tissues of the fused image through the discrete Morse theory and outputting a pancreatic topological structure image; The clustering subdivision module is used for performing unsupervised clustering on the topological structure images by adopting a topological perception diffusion condensation method to distinguish pancreas from background areas, introducing a labeling sample to construct a semi-supervised loss function, optimizing a clustering center, subdividing edema, survival and necrosis tissues, and outputting a preliminary segmentation mask; The mechanical correction module is used for constructing a dynamic mechanical finite element analysis model, distributing differential mechanical parameters for each region of the preliminary segmentation mask according to the elastic modulus and Poisson ratio measured data of pancreatic tissues in different courses, simulating mechanical properties of the tissues, correcting boundary deformation and outputting an optimized mask; The pixel optimization module is used for carrying out fine correction on the pixels of the optimization mask by adopting an entropy value optimization threshold method, determining and distinguishing pseudocyst and peripancreatic exudate boundaries, and outputting an area segmentation mask; And the quantization output module is used for automatically calculating the necrosis volume ratio and the exudation maximum cross-sectional area according to the region segmentation mask and outputting a region segmentation result.

Description

Intelligent segmentation method and system for CT (computed tomography) image of acute pancreatitis Technical Field The invention belongs to the technical field of image segmentation analysis, and particularly relates to an intelligent segmentation method and system for CT images of acute pancreatitis. Background The intelligent segmentation of the CT image of the acute pancreatitis is a technology for automatically and finely dividing the abdomen enhanced CT image of the acute pancreatitis patient by combining a medical imaging technology, computer vision and an artificial intelligent algorithm, and has the core aims of accurately identifying pathological areas such as pancreas parenchyma, edema areas, necrosis areas, peripancreatic exudates, pseudocysts and the like, and providing a quantitative basis for disease classification and treatment scheme formulation. The traditional CT image segmentation method for acute pancreatitis has the defects that firstly, the model is trained by manually labeling samples, the labeling cost is high, the generalization capability is weak, and the segmentation precision of rare severe cases is greatly reduced. Only by means of the characteristics of the shallow layers such as gray scale and texture, the device is easy to be interfered by intestinal qi and blood vessels, and the gray scale areas such as edema and necrosis are difficult to distinguish. The mechanical deformation of tissues caused by pancreatic inflammation is not considered, and the deviation between the segmentation boundary and the real anatomical morphology is large. The pseudocyst and peripancreatic exudates cannot be distinguished accurately, and the two pathological areas are easily confused. The algorithm has complex flow and large calculation amount, and is difficult to be deployed to a clinical workstation to realize real-time segmentation. Disclosure of Invention In order to make up for the defects of the prior art, the invention provides an intelligent segmentation method for CT images of acute pancreatitis. The method is mainly used for solving the problem that the intelligent segmentation boundary of the traditional CT image for acute pancreatitis has large deviation from the true anatomical form. The intelligent segmentation method for the CT image of the acute pancreatitis comprises the following steps of S1, simulating a normal pancreas CT image according to physical fluid diffusion to generate a virtual lesion image set, and outputting the virtual CT image set. And S2, fusing the virtual CT image set with the real CT image, extracting topological structure features of pancreatic tissues of the fused image through a discrete Morse theory, and outputting a pancreatic topological structure image. And S3, performing unsupervised clustering on the topological structure image by adopting a topological perception diffusion agglomeration method, and distinguishing pancreas and background areas. And introducing a labeling sample to construct a semi-supervised loss function, optimizing a clustering center, subdividing edema, survival and necrosis tissues, and outputting a preliminary segmentation mask. S4, constructing a dynamic mechanical finite element analysis model, distributing differential mechanical parameters for each region of the preliminary segmentation mask according to the elastic modulus and Poisson' S ratio measured data of pancreatic tissues in different courses, simulating mechanical properties of the tissues, correcting boundary deformation, and outputting an optimized mask. And S5, carrying out refined correction on pixels of the optimized mask by adopting an entropy value optimization threshold method, determining and distinguishing boundary between pseudocyst and peripancreatic exudate, and outputting an area segmentation mask. And S6, automatically calculating the necrosis volume ratio and the exudation maximum cross-sectional area according to the region segmentation mask, and outputting a region segmentation result. According to the intelligent segmentation method for acute pancreatitis CT images provided by the invention, in step S1, the specific steps of outputting a virtual CT image set are as follows: And S11, acquiring a normal pancreas CT image and corresponding anatomical structure labels, extracting basic characteristics such as CT value distribution, contour morphology and the like of a pancreas parenchymal region, and outputting a normal pancreas image characteristic set. And S12, setting fluid diffusion parameters of edema, necrosis and exudation according to the pathological evolution rule of the acute pancreatitis, taking a normal pancreas image feature set as a diffusion initial field, simulating the fluid diffusion process of a pathological change region, and outputting a pathological change embryonic image. And S13, combining the disease course characteristics of the mild symptoms and the severe symptoms, adjusting the time step and the range of diffusion simula