Research On Three-Dimensional Modeling Method Of Medical Image Based On Statistical Shape Model

With the rapid increase in the number of medical images,the deep interpretation and modeling of medical images have become an urgent problem to be solved in modern medicine.Traditional X-ray computed tomography(CT)and magnetic resonance imaging(MRI)can capture a series of two-dimensional(2D)images of the patient's location.Doctors can judge the characteristics and location of the patient's position by observing these 2D images.However,it is difficult to accurately diagnose the patient's condition only by observing 2D images,and the diagnosis process mainly depends on the doctor's personal experience,thus there is a certain risk of misdiagnosis.The three-dimensional(3D)model of the patient's location can better visualize the data,and help doctors directly interact with the 3D model through the computer.At the same time,the qualitative and quantitative analysis of medical image data can effectively reduce the risk of misdiagnosis and provide a powerful basis for doctors to formulate subsequent treatment plans.Therefore,it has become the core issue of medical image processing to construct the corresponding 3D model according to the 2D plane image of the patient.Aiming at the problems existing in the construction of 3D model,such as segmentation region of interest,the establishment of correspondence between template specimen and target specimens,the fitting of 3D model and test specimens,and the construction of 3D prediction model based on defect specimens,this dissertation takes statistical shape model(SSM)as the basis,analyzes the characteristics of different methods to construct 3D model,and puts forward corresponding improvement methods,so as to realize 3D modeling of medical images.The main research contents are as follows:(1)Aiming at the problem that the region of interest can not be accurately segmented from the target image in the process of medical image modeling,a medical image segmentation method based on random forest regression-statistical shape model(RFR-SSM)is proposed.Firstly,random forest regression is used to train the input training specimens,and split at the root node of each training specimens.When the number of specimens contained in all leaf nodes is less than the predefined observation value,the split stops;Then the decision tree is established according to different classification features,and the output of multiple decision trees is weighted to generate the prediction function;Finally,the training specimens in different positions and directions in the reference coordinate system are iteratively optimized,and the optimized training specimens are converted back to the image coordinate system to complete the segmentation region of interest in the target image.(2)Aiming at the problem that the correspondence relationship between the template specimen and the target specimens in the process of constructing the anatomical structure model of patients' s location,a 3D anatomical structure model of the patient automatically constructed based on statistical shape model(3DASMPACB-SSM)method is proposed.Firstly,the input data is preprocessed to enhance the region of interest,and the region of interest is segmented from the CT scan image of the patient by RFR-SSM;Then,the segmented region of interest is used as the training specimens,and the training specimens are meshed by triangles.At the same time,the vertices contraction strategy is introduced to iteratively contract the triangle mesh,which effectively reduces the number of triangle mesh vertices in the training specimens without distortion of the training specimens;Finally,b-spline free form deformation(BSFFD)is used to establish the correspondence relationship between the template specimen and the target specimens,and the mean model and deformation model of the anatomical structure is generated by principal component analysis(PCA).3DASMPACB-SSM can effectively construct a 3D anatomical structure model of the patient's location,and evaluate the performance of the 3D anatomical structure model through four evaluation indicators of compactness,specificity,generality and representation.(3)In order to solve the problem of poor fitting effect between the 3D anatomical structure model of patients and the test specimens,a 3D model fitting based on point distribution model(3DMFB-PDM)is proposed.Firstly,the training specimens set is processed,and the template specimen is aligned with the target specimens by procrustes analysis(PA)to reduce the adverse effects of training specimens due to rotation and scale changes;Then the correspondence relationship between the training specimens is established,and the point distribution model of the patient's location is expressed by normal distribution according to the correspondence relationship;Finally,the distance between the feature points in the point distribution model and the corresponding points in the test specimens is calculated.Mahalanobis distance(MD)is introduced as an additional item into the objective function,and the nonlinear equations are transformed into linear equations to minimize the distance between of them.At the same time,the model parameters of the point distribution model are constantly adjusted according to the minimum distance,so that the point distribution model and the test specimens have the minimum fitting error.3DMFB-PDM can effectively calculate the minimum fitting error between the point distribution model and the test specimens.Furthermore,the maximum error,mean error and root mean square error are used as evaluation indicators to measure the fitting degree.(4)In order to solve the problem of data defect in the process of constructing the prediction model of patient's location,a 3D anatomical structure prediction model based on defect specimens(3DASPMB-DS)method is proposed.Firstly,the input dataset is segmented by RFR-SSM,and the segmented region of interest is used as the training specimens set;Secondly,the training specimens set is divided into healthy specimens set and defect specimens set,and the possible defects of specimens in defect specimens set in all cases are manually simulated;Then the mean specimen of healthy specimens is used as template specimen,and all defect specimens are used as target specimens.The template specimen is aligned with the target specimens by iterative closest point(ICP).Finally,the correspondence relationship between the template specimen and the target specimens is established,and the prediction model of the patient's location is generated according to the correspondence relationship.The fitting error between the prediction model and the test specimens is evaluated by similarity measure function.3DASPMB-DS can effectively construct the prediction model of patient's location according to the defect specimens,and the performance of the prediction model is evaluated by compactness,specificity and generality.