Detection And Analysis Of Morphological Features Of Cerebral Cortex Based On Tetrahedral Mesh Model

Alzheimer’s disease(AD)is a neurodegenerative disease,and changes in cognitive abilities caused by AD are closely related to changes in cerebral cortex morphology.In order to effectively intervene in the early stages of AD,it is necessary to accurately characterize and identify the changes in cerebral cortex morphology.The paper starts with the detection of morphological features of the cerebral cortex and uses MRI images to proposes an algorithm for estimating the thickness of the cerebral cortex based on the tetrahedral mesh model.Using parity rules and contraction strategies,the algorithm detects and repairs cross-overlapping regions generated by the FreeSurfer software segmentation of the cerebral cortex,and then constructs a tetrahedral mesh model that reflects the inherent morphological characteristics of brain MRI images.The Laplace equation under Dirichlet boundary conditions calculated by the finite element method,constructs the steady-state field distribution in the cerebral cortex.Using the constructed local isothermal surface strategy and linear geometry determine the direction of the gradient line,and then execution efficiency of the algorithm can be improved by the half-surface data storage structure.Finally,the thickness of the cerebral cortex is calculated accurately and effectively by the length of each gradient line,and the statistical methods can analyze the differences of cerebral cortical thickness feature among AD disease(110),mild cognitive impairment(MCI,101)and normal people(CTL,128).Compared with the FreeSurfer method and the method of measuring the thickness of the cerebral cortex based on the Laplace equation using cubic voxels,the proposed thickness feature measurement algorithm can accurately capture the thickness features of the cerebral cortex and effectively improve the accuracy of the thickness measurement,and has a strong statistical analysis ability in detecting regions with significant differences in the thickness feature of the cerebral cortex.Then,using the measured cortical thickness of AD,MCI and CTL,the paper proposes a feature selection method(kROI method)based on feature reconstruction and classification between groups.Statistical one-sided t-test and K-nearest neighbor method(KNN method)are used to obtain prior knowledge of classification,and regions of interest(ROI)with large thickness differences between groups are extracted.The linear discriminant analysis(LDA)algorithm is fused to construct the kROI-LDA fusion algorithm,which further improves the accuracy of discrimination and classification.Finally,the SVM is used to classify and predict the features that are features of the whole brain,features of kROI method selection,features of LDA dimension reduction based on the whole brain and features of kROI-LDA method construction.The results prove that the kROI-LDA method can effectively reduce the impact of redundant information and noise on the classification results and improve the classification accuracy between different groups.