A Study Of MRI-negative Epilepsy Foci Based On Quantitative Analysis Of Intelligent Cortex

Epilepsy is one of the craniocerebral nervous system diseases with a large number of patients in clinical medicine.Magnetic Resonance Imaging(MRI)is an important means of epilepsy diagnosis,but in fact,there are still some shortcomings:there is a large proportion of patients in clinical MRI cranio-brain examination show conventional “MRI negative”,that is,doctors can't find obvious abnormal structures.Using the very popular image processing technology to intelligently post-process MRI negative neuroimaging and integrating with machine learning technology to form an intelligent and objective form of auxiliary diagnosis can improve the efficiency and accuracy of clinicians in the diagnosis of epileptic brain diseases,which also will greatly promote the research of the brain subject with high application value.In this paper,the morphological characteristics and classification techniques of the craniocerebral cortex were studied by using conventional "MRI negative" epilepsy images.Firstly,we systematically studied the brain theoretical basis,including the structure of normal brain tissue and abnormal epilepsy tissue,as well as the imaging principle and scanning characteristics of MRI imaging technology.Secondly,an analysis method based on surface morphology(SBM)was used to extract and analyze the cranial tissue structure characteristics of MRI images,including preprocessing operations such as format conversion,standardization,segmentation and smoothing,as well as post-processing operations for quantifying cortical features of cortical thickness,groove depth,mean curvature and fractal dimensions.Then,after establishing the correlation coefficient matrix,it is found that there are many useless and redundant features between the data,so feature selection is needed.Principal Component Analysis(PCA)was mainly used and founded optimal features subset.Finally,single-kernel support vector machine(SVM)classifier and multi-kernel SVM classifier were used to classify the feature subset.The Grid search method and Particle Swarm Optimization algorithm were used to optimize the parameters of RBF kernel function,and the improved Particle Swarm Optimization algorithm was introduced to optimize the weight coefficients of Polynomial kernel function and RBF weight synthesis of multi-kernel SVM.The experimental results showed that the morphological characteristics of cortical surface thickness,groove depth and the average curvature can effectively detect epilepsy abnormality.The single-kernel SVM model achieved the classification accuracy of 71.4%-84.1%,with an average of 77.3%.The multi-kernel SVM classification model established by the weighted mapping of d=3's polynomial kernel function and RBF achieved a higher accuracy of 85.7% on the train set and 81.8% on the test set.In addition,the optimal feature subset corresponds to the brain region and medical t-test methods overlap with the parietal lobe,frontal gyrus,temporal gyrus and other significant brain regions where may be present lesions.