Research On Automatic Epilepsy Detection Based On PET/CT Brain Images

PET/CT plays an important role in the diagnosis of epilepsy.However,the recognition of the epilepsy lesions mainly relies on the clinicians,who needs rich clinical experiences,high professional level and a lot of time to reading the PET/CT images.Based on biomedical theor y and computer image processing technology,the research is aimed to automatic detect the epi leptic on PET/CT improving the accuracy in diagnosis of epilepsy and reducing the doctor rea ding time.Therefore,this paper has important researching significance and application value.Based on the unique feature in PET/CT and pathological characteristics of epilepsy,this paper efficiently and accurately detects the epilepsy lesions.For the innovation,the quantum space half brain symmetry feature recognition method is used to detect the whole brain PET images of epilepsy lesions.The method includes the following four parts.(1)Calculating all the voxel in SUV and setting up the third order tensor;(2)Extracting half brain symmetry feature and building the half brain symmetry tensor model;(3)Using the method of multiple linear principal component analysis(MPCA)to select the features of half brain symmetry tensor model;(4)Based on the support vector machine(SVM)classifier detecting the epilepsy.For the accurate position of epilepsy lesions,the research detects the ROI of focal epilepsy.Firstly,based on gray level co-occurrence matrix and neighborhood grayscale color difference matrix,extracting a variety of texture feature on each ROI;Secondly through SFFS+KNN feature selection method,selecting the optimal feature sets;Using the DT-KNN classification algorithm detecting the epilepsy.For the above algorithms,this paper has carried on the experimental verification and the analysis of contrast.In the whole brain PET image epilepsy detection experiments,the accuracy is significantly higher than other traditional epilepsy detection method.In the ROI epilepsy detection experiments,DT-KNN classification algorithm is significantly higher than other classification algorithms in classification accuracy.In the end,the paper further developed the visualization software for the epilepsy automatic detection for the clinicians.