Epilepsy is a chronic disease of the nervous system.Epileptic seizures can cause serious cognitive behavior and functional disorders,threatening the physical and mental health and quality of life of patients.Focal cortical dysplasia(FCD)is an important cause of epilepsy and the most common one of drug-refractory epilepsy(DRE).Currently,epilepsy surgery is a suitable option for the FCD treatment.By removing the lesion area of FCD,seizures frequency can be reduced and patients can even be seizure-free.Therefore,accurate detection and localization of FCD lesions in preoperative evaluation is of great significance,which affects surgical decisions and implantation strategies of intracranial electroencephalography,improving the therapeutic effect.Traditional FCD detection methods are mostly based on conventional visual review.Recently morphological analysis has been gradually applied to automatic detection of FCD while multimodal images reflecting different information in clinical diagnosis of brain diseases.Magnetic resonance imaging(MRI)is often used in the diagnosis and treatment of FCD to observe the structure of the brain and positron emission computed tomography is gradually applied in the diagnosis of FCD because it can reflect the brain metabolism.Therefore,this paper studied on the detection of FCD abnormal signals automatically based on the morphology analysis and multimodal images.The details are as follows:Firstly,the medical imaging pre-processing procedure was conducted through SPM12 and FreeSurfer.Because of the easy occurrence in whole-brain and the difficulty in lesion boundary,this paper proposed an idea of FCD detection based on multiscale analysis,which was divided into the brain-wise detection,hemi-wise detection,lobe-wise detection and vertex-wise detection from coarse to fine.Secondly,in our research on the FCD coarsely-detection,the combination of Region of Interest(ROI)analysis and morphometry analysis was firstly used to extract features.Then,a feature selection algorithm combined filtering method and packaging method was proposed in this paper to avoid possibly adverse effects on model performance due to high-dimensional features.The filtering method is the Pearson Correlation Coefficient Analysis(PCCAs)and Max-relevance and min-redundancy(mRMR)and the packaging method Recursive Feature Elimination(RFE).Finally,the universality and effectiveness of the proposed method were verified from three aspects:the different task scenario,different feature selection methods,different classifiers.Then,to improve the detection results,the FCD finely-detection was studied furtherly at vertex-wise level.The functional modality PET was introduced in this part and the multi-modal feature extraction method was designed.Then a fine-detection FCD model was constructed using the four-layer neural network(ANN),and the results of model classification were postprocessed to obtain the visualization results of lesion detection,realizing the coarse segmentation of FCD lesion.Finally,the feasibility of the proposed method was verified by comparing the performance of fine-detection model constructed by different modal feature combinations.The proposed method was compared with traditional visual examination method,voxel-based Morphometry(VBM)and deep learning method to verify the effectiveness of the proposed method.To sum up,in this paper,the feasibility of combining morphological analysis and ROI analysis was verified by the study of coarsely FCD detection method,and the effectiveness of the proposed hybrid feature extraction algorithm was also verified.The necessity and effectiveness of multimodal image for FCD detection are verified by the study of fine FCD detection method,the FCD detection method at vertex-wise level is verified.Moreover,our paper studied the FCD detection problem based on multiscale analysis,which implemented the gradual refinement from patient detection to lesion localization.The FCD detection at the vertex-wise combined with post-processing method can describe the lesion boundary of FCD,which has a certain auxiliary role in preoperative evaluation of FCD. |