Brain Image Pattern Recognition In Epilepsy Based On Individualized Functional Brain Atlas

Individual variability in human brain functional activity is a considerable factor that linking to the distribution and location of individual-specific functional network.Restingstate functional magnetic resonance imaging(rfMRI)is a suited measure to study the neural basis of neurological disorders and psychosis with excellent spatial resolution and considerable time specification in a noninvasive manner.With the aid of rfMRI,the human brain was parcellated to different community which account different cognitive function.Based on group-level atlas,neuroimaging research focused on epilepsy contributed significant knowledge of the aberrant functional network architecture in epileptic brain.But these research methods ignored the existence of individual variability resulting to obscure details.Recently proposed advanced processing method parcellated the individual-specific function area with considering the difference in functional architecture between each brain.These pipelines yield personalized functional atlas which partition the cerebral cortex according to individual's unique connectivity profile.Some researches about schizophrenia and autism already took advantage of the strength of the new technique,whereas none study devoted to epilepsy.Hence this dissertation focused on epilepsy imaging pattern recognition based on individualized function parcellation.The work introduced in this dissertation mainly includes the following two parts:Based on the individual-level functional network parcellation,using an iteration algorithm,and taking into account the variability of network distribution among individuals,we identified the brain parcellations with consistent functions after the algorithm converges.Then we utilized machine learning technique to obtain the key feature from functional connectivity metric.We found the individualized pipeline yield more robust classification model and better performance.This section we utilized a different process strategy namely personalized intrinsic network topology(PINT)which only modified the location of the region of interest(ROI)with a fixed amount.We enrolled a group of genetic(idiopathic)generalized epilepsies with generalized tonic-clonic seizure(GGE-GTCS)patients and a group of matched healthy controls.Both groups underwent the whole process pipeline and obtained the personalized ROIs.Based on these individual-specific locations we extracted connectivity features and input to classification model to identify the epilepsy patients from others.The personalized process strategy has better performance and reliability compared to group-level analyses.