The Study Of Temporal Lobe Epilepsy Classification Based On Neuropsychology Tests And Mri

Temporal lobe epilepsy(TLE)is the most common form of drug-resistant epilepsy,with an incidence of more than 50% among patients with epilepsy.Epilepsy surgery is the most effective treatment to control seizure for patients with drug-resistant epilepsy.Successful surgical outcome depends on accurate delineation of the epileptogenic zone and localization of the vital eloquent cortex.As the etiology,pathogenesis and clinical symptoms are complicated,diagnosis of TLE and epileptogenic zone localization need to be accomplished with comprehensive evaluations of different diagnostic tools.Neuropsychology tests and magnetic resonance imaging(MRI)are the common and valuable methods for TLE diagnosis in clinical practice.However,currently the interpretation of neuroimaging and neuropsychology relies immensely on clinicians’ experience,which is complicated,subjective and time-consuming.In order to assist doctors in presurgical evaluation,lower the burden of clinicians,improve long-term postoperative seizure-free rate,and improve the quality of patients’ life,here,we proposed to use machine learning algorithms with the neuropsychological data and MRI to classify TLE,and to explore the relationship between MRI and the neuropsychological tests.42 epilepsy patients underwent neuropsychological tests and MRI scans before surgery,including 23 with TLE(aged 33.4±2.2 years)and 19 with extratemporal lobe epilepsy(extraTLE)(aged 31.4±2.0 years).Firstly,machine learning approaches with the neuropsychological tests were employed to classify TLE,then the neuropsychology tests with highest frequency in the training sets were selected as the significant neurologic signatures for the diagnosis of TLE.Secondly,generalized linear model was used to analyze the brain region differences between TLE and extraTLE,and stepwise multiple regression was performed to analyze the relationship between brain alterations and the neuropsychological status.Thirdly,we applied machine learning algorithms on the neuropsychological data and MRI to explore whether machine learning with multimodel features could improve the accuracy of TLE classification.The results showed that:(1)The accuracy of classification using support vector machine(SVM),logistic regression and random forest algorithm was 75.5%,81.0% and 80.3% respectively,and the area under the receiver operating characteristic(ROC)curve(AUC)was 0.81 and 0.79 and 0.84 respectively.No significant differences were detected when the AUC of logistic regression,random forest compared with the AUC of SVM.(2)13 neuropsychological tests were selected as the significant neuropsychological signatures for the diagnosis of TLE,including arithmetic test,digital span and digital symbol in the Wechsler Adult Intelligence Scale,immediate recall,delayed recall and maximum capacity in the abstract figure learning tests,L or R orientation test,selective attention test,category verbal fluency testing,component verbal fluency testing,and single word test,single color test and double colors test in the Stroop test.(3)Gray matter volume was decreased in the parahippocampal and fusiform of TLE patients compared with the extraTLE.Significant abnormalities of cortical surface features such as cortical thickness,sulcal depth and mean curvature between two groups were widespread and disperse,which were found in the region of temporal lobe,frontal lobe and the parietal lobe.(4)Differences in memory functions between two groups were associated with abnormalities in the left precuneus and bankssts gyrus.Differences in working memory functions and executive functions were associated with abnormalities in the region of parietal lobe,frontal lobe and temporal lobe.And differences in visuoperceptual skills were associated with abnormalities in the left superiortemporal gyrus.(5)When neuropsychological test features and the gray matter volume features were combined to classify TLE,the accuracy and AUC were 79.6%~85.7%,0.82~0.88,respectively,higher than the classification result when neuropsychological tests were the only feature.Taken together,using machine learning methods with the neuropsychological data and MRI for TLE classification can achieve higher accuracy compared to the similar studies as ever reported.In addition,explaining the mechanism of cognitive impairment using neuroimaging information could assist clinicans in presurgical evaluation of TLE to a certain extent.