A specific model of resting-state functional brain network in MRI-negative temporal lobe epilepsy.

Purpose: Without any visible indicator on structure magnetic resonance imaging (MRI), the diagnosis of MRI-negative temporal lobe epilepsy (NTLE) gets harder. By considering healthy control (HC), a specific functional connectivity (FC) model was constructed in a network topology to improve FC computation to a high-level. Methods: MRI data of 20 NTLE patients and 60 HC were pre-processed. Relative to HC, a network-level specific FC model of each network index was built to score the network functions for each NTLE patient. The specific brain areas (regarded as ROIs) were extracted for NTLE by sensitivity analysis of scores. By considering scores of specific ROIs as feature vectors to input into a SVM respectively, a specific NTLE classifier was constructed. Both 10-fold cross validation and hold-out method were utilized to validate the classification and to evaluate the effectiveness of our specific FC models. Simultaneously, the specific FC model was compared to the conventional FC model of Pearson correlation. Results: By the constructed model for specific FC at a network-level, 11 specific ROIs, such as, frontal lobe, temporal lobe, parietal lobe, hippocampus, and occipital lobe, were extracted for NTLE. Accuracy of our specific NTLE classifier could reach up nearly 93 %, over 6 % greater than conventional FC model of Pearson correlation. Conclusions: The network-level specific FC model might provide a new methodology for machine-aiding detection of functional abnormal lesions of NTLE by resting-state functional MRI.