dataset_Deep Learning for the Diagnosis of Mesial Temporal Lobe Epilepsy

Two or three continuous coronal sections of T2WI at the level of the anterior commissure from patients with MTLE and those with other diseases such as psychogenic nonepileptic seizures (control subjects) were prepared for AI analysis. MR images were allocated to two datasets: learning and test datasets. The learning data were used for deep learning, whereas the test data were used for analyzing the diagnostic accuracy (i.e., validation). The test data included 30% of all data. MRI data of patients with MTLE with the hippocampus as the epileptogenic area included four images on T2WI (25.0%) and six images on FLAIR (21.4%). MRI data of control subjects included 77 images on T2WI (36.5%) and 77 images on FLAIR (27.7%). After the deep learning process, the diagnostic accuracy was calculated using the test data from T2-weighted images and FLAIR images.