Deep Learning Brain MRI Segmentation and 3D Reconstruction: Evaluation of Hippocampal Atrophy in Mesial Temporal Lobe Epilepsy

In the present paper, we present an automated approach to analyse the lateralization of the epileptogenic focus within the mesial temporal lobe (mTLE) using brain MRI scans. The proposed method encompasses an initial segmentation stage that utilizes a deep convolutional neural network (CNN), followed by a 3D reconstruction and volume computation for both the right and left hippocampus. Our comprehensive approach involves preprocessing the database, employing augmentation techniques, and evaluating outcomes using standard metrics. We validated our method using a dataset containing MRI images from 50 patients, resulting in encouraging findings. Specifically, our approach achieved an 0.84 Dice score, surpassing the values previously reported for this dataset in the literature. Moreover, our method attained an 89% sensitivity and a Hausdorff distance of 2.59.