Efficient Detection of Focal Cortical Dysplasia Using Novel Two Fold Attention Mechanism

Abstract

Abstract Focal Cortical Dysplasia (FCD) is a malformation of cortical development that leads to frequent pharmacological pediatric epilepsy. The only treatment for FCD is surgery, and the use of imaging techniques helps physicians plan the surgery. Magnetic Resonance Imaging (MRI) is an effective tool to predict the FCD lesion. The automatic segmentation of FCD lesions technique may help locate the lesions in the patient’s MRI slices. This research work proposes a novel two fold attention mechanism namely Hybrid Attention Gate_U shaped encoder decoder Network (HAG_UNET) model to detect the FCD accurately. The proposed model exploiting its novel attention mechanism is effective in accurate FCD lesion detection. The model tends to identify crucial features for FCD detection using the proposed novel two fold attention mechanism compared to state of art model. Experiment are done in python using standard datasets. A total of 11 subjects are used for the experiment. Metrics, namely IOU, precision, recall, and F1_score, are used for evaluation. Compared to UNET, the proposed model showed 5.89%, 4.92% and 3.15% improvement in terms of IOU, Recall and F1_score respectively. Compared to Attention_UNET, the proposed model showed 5.03%, 4.9%, 1.34% and 2.1% improvements in terms of IOU, Recall, Precision and F1_score respectively.