Speeded up robust features trailed GCN for seizure identification during pregnancy

Abstract

In this work, an efficient computational framework has been designed for seizure identification using MRI analysis. The inputs being brain MRI of pregnant women and corresponding outputs being the seizure or no seizure label. The framework is implemented in two phases. First, the informative speeded up robust features (SURF) are extracted from the MRI. Second, these features are further mapped to a graph convolutional neural network (GCN). The maximal clique is generated out of these intermediate features and subjected to convolutional neural network (CNN) architecture for classification. The maximal clique acts as an efficient tool for representing final and fine-tuned feature points through combined graph convolution and thus contributes towards efficient classification. The designed framework is validated through benchmark dataset images presented by NITRC. Experimental evaluation is made on samples of ‘male’, ‘female’ and ‘female with pregnancy’. The overall rate of accuracy stands at 96%, 95%, and 95% respectively.

URN:NBN:sciencein.jist.2024.v12.810