TLENet-Transformer for Epilepsy Detection with Futuristic IoMT Plans

Temporal Lobe Epilepsy (TLE) is the most prevalent subtype of focal epilepsy, often posing diagnostic difficulties due to its complex clinical features. To address these challenges, this study proposes TLENet-Transformer, a novel deep learning architecture tailored for precise TLE identification. The study begins with skull stripping, bias field correction, z-score normalization, and Gaussian filtering; these methods are implemented to enhance image quality. Following this, a segmentation method based on an optimized Fuzzy C-Means clustering algorithm is employed, providing accurate delineation of regions. At this point, the segmented images are subsequently mapped to a pre-labelled Automated Anatomical Labelling (AAL) template to detect distinct segmented ROIs. Developing hybrid models that combine DL with Transformer and attention-based techniques could improve feature extraction. The classification module integrates a hybrid deep learning approach, combining 3D Convolutional Neural Networks for spatial feature extraction, Spatio-Temporal Long Short-Term Memory units for capturing temporal dynamics, and Capsule Networks to preserve hierarchical features. This multi-layered architecture allows the model to learn complex patterns associated with affected regions. TLENet-Transformer gives a high diagnostic accuracy of 98.16%, outperforming conventional models in both sensitivity and specificity. Furthermore, the study explores the integration of Internet of Medical Things (IoMT) technologies to enable real-time remote monitoring as the future scope of work.