TEG-Net: A Temporal Entropy-Guided Multimodal Deep Learning Framework for Early Alzheimer’s Disease Prediction Using EEG and MRI

Alzheimer disease (AD) is an irreversible neurodegenerative condition that severely affects the cognition, memory and normal functioning ensuring timely detection is important to provide effective intervention.This paper tackles the limitations of traditional AD diagnosis, including the inability to diagnose early, heterogeneity of data, and computational overheads of current Machine Learning (ML) and Deep Learning (DL) models.The main aim is to create an efficient, multimodal model, namely, Temporal Entropy Guided Multimodal Network (TEG-Net) to combine the EEG-based temporal entropy measurements with the MRI based structural dynamics.The suggested algorithm estimates Sample and Permutation Entropy of EEG signals to measure the loss in neural complexity, and a 3D CNN with a very small number of nodes recognizes the existence of spatial degeneration in MRI images.Entropy-based attention causes dynamic weighting of MRI features to allow adaptive multimodal fusion to enhance the classification of a healthy control, Mild Cognitive Impairment (MCI) and AD group.Experimental findings show that TEG-Net is more effective than single-modality and traditional multimodal methodology with 96.4 percent accuracy, and 0.98 AUC, and especially higher sensitivity in early MCI detection.The authors find that adaptive fusion driven by entropy improves biologically meaningful crossmodal interactions, generalization, and allows computationally efficient, clinically relevant prediction of AD, which leads to application of adaptive fusion to real-time diagnostic systems and the incorporation of other modalities like PET imaging in the future.