Early Epilepsy Diagnosis using Epistemic Neural Networks with Electric Eel Foraging Optimization and Secure EEG Data Sharing via Blockchain

Epilepsy is still quite common, affecting over 50 million individuals worldwide, despite the tremendous advances in contemporary therapy. Early detection and treatment can greatly lower the chance of long-term harm and raise living standards for those with epilepsy. Electroencephalography (EEG), a non-invasive method that tracks electrical activity in the brain, is frequently used to diagnose this neurological disorder. Yet, using EEG data for diagnosis and study raises a number of issues with patient privacy and security protocols for such private digital data. An early epilepsy diagnostic system and a decentralized, secure environment for exchanging EEG data based on a block chain-based system with DL integration are presented in this work. The proposed method makes it possible to maintain patients’ personal data confidentiality and, at the same time, increase the efficiency of early epilepsy diagnosis. The most frequently used preprocessing technique in the strategy is Multiple Discrete Orthonormal S-Transforms (MDOSTs), and for feature extraction, Memory Efficient Vision Transformer-based Feature Extraction is employed. In classification, we use an Epistemic Neural Network (ENN) to achieve high precision and reliable prediction through the Electric Eel Foraging Optimization (EEFO) algorithm. The storage and dissemination of the data are both proactively secured through blockchain; through smart contracts, patients completely govern their data. The analysis of the Temple University Hospital EEG Corpus (TUH EEG) dataset proves that this method is diagnosed more than the traditional one, while the protection level of this data is also high. By achieving above 99% performance on all evaluation measures, such as F1-score, accuracy, recall, specificity, and precision, the suggested method demonstrates its remarkable efficacy in both secure EEG data sharing and early epilepsy detection. The method's higher performance over conventional approaches can be attributed to its strong deep learning models and block chain integration.