AI-Enhanced BCI Model for Accurate Differentiation between Brain Tumor–Induced Seizures and Epileptic Seizures Using EEG Signatures

Abstract

Due to EEG overlap, brain tumour-induced seizures (BTIS) and primary epileptic seizures (ES) are difficult to distinguish. This paper introduces an AI-enhanced BCI model that uses hybrid deep learning architectures to differentiate accurately. We used CNNs for spatial feature extraction and Transformer-based attention mechanisms for temporal dependency modelling to decompose multi-channel EEG data using wavelet and Fourier methods. On a multi-centre dataset of 1,240 EEG sessions (500 subjects: 125 healthy controls, 187 epileptic, 188 tumour-induced seizure cases), our model had 96.7% sensitivity, 94.5% specificity, and 95.6% accuracy. The framework performed well across patient demographics and seizure types. Real-time lightweight architecture detection latency was 8.3 seconds with 0.62 false alarms per day. This study fills a clinical need by offering an automated, interpretable diagnostic tool that may improve seizure treatment and tumour-related neurological consequences.