Enhancing EEG signal analysis with geometry invariants for multichannel fusion

Automated computer-aided diagnosis (CAD) has become an essential approach in the early detection of health issues. One of the significant benefits of this approach is high accuracy and low computational complexity without sacrificing model performance. Electroencephalogram (EEG) signals with seizure detection are one of the critical areas where CAD systems have been developed. In this study, we proposed a CAD system for seizure detection that prioritizes optimizing the solution's complexity. The proposed approach combines geometry invariants multi-channel fusion and amplitude normalization for input data preparation, and experiments on the frequency domain and CNN architecture for reducing complexity. Furthermore, the study includes explainability experiments that should aim to interpret not only the performance of the model but also the analysis of the patterns that contributed to the obtained results. The results demonstrate the effectiveness of the proposed model and its suitability for decision support in both clinical and home environments.