A Bioinspired Multimodal CNN-LSTM Network for EEG Analysis of Patients in Coma

Electroencephalography (EEG) is widely used for diagnosis and evaluation of neurological diseases, despite challenges from its high-dimensional and noisy temporal data, which complicate accurate brain signal classification. This study proposes a multimodal deep learning model combining Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) layers to classify EEG signals, integrating patient information as additional modalities. CNN layers effectively extract spatial features and reduce EEG data dimensionality, while LSTM layers capture long-term temporal dependencies. A Genetic Algorithm (GA) selects relevant multimodal features and optimizes CNN-LSTM hyperparameters. The model was applied to outcome classification in comatose patients, achieving improved classifier performance compared to unimodal approaches. Experimental results demonstrate that multimodal integration and GA optimization significantly enhance accuracy, robustness, and generalization. The architecture shows promise for broader EEG classification tasks, potentially advancing clinical decision support based on EEG signals.