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

Area of Science:

• Neuroscience and Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Electroencephalography (EEG) is crucial for neurological disease diagnosis but faces challenges due to high-dimensional, noisy data.
• Accurate brain signal classification is complex, hindering effective clinical application of EEG.

Purpose of the Study:

• To develop a multimodal deep learning model integrating EEG with patient data for enhanced signal classification.
• To improve the accuracy and robustness of EEG-based neurological outcome prediction.

Main Methods:

• A hybrid deep learning architecture combining Convolutional Neural Network (CNN) for spatial feature extraction and Long Short-Term Memory (LSTM) for temporal dependency analysis.
• Integration of patient information as an additional data modality.
• Utilization of a Genetic Algorithm (GA) for feature selection and hyperparameter optimization of the CNN-LSTM model.

Main Results:

• The multimodal CNN-LSTM model demonstrated superior performance in classifying EEG signals compared to unimodal approaches.
• Significant improvements in accuracy, robustness, and generalization were observed with multimodal integration and GA optimization.
• Successful application to outcome classification in comatose patients, showing enhanced predictive power.

Conclusions:

• Multimodal deep learning, enhanced by GA optimization, offers a promising approach for accurate EEG signal classification.
• This architecture holds potential for advancing clinical decision support systems in neurology.
• The model's effectiveness suggests broader applicability to various EEG-based diagnostic and evaluative tasks.