EEG Seizure Detection using Convolutional Neural Network With Grad Cam

The project’s main goal is to employ deep learning techniques to analyze EEG readings and detect seizure episodes. EEG data is first preprocessed and split into sets for training and validation. Next, a Convolutional Neural Network (CNN) model with several convolutional and dense layers is built. Using an Adam optimizer, binary cross-entropy loss is optimized as part of the model’s training process. Furthermore, classification reports and accuracy metrics are used to assess the performance of the model. Weight extraction from the first convolutional layer is used to visualize feature importance for interpretability. Grad-CAM heatmaps are produced to identify the EEG signal regions that are most important for seizure prediction. This method helps detect pertinent EEG patterns linked to seizures and comprehend the model’s decision-making process. All things considered, the experiment shows how well deep learning works for EEG-based seizure identification and offers insights into the significance of features using Grad-CAM visualization approaches. The method has potential for real-time seizure monitoring and may aid in the creation of assistive devices for the treatment of epilepsy.