Attention-based Deep Learning for Epileptic Seizure Type Detection

The majority of the research studies surrounded by detection of seizures rather than discriminating seizure types using EEG. Identifying specific seizure types is crucial for diagnosis and treatment. EEG signal variability among seizure types presents a challenge, but deep neural networks, especially with attention mechanisms, show promise for accurate analysis. In this work, a well-recognized recurrent network, Long Short-Term Memory lined with self-attention mechanism has been designed to categorize four types of seizure, including focal non-specific, generalized non-specific, simple partial, and complex partial seizures. For validation, the Temple University Hospital EEG database has been used. The idea has attained classification accuracy, specificity, precision, and F1-score of 92.2%, 97.5%, 92.1 %, and 92.2% respectively. Additionally, it has displayed notable area under the ROC. In a comparative study, this method has demonstrated superior performance.