Neural Memory Networks for Seizure Type Classification.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Artificial Intelligence in Medicine

Background:

• Accurate seizure type classification is crucial for epilepsy diagnosis, treatment, and research.
• Current methods rely on expert epileptologists, highlighting a need for automated solutions.
• Machine learning and deep learning show promise but lack a definitive automated classification method.

Purpose of the Study:

• To introduce a novel approach for automated seizure type classification using electrophysiological data.
• To enhance traditional deep learning architectures with neural memory networks (NMNs).
• To evaluate the performance of NMNs against existing deep learning techniques for seizure classification.

Main Methods:

• Exploration of traditional deep learning models (CNNs, RNNs).
• Enhancement of deep learning architectures with external memory modules and trainable neural plasticity.
• Utilizing the TUH EEG Seizure Corpus with IBM TUSZ preprocessed data for evaluation.

Main Results:

• The proposed neural memory network model achieved a state-of-the-art weighted F1 score of 0.945.
• Demonstrated superior performance compared to traditional deep learning techniques for seizure type classification.
• Successfully classified seizure types using electroencephalogram (EEG) data.

Conclusions:

• Neural memory networks offer a powerful tool for automated seizure type classification.
• This approach has significant potential to advance epilepsy research and clinical practice.
• The methodology can be broadly applied to biomedical research and signal analysis problems.