[Multi-task learning with sleep features for interictal epileptiform discharge detection: a model development and...

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Context:

• Epilepsy diagnosis relies heavily on interpreting electroencephalogram (EEG) data, specifically identifying interictal epileptiform discharges (IED).
• Current automated IED detection methods often lack precision and can be improved by integrating additional relevant features.
• Sleep patterns within EEG signals contain valuable information that can aid in IED detection.

Purpose:

• To develop and validate an automated detection model for interictal epileptiform discharges (IED) using a multi-task learning algorithm.
• To integrate sleep features into the detection model to enhance the precision of electroencephalogram (EEG) interpretation.
• To provide a more robust tool for clinical practice in epilepsy diagnosis.

Summary:

• A multi-task learning model, Siamese-ES, was developed using convolutional neural networks, integrating deep sleep features with IED features extracted from EEG recordings.
• The model was trained and validated on a dataset of 265,551 EEG samples from 150 patients, comparing its performance against classic single-task models.
• Ablation experiments confirmed the effectiveness of integrating sleep features and the multi-task learning approach, showing improved precision, specificity, F1 score, and AUC.

Impact:

• The Siamese-ES model demonstrates significantly improved performance in IED detection, offering more precise EEG interpretation support.
• This advancement can lead to more accurate and efficient epilepsy diagnosis in clinical settings.
• The study provides a foundation for future research into IED detection models tailored for diverse clinical scenarios.