Epileptic focus localization using transfer learning on multi-modal EEG.

Area of Science:

• Neurology
• Medical Imaging
• Machine Learning

Background:

• Epilepsy affects millions, with a significant portion experiencing drug-resistant forms.
• Current treatments include drug therapy and surgical resection, but localizing the epileptic focus remains challenging for ~1/3 of patients.
• Accurate localization of the epileptic focus is crucial for effective surgical intervention.

Purpose of the Study:

• To develop and validate a novel transfer learning method for localizing drug-resistant epileptic foci.
• To improve the accuracy and efficiency of epileptic focus identification using multi-modal electroencephalography (EEG).
• To provide a tool that assists physicians in clinical diagnosis and treatment planning for epilepsy.

Main Methods:

• Proposed a transfer learning approach utilizing intracranial EEG (iEEG) and surface EEG (sEEG) data.
• Validated a pre-trained model using 10-fold cross-validation on the Bern-Barcelona and Bonn datasets.
• Fine-tuned the model with epilepsy data from Chongqing Medical University and tested using leave-one-out cross-validation.

Main Results:

• The pre-trained model achieved high accuracy rates of 94.50% (Bern-Barcelona) and 97.50% (Bonn).
• Demonstrated superior performance over state-of-the-art baselines in accuracy, sensitivity, and negative predictive value.
• Achieved an average accuracy of 90.15% on the Chongqing Medical University dataset, highlighting significant feature differences between epileptic and non-epileptic channels.

Conclusions:

• The proposed transfer learning method effectively localizes epileptic foci by accurately classifying epileptic and non-epileptic channels using neural networks.
• The model's superior performance indicates its potential as a valuable tool for clinical diagnosis of epilepsy.
• This technique can significantly aid physicians in the precise localization of epileptic foci, improving patient outcomes.