Deep Learning Approach for Epileptic Focus Localization.

Area of Science:

• Neuroscience
• Medical Engineering
• Artificial Intelligence

Background:

• Epileptic focus localization is vital for successful epilepsy surgery, relying heavily on intracranial electroencephalogram (iEEG) data.
• Accurate characterization of the epileptogenic zone is necessary for surgical removal of seizure-causing regions.
• Current methods often require manual feature extraction from non-stationary iEEG recordings.

Purpose of the Study:

• To propose two novel deep learning-based methods for accurate automatic epileptic focus localization.
• To automate feature extraction and classification processes for iEEG signals.
• To enhance the precision of identifying the seizure source for surgical planning.

Main Methods:

• A semi-supervised learning approach using a deep convolutional autoencoder and a multi-layer perceptron classifier.
• An unsupervised learning approach combining a deep convolutional variational autoencoder with K-means clustering.
• Implementation of the semi-supervised model's inference network on Field-Programmable Gate Array (FPGA).

Main Results:

• Both proposed methods demonstrated high classification accuracy in localizing the epileptic focus.
• The unsupervised clustering method effectively separated iEEG signals based on seizure source.
• The deep learning approaches automated feature extraction and dimensionality reduction.

Conclusions:

• The developed deep learning methods offer accurate and automated solutions for epileptic focus localization.
• These techniques can significantly aid clinicians in surgical decision-making for epilepsy patients.
• The study highlights the potential of AI in improving epilepsy treatment outcomes.