Prediction of Epilepsy Based on Tensor Decomposition and Functional Brain Network.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Informatics

Background:

• Epilepsy affects 65 million globally, necessitating advanced diagnostic tools.
• Network-based analyses aid seizure investigation, but struggle with dynamic brain networks.
• Traditional graph theory methods have limitations in capturing the temporal dynamics of functional brain networks.

Purpose of the Study:

• To present a novel approach for epilepsy prediction by analyzing dynamic functional brain networks.
• To introduce tensor decomposition as a feature extraction method for dynamic brain networks.
• To evaluate the efficacy of tensor decomposition features against traditional graph metrics in epilepsy prediction.

Main Methods:

• Dynamic functional brain networks were constructed by time-axis stacking.
• Tensor decomposition was employed to extract salient features from these dynamic networks.
• An Extreme Learning Machine (ELM) classifier was utilized for epilepsy prediction.

Main Results:

• Features extracted via tensor decomposition demonstrated superior performance in epilepsy prediction.
• Accuracy and F1 scores were significantly higher using tensor decomposition features compared to degree and clustering coefficients.
• The proposed method offers a more comprehensive analysis of dynamic functional brain networks for neurological disorder prediction.

Conclusions:

• Tensor decomposition is an effective method for extracting features from dynamic functional brain networks.
• This approach enhances the accuracy and comprehensiveness of epilepsy prediction.
• The findings suggest a promising direction for the neuroimaging analysis of neurological disorders.