Propagation mapping using iterative independent component analysis for seizure onset zone localization in temporal...

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Epilepsy impacts 70 million globally, with one-third experiencing drug-resistant forms requiring surgery.
• Accurate seizure onset zone (SOZ) localization is critical for surgical success but remains a significant challenge.

Purpose of the Study:

• To develop and validate a novel, data-driven method for precisely mapping seizure propagation and localizing the SOZ in drug-resistant temporal lobe epilepsy (TLE).

Main Methods:

• Proposed an iterative independent component analysis (ICA) approach to map seizure propagation patterns.
• Applied machine learning models to spatial propagation profiles derived from ICA to distinguish SOZ.
• Compared the performance of the ICA-based method against traditional iEEG source imaging.

Main Results:

• Seizure propagation features significantly differentiated SOZ from non-SOZ contacts in seizure-free patients (AUC = 0.85), outperforming iEEG source imaging (AUC = 0.73).
• The ICA method demonstrated robust accuracy (AUC = 0.84) in mesial TLE, where iEEG source imaging performance decreased significantly (AUC = 0.64).

Conclusions:

• Introduced a data-driven strategy for characterizing seizure propagation, offering a potential advancement in SOZ localization, particularly for deep brain origins.
• The iterative ICA method provides an adaptive, purely data-driven alternative to expert-reliant traditional techniques.