Machine learning localization to identify the epileptogenic side in mesial temporal lobe epilepsy

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence

Background:

• Mesial temporal sclerosis (MTS) is the primary cause of drug-resistant mesial temporal lobe epilepsy (mTLE) in adults.
• Standard MRI protocols may miss subtle hippocampal changes in MTS, necessitating advanced quantitative methods like T1-weighted MRI.
• Machine learning (ML) offers a promising approach to enhance lesion detection and lateralization in mTLE.

Purpose of the Study:

• To apply ML techniques to MRI data for improved accuracy in lateralizing mTLE.
• To investigate the utility of quantitative MRI analysis, specifically T1-weighted axial series, in detecting subtle MTS.
• To enhance diagnostic capabilities for surgical planning in mTLE patients.

Main Methods:

• Utilized high-resolution T1-weighted axial brain MRI scans from 104 mTLE patients.
• Employed FreeSurfer software to extract features and compute lateralization indices (LI) for key brain regions.
• Trained an ML model using LI values from selected regions of interest (ROIs) for lesion lateralization.

Main Results:

• The ML model achieved a high test accuracy of 92.38% for mTLE lateralization.
• The model demonstrated a strong performance with an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.97.
• Successfully demonstrated ML's efficacy in detecting MTS from thin-slice T1 images.

Conclusions:

• ML effectively aids in detecting mesial temporal sclerosis (MTS) from MRI, crucial for drug-resistant mesial temporal lobe epilepsy (mTLE).
• The developed ML model shows significant potential for improving surgical planning and treatment strategies in mTLE patients.
• Further research is needed to validate and enhance the model's clinical applicability and robustness.