Machine Learning Quantitative Analysis of FDG PET Images of Medial Temporal Lobe Epilepsy Patients

Area of Science:

• Neuroimaging
• Epileptology
• Artificial Intelligence in Medicine

Background:

• 18F-FDG PET is a key tool for localizing the epileptogenic zone in epilepsy surgery.
• Accurate lateralization of epileptogenic foci is crucial for successful surgical outcomes in medial temporal lobe epilepsy (MTLE).
• Visual inspection of 18F-FDG PET scans has limitations in precisely identifying the side of epileptogenesis.

Purpose of the Study:

• To develop a quantitative algorithm for lateralizing epileptogenic foci using 18F-FDG PET.
• To evaluate the efficacy of machine learning applied to 18F-FDG PET data for classifying the side of epileptogenesis in MTLE patients.

Main Methods:

• Retrospective review of MTLE patients who underwent epilepsy surgery.
• Visual inspection by clinicians to determine the side of MTLE epileptogenesis.
• Extraction of normalized 18F-FDG PET uptake from ROIs using Automated Anatomical Labeling and FreeSurfer aparc + aseg atlases.
• Calculation of lateralization indices and application to machine learning models for classification.

Main Results:

• Visual analysis achieved a 75.3% hit rate for lateralization.
• Machine learning models correctly lateralized 100% of cases using Automated Anatomical Labeling and 82.6% using FreeSurfer aparc + aseg in a subset of difficult cases.
• The testing set demonstrated 100% lateralization accuracy with both parcellation paradigms.

Conclusions:

• Machine-assisted interpretation of 18F-FDG PET data offers superior accuracy for lateralizing MTLE epileptogenesis compared to visual analysis.
• Analyzing regions associated with MTLE, beyond just hippocampal regions, enhances diagnostic performance.
• This quantitative, machine learning-based approach holds significant promise for improving epilepsy surgery planning.