Robust surface-based multi-template automated algorithm to segment healthy and pathological hippocampi.

Area of Science:

• Neuroimaging and Medical Image Analysis
• Epilepsy Research
• Computational Anatomy

Background:

• Temporal lobe epilepsy (TLE) is the most common drug-resistant epilepsy, often linked to hippocampal atrophy and atypical morphologies.
• Current automatic hippocampal segmentation methods yield unsatisfactory results for TLE patients.
• Accurate hippocampal segmentation is crucial as atrophy presence predicts favorable surgical outcomes.

Purpose of the Study:

• To develop a novel, automated hippocampal segmentation method (SurfMulti) for TLE patients.
• To improve the accuracy and reliability of hippocampal segmentation in the presence of atrophy and shape variability.
• To provide a robust tool for presurgical evaluation in TLE.

Main Methods:

• Proposed SurfMulti, a surface-based approach for statistical estimation of locoregional texture and shape.
• Utilized a multi-template library from controls and TLE patients to handle inter-subject variability, including shape variants.
• Compared SurfMulti performance against state-of-the-art volume-based methods and manual volumetry.

Main Results:

• SurfMulti achieved high segmentation performance (Dice index: 86.1%), comparable to controls (87.5%) and superior to existing methods.
• The method demonstrated sensitivity in detecting atrophy similar to manual volumetry.
• SurfMulti effectively accounts for inter-subject shape variability.

Conclusions:

• SurfMulti is a robust automated algorithm for hippocampal segmentation in TLE patients with atrophy.
• It offers a reliable surrogate for time-consuming manual volumetry in presurgical evaluations.
• The algorithm's accuracy supports its use in predicting surgical outcomes for TLE.