Toward reliable thalamic segmentation: an evaluation of automated methods for structural MRI

Area of Science:

• Neuroimaging
• Brain Anatomy
• Medical Image Analysis

Background:

• Automated thalamic nuclear segmentation advances neuroimaging by treating the thalamus as distinct nuclei within brain circuits.
• Current reliance on FreeSurfer's T1-weighted MRI segmentation is limited by poor intrathalamic nuclear contrast.
• Existing segmentation methods like HIPS-THOMAS and FreeSurfer updates lack comprehensive comparisons and rely on limited atlases.

Purpose of the Study:

• To compare the performance of HIPS-THOMAS, standard FreeSurfer, and a recent FreeSurfer update for thalamic nuclear segmentation.
• To evaluate these methods using two thalamic atlases across healthy adults and patients with autoimmune limbic encephalitis.

Main Methods:

• Comparative analysis of three automated thalamic segmentation techniques: HIPS-THOMAS, standard FreeSurfer, and a FreeSurfer update.
• Utilized two cohorts of healthy adults and one cohort of patients with chronic autoimmune limbic encephalitis.
• Assessed segmentation accuracy against two distinct thalamic atlases.

Main Results:

• HIPS-THOMAS outperformed both standard FreeSurfer and its diffusion-based update in healthy adults.
• The FreeSurfer update showed limited improvements, primarily in a few specific thalamic nuclei.
• Standard FreeSurfer exhibited underperformance in differentiating patients from controls, particularly affecting the anteroventral and pulvinar nuclei.

Conclusions:

• HIPS-THOMAS is a more accurate method for automated thalamic nuclear segmentation compared to FreeSurfer variants using structural MRI.
• Standard FreeSurfer's limitations impact clinical applications, especially in identifying specific nuclei affected in certain neurological conditions.
• Provides evidence-based recommendations for selecting automated thalamic segmentation methods in structural neuroimaging studies.