Thalamus segmentation using multi-modal feature classification: Validation and pilot study of an age-matched cohort

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Accurate thalamus segmentation is crucial for understanding volume changes related to neurological conditions, injury, and aging.
• Existing segmentation methods may lack precision or require extensive manual input.

Purpose of the Study:

• To develop and validate an automated algorithm for precise thalamus segmentation.
• To incorporate diffusion tensor imaging (DTI) features and multi-atlas priors for improved accuracy.

Main Methods:

• A novel algorithm was created using DTI features (fractional anisotropy, mean diffusivity, fiber orientation) and averaged thalamus priors from registered atlases.
• A random forest classifier was trained to distinguish thalamus voxels from background.
• The method was validated using leave-one-out cross-validation on nine subjects.

Main Results:

• The proposed algorithm achieved high mean Dice scores: 0.878 for the left thalamus and 0.890 for the right thalamus.
• These scores surpassed those of three state-of-the-art comparison methods.
• A pilot study showed trends of larger thalamic volumes in healthy controls compared to multiple sclerosis patients.

Conclusions:

• The developed automated segmentation method is accurate and efficient for thalamus volume analysis.
• This technique shows promise for clinical applications, such as differentiating between healthy individuals and patients with neurological disorders like MS.
• Further research with larger cohorts is warranted to confirm statistically significant differences in thalamic volume in MS patients.