Thalamus parcellation using multi-modal feature classification and thalamic nuclei priors

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Thalamus segmentation is crucial for tracking neurodegenerative diseases.
• Current methods often rely solely on T1-weighted MRI and require manual input.
• Accurate parcellation of small thalamic nuclei is particularly challenging.

Purpose of the Study:

• To develop an automated algorithm for segmenting the thalamus and its nuclei.
• To improve the accuracy and efficiency of thalamic parcellation compared to existing methods.

Main Methods:

• Utilized diffusion tensor imaging (DTI) features and thalamic nuclei location priors.
• Employed a hierarchical random forest classifier for thalamus localization.
• Applied a second random forest classifier for classifying individual thalamic nuclei.

Main Results:

• The proposed algorithm achieved higher Dice scores for whole thalamus and several nuclei segmentation compared to state-of-the-art methods.
• Demonstrated improved accuracy in locating smaller, challenging nuclei like the lateral and medial geniculates.

Conclusions:

• The developed automated segmentation algorithm offers a more accurate and efficient approach for thalamus and nuclei parcellation.
• This method has significant potential for quantifying volumetric changes in neurodegenerative diseases.