THALAMIC PARCELLATION FROM MULTI-MODAL DATA USING RANDOM FOREST LEARNING.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• The thalamus, a subcortical gray matter structure, comprises nuclei crucial for communication between the cerebral cortex and midbrain.
• These nuclei are significantly impacted in neurodegenerative conditions like multiple sclerosis and Alzheimer's disease.
• Thalamic parcellation is challenging due to limited contrast in standard magnetic resonance (MR) imaging.

Purpose of the Study:

• To develop an advanced method for differentiating and mapping thalamic nuclei.
• To combine diffusion tensor imaging (DTI) data with structural MR imaging for improved parcellation.
• To enhance the understanding of thalamic involvement in neurodegenerative diseases through precise anatomical segmentation.

Main Methods:

• Utilized multi-modal features including spatial location, fiber orientation from DTI, and structural MR data.
• Developed random forest learners trained on voxel-wise multi-dimensional features and manual classifications.
• Integrated trained learners with a multiple object level set model for automated parcellation.

Main Results:

• Achieved effective differentiation of thalamus from background and individual thalamic nuclei.
• Demonstrated the efficacy and reproducibility of the developed multi-modal approach.
• Quantitatively validated results through cross-validation on a dataset of twenty subjects.

Conclusions:

• The proposed method significantly improves the accuracy and reliability of thalamic nuclei segmentation.
• This technique offers a valuable tool for research into neurodegenerative diseases affecting the thalamus.
• The combination of DTI, structural MRI, and machine learning provides a robust framework for complex brain structure parcellation.