Thalamic nuclei segmentation from T1-weighted MRI: Unifying and benchmarking state-of-the-art methods.

Area of Science:

• Neuroimaging
• Neuroanatomy
• Neurology

Background:

• Thalamic nuclei are vital for cognition and sensorimotor functions.
• Neurological conditions often involve the thalamus.
• Accurate thalamic nuclei segmentation in neuroimaging is challenging due to technical difficulties and lack of standardized nomenclature.

Purpose of the Study:

• To benchmark four state-of-the-art T1 MRI thalamic segmentation methods.
• To compare segmentation accuracy against the Morel stereotaxic atlas.
• To evaluate methods' ability to detect thalamic degeneration in Alzheimer's disease progression and distinguish patient groups from healthy controls.

Main Methods:

• Utilized data from healthy young adults (Human Connectome Project, n=100) and older adults, mild cognitive impairment, and Alzheimer's disease patients (Alzheimer's Disease Neuroimaging Initiative, n=540).
• Benchmarked FreeSurfer, HIPS-THOMAS, SCS-CNN, and T1-THOMAS segmentation methods.
• Compared segmentations using overlap and dissimilarity metrics against the Morel atlas and quantified diagnostic discrimination based on thalamic volumes.

Main Results:

• HIPS-THOMAS provided the most effective segmentations of individual thalamic nuclei compared to the Morel atlas.
• HIPS-THOMAS accurately discriminated healthy controls from mild cognitive impairment and Alzheimer's disease patients using individual nucleus volumes.
• SCS-CNN was most accurate in classifying healthy controls when using whole thalamus volumes.

Conclusions:

• This study systematically compares thalamic segmentation methods under a unified framework for the first time.
• HIPS-THOMAS is recommended for studies focusing on specific thalamic nuclei's functional relevance.
• SCS-CNN may be suitable for whole thalamus analyses in disease classification.