Background: Accurate thalamic nuclei segmentation is critical for neuroscience research and clinical interventions such as deep brain stimulation and magnetic resonance guided focused ultrasound. Cortical based parcellation has been widely used for two decades, yet its anatomical validity remains uncertain compared with newer imaging approaches. Methods: We analyzed high resolution diffusion magnetic resonance imaging (MRI) and T1 weighted data from 67 healthy young adults in the Human Connectome Project. Cortical based parcellation was performed using probabilistic tractography with cortical targets derived from the HCP MMP1 atlas, generating 8, 11, and 23 region parcellations. Results were compared against three state of the art methods: orientation distribution function (ODF) clustering, track density imaging (TDI), and the structural MRI based segmentation. Group level analyses were conducted in Montreal Neurological Institute and Hospital (MNI) space, and Dice overlap coefficients were calculated against the histology based Morel atlas. Results: Cortical based parcellation demonstrated limited anatomical precision, with increasing cortical target counts introducing greater variability and noise without improving nuclear boundary definition. ODF clustering and TDI recovered subdivisions consistent with cytoarchitectonic patterns, particularly in the pulvinar and mediodorsal nuclei. Structural MRI based segmentation achieved the highest overall Dice coefficients, closely approximating Morel defined boundaries, while Cortical based parcellation consistently underperformed across nuclei. Conclusion: Despite methodological advances, Cortical based parcellation remains constrained in its ability to delineate thalamic nuclei with histological accuracy. By contrast, structural and diffusion microstructural (ODF, TDI) approaches provide superior nuclear localization. These findings highlight the need for hybrid workflows that integrate structural and diffusion based information to enable more reliable thalamic segmentation for research and clinical targeting applications.
Nguyen et al. (Wed,) studied this question.