Abstract
Despite extensive neuroimaging research of primary sensory cortices involved in auditory and visual functions, subcortical structures within these domains, such as the inferior and superior colliculi, the medial and lateral geniculate nuclei and the superior olivary complex, are currently understudied with magnetic resonance imaging (MRI) in living humans. This is because a precise localization of these nuclei is hampered by the limited contrast and sensitivity of conventional neuroimaging methods for deep brain nuclei. In this work, we used 7 Tesla multi-modal (T2-weighted and diffusion fractional anisotropy) 1.1 mm isotropic resolution MRI to achieve high sensitivity and contrast for single-subject brainstem and thalamic nuclei delineation. After precise coregistration to stereotactic space, we generated an in vivo human probabilistic atlas of auditory (medial geniculate nucleus, inferior colliculus, and superior olivary complex) and visual (lateral geniculate nucleus and superior colliculus) subcortical nuclei. We foresee the use of this atlas as a tool to precisely identify the location and shape of auditory/visual deep nuclei in research as well as clinical human studies.
Introduction
While neuroimaging has extensively mapped the primary sensory cortices—the ultimate destinations for sensory data—the subcortical "relay stations" remain significantly understudied in living humans. Structures such as the inferior colliculus (IC) and the lateral geniculate nucleus (LG) are buried deep within the brainstem and thalamus. On standard 1.5T or 3T MRI scanners, these tiny nuclei often lack the contrast and spatial resolution necessary for precise localization.
This lack of clarity makes it difficult for researchers to study how these nuclei function in healthy individuals or how they degrade in conditions like glaucoma, Parkinson’s disease, or dyslexia. Historically, scientists have relied on post-mortem histological atlases, which do not always translate accurately to the living human brain. This study addresses these gaps by utilizing ultra-high-field 7 Tesla (7T) MRI to create a high-resolution, in vivo probabilistic atlas.
Methodology
The researchers leveraged 7T MRI to overcome the limitations of standard clinical imaging, providing a ~2.2x increase in sensitivity compared to 3T. The study utilized a dataset of 12 healthy subjects (6 males, 6 females, average age 28 ± 1 year).
The core of the approach involved:
- Multi-modal Imaging: The study combined T2-weighted images (optimal for the IC, SC, and MG) with Diffusion Fractional Anisotropy (FA) maps, which highlight white matter architecture to make the LG and SOC visible.
- High Resolution: Scans were performed at 1.1 mm isotropic resolution, a significant improvement over standard methods.
- Custom Hardware: A custom-built 32-channel receive coil was used to maximize sensitivity in the brainstem area.
- Manual Segmentation: Two independent raters manually delineated the nuclei in each subject. To ensure high stringency, the final atlas labels only included voxels where both raters agreed.
- Stereotactic Alignment: Individual maps were coregistered to MNI152 space, creating a probabilistic "heat map" of where these nuclei are most likely to be located in the general population.
Results
The researchers successfully created a 3D probabilistic atlas for the IC, SC, MG, LG, and SOC. Validation showed high inter-rater agreement, with Dice similarity coefficients consistently above 0.63 and Hausdorff distances (a measure of boundary error) remaining below the 1.1 mm imaging resolution.
The study provided mean volumes for each nucleus (e.g., the Right LG at ~101 mm³ and the SOC at ~16 mm³). Notably, the right LG was found to be slightly larger than the left, supporting previous findings of hemispheric asymmetry. The volumes calculated in vivo closely matched historical data from post-mortem brain slices, validating that 7T MRI can reach a level of precision previously only possible through invasive means.
Discussion and Future Work
A major takeaway is the necessity of multi-modal imaging; no single MRI modality can visualize all these structures simultaneously. For instance, the Superior Olivary Complex (SOC) is virtually invisible on T2-weighted scans but becomes clear on FA maps.
While 7T MRI acts as a powerful "microscope" for the living brain, limitations remain. The current atlas is based on a small group of healthy young adults; future versions must include older populations and individuals with specific pathologies. Furthermore, even at 1.1 mm resolution, 7T scanners cannot yet resolve the six individual cell layers of the LG.
Ultimately, this atlas serves as a tool to improve the accuracy of deep brain stimulation, neurosurgical planning, and the tracking of neurodegenerative diseases. The next step involves integrating this atlas into standard neuroimaging software pipelines for broader scientific use.
Summary updated