Researchers from MIT, Harvard, and Massachusetts General Hospital have developed new AI-powered software, the BrainStem Bundle Tool (BSBT), capable of automatically segmenting eight distinct white matter bundles in the brainstem from diffusion MRI sequences.
This advancement addresses previous imaging limitations that hindered assessment of these crucial neural cables, which are essential for functions such as consciousness, sleep, and breathing.
Study Findings
Published in the Proceedings of the National Academy of Sciences, the study, led by MIT graduate student Mark Olchanyi, reported that BSBT identified distinct patterns of structural changes in patients with Parkinson's disease, multiple sclerosis (MS), and traumatic brain injury (TBI). It also provided insights into Alzheimer's disease.
The tool was retrospectively used to track bundle healing in a coma patient, reflecting a 7-month recovery process.
Algorithm Development
BSBT functions by tracing fiber bundles that extend into the brainstem from higher brain regions, generating a probabilistic fiber map. A convolutional neural network then combines this map with other imaging data to distinguish the eight bundles.
The neural network was trained using 30 diffusion MRI scans from the Human Connectome Project, which were manually annotated. Validation involved testing the BSBT's output against microscopic dissections of post-mortem human brains and ultra-high-resolution imaging.
Potential Biomarkers and Diagnostic Insights
The research team applied BSBT to diffusion MRI scans from patients with Alzheimer's, Parkinson's, MS, and TBI to assess its ability to track how bundle volume and structure varied with disease or injury. The tool measured bundle volume and 'fractional anisotropy' (FA), an indicator of white matter structural integrity.
Consistent patterns of changes were observed across different conditions:
- Alzheimer's: One bundle showed significant decline.
- Parkinson's: Reduced FA in three bundles and volume loss in another over a two-year period.
- MS: Greatest FA reductions in four bundles and volume loss in three.
- TBI: FA reductions in the majority of bundles, but no significant volume loss.
The study indicated that BSBT was more accurate than other classifier methods in differentiating between patients with health conditions and controls.
The authors suggested that BSBT could serve as a key adjunct to current diagnostic imaging methods by providing a detailed assessment of brainstem white matter structure and longitudinal information.
Tracking Coma Recovery
In a case involving a 29-year-old man in a 7-month coma due to severe TBI, BSBT analysis of his scans revealed that his brainstem bundles had been displaced but not severed. Over the course of his coma, the nerve bundle lesions decreased significantly in volume, and the bundles repositioned.
The authors propose that BSBT holds prognostic potential by identifying preserved brainstem bundles, which could facilitate coma recovery.
The tool has been made publicly available.