Longitudinal Boundary Sharpness Coefficient Slopes Predict Time to Alzheimer's Disease Conversion in Mild Cognitive Impairment: A Survival Analysis Using the ADNI Cohort
A new AI survival model analyzes MRI boundary degradation to forecast when MCI patients will develop Alzheimer's.
Researcher Ishaan Cherukuri has developed a novel AI-powered survival analysis model that significantly improves predictions of when patients with mild cognitive impairment (MCI) will progress to Alzheimer's disease. The key innovation lies in analyzing how the Boundary Sharpness Coefficient (BSC)—a measure of how well-defined the gray-white matter boundary appears on structural MRI—changes over time. By tracking the rate of boundary degradation (the BSC slope) rather than relying on single timepoint measurements, the model captures the neurodegenerative process more effectively.
The study analyzed 1,824 T1-weighted MRI scans from 450 subjects in the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort, including 95 who converted from MCI to Alzheimer's over a mean follow-up of 4.84 years. The Random Survival Forest model, trained on these temporal slope features, achieved a test concordance index (C-index) of 0.63 for predicting time-to-conversion. This represents a dramatic 163% improvement over baseline parametric models, which scored only 0.24. Unlike previous deep learning approaches (CNNs and RNNs) that achieved high classification accuracy but disregarded specific brain regions, this method focuses specifically on the vulnerable gray-white matter interface.
This breakthrough matters because current prediction methods face significant limitations, creating uncertainty that hampers clinical trial enrollment and delays treatment. The new approach uses standard structural MRI, which costs $800-$1,500—a fraction of the $5,000-$7,000 for PET imaging—and doesn't require invasive cerebrospinal fluid collection. By providing more accurate, individualized risk timelines, these temporal biomarkers could transform patient screening, risk assessment, and enable earlier therapeutic interventions when treatments become available.
- Model analyzes Boundary Sharpness Coefficient (BSC) slopes from MRI scans to measure gray-white matter boundary degradation rates over time
- Achieved test C-index of 0.63—163% improvement over baseline models (0.24)—using Random Survival Forest on 1,824 scans from 450 ADNI subjects
- Uses affordable structural MRI ($800-$1,500) instead of expensive PET imaging ($5,000-$7,000), enabling more accessible Alzheimer's progression forecasting
Why It Matters
Enables earlier, more affordable Alzheimer's risk assessment using routine MRI scans, potentially transforming clinical trial design and patient care.