Generated blood-volume maps improve brain age estimates
Among cognitively normal controls, average age-estimation error was 3.95 years with the combined model, versus 4.10 years using structural scans alone.
Our read
Biomarkers- Studied:
- Computational study of cognitively normal controls, against structural MRI-only models
- What changed:
- Brain age estimation error was lower with generated maps and sex included: 3.95 years against 4.10 years with structural MRI alone
- The catch:
- Generated blood-volume maps can re-encode anatomy rather than independently measure vascular physiology
- In the record:
- The 37th paper on BrainAGE with results in people; 66 on record in all.
- Why it’s aging research:
- It builds or tests a tool for aging research.
Editor of record: Boris Djordjevic · AgingResearch.News
Researchers used 2,851 human brain scans from 13 public datasets, including scans from 1,507 females, to develop a brain-age prediction model. This modeling study combined predictions from structural magnetic resonance imaging (MRI) scans and computer-generated maps of brain blood volume. Those maps came from the same scans, without injected contrast material. The final model also included sex.
Among controls with normal thinking and memory, average prediction error was 3.95 years, compared with 4.10 years for structural scans alone and 4.49 years for generated maps alone. The researchers separately evaluated 1,233 people with normal cognition, mild cognitive impairment—problems with memory or thinking—or Alzheimer’s disease. These participants were excluded from model development. The gap between estimated brain age and actual age rose across those groups and was associated with worse cognitive scores.
Why it matters
Brain age estimates aim to capture age-related changes relevant to dementia risk. The authors suggest that this approach could support early detection and monitoring of treatment responses.
Caveats
The blood-volume maps were computer-generated proxies, not direct measurements. Associations with diagnoses and cognitive scores do not establish whether using the model improves diagnosis or care.
From the paper
- Result
The combined T1w + DeepCBV model (TD-model) showed a substantial reduction in RSS compared with the T1w-only model (RSS difference 95% CI = [349, 1719]; bootstrap P = 0.001), corresponding to a large effect size (Cohen’s d (d) = 2.63, 95% CI: 0.98 to 4.80).
Quoted word for word from the full text on PubMed Central.
The paper
Enhancing brain age estimation with structural MRI and synthesized cerebral blood volume maps
Columbia University
Brain Communications, Jul 22, 2026, Open access
Comments
Be the first to comment.