MCH-Guard, a multimodal machine-learning framework, stratifies cerebral microhemorrhage risk in Alzheimer's Disease Neuroimaging Initiative participants with high accuracy (AUC = 0.86 for the comprehensive model) and identifies a transient vascular instability phenotype strongly predicted by hepatic factors.
The comprehensive MCH-Guard model detected baseline cerebral microhemorrhage (MCH) with high accuracy.
A minimal model using only demographics and clinical history achieved robust MCH detection performance.
Longitudinal models predicted time-to-MCH incidence with substantial explanatory power.
MCH-Guard identified a transient vascular instability phenotype characterized by fluctuating MCH status.
MCH-Guard was developed as a clinical decision-support tool for both spontaneous MCH and ARIA-H surveillance during anti-amyloid therapy.
This research suggests that a new computer-based tool called MCH-Guard can accurately identify Alzheimer's disease patients who are at risk for tiny brain bleeds called cerebral microhemorrhages (MCH). These small bleeds have become increasingly important to monitor because new Alzheimer's treatments that clear amyloid plaques from the brain can cause a side effect called ARIA-H (amyloid-related imaging abnormalities with hemosiderin deposits), which involves similar bleeding patterns. The tool was tested on over 800 participants from a major Alzheimer's research database and used combinations of medical history, lab tests, and brain imaging to assess risk. One of the most practically significant findings is that even a simplified version of the tool — using only basic patient demographics and medical history without any specialized lab tests or brain scans — still performed reasonably well (AUC of 0.72 out of a perfect 1.0). This suggests the approach could be useful in clinics that don't have access to advanced imaging equipment. The full model, which incorporated all available data types, performed even better (AUC of 0.86). The tool could also predict how long it would take for new microhemorrhages to appear and assess a patient's four-year risk. The study also uncovered a previously underappreciated pattern: some patients have a 'transient vascular instability' phenotype where microhemorrhages seem to appear and disappear over time rather than steadily accumulating. This fluctuating pattern was linked to liver-related (hepatic) factors. This finding matters because such fluctuations could confuse doctors and researchers trying to determine whether a treatment is causing harm or whether natural bleeding processes are occurring — making it an important consideration for the safe monitoring of patients receiving anti-amyloid Alzheimer's therapies.
Gel A, Phillips E, Hausle I, Thropp P, Tosun D. (2026). MCH-Guard: Multimodal machine learning framework for risk stratification of cerebral microhemorrhage risk in the Alzheimer's Disease Neuroimaging Initiative.. Alzheimer's & dementia : the journal of the Alzheimer's Association. https://doi.org/10.1002/alz.71818