Deep learning-based brain age predicts stroke recurrence in acute ischemic cerebrovascular disease

Acute ischemic cerebrovascular disease (AICVD) has a high recurrence rate, highlighting the need for improved risk stratification beyond traditional factors. The authors developed the Mask-based Brain Age estimation Network (MBA Net), a deep learning framework designed to estimate brain age in patients with acute stroke. The model uniquely isolates contextual brain age (CBA) by applying a masking technique to T2-FLAIR images to exclude acute infarct regions, aiming to capture stable markers of brain health rather than the acute stroke injury itself. The model was trained on 5,353 healthy individuals and then applied to 10,890 patients from the Third China National Stroke Registry (CNSR-III).

The researchers found that the brain age gap (BAG)—the difference between the estimated contextual brain age and the patient's chronological age—serves as an independent predictor of stroke recurrence at 3 months and 5 years. Patients with an accelerated BAG (higher brain age than chronological age) showed a consistently higher risk of recurrent stroke, composite vascular events, and new ischemic stroke. Notably, the model demonstrated superior predictive performance compared to chronological age alone and significantly improved the discriminative accuracy of established clinical stroke risk scores, such as the Essen Stroke Risk Score (ESRS) and the Stroke Prognosis Instrument (SPI).

Traditional risk stratification often relies on chronological age and static clinical history. This study provides a proof-of-concept for integrating AI-derived neuroimaging biomarkers into clinical practice to enhance precision medicine. By demonstrating that a model can accurately predict future risk by focusing on the non-infarcted brain, the authors suggest that brain aging metrics can capture latent vulnerabilities that traditional models miss, offering a potential tool to optimize secondary stroke prevention strategies.