Introduction: Brain age gap estimation (BrainAGE) has emerged as a promising biomarker of brain health, with MRI-based approaches achieving high accuracy. However, MRI’s limited availability and acquisition time restrict its use in acute stroke. We propose a deep learning method for estimating brain age from urgently acquired non-contrast CT (NCCT), providing a biomarker adapted to time-critical clinical workflows. Methods: We conducted a prospective cohort study at two comprehensive stroke centers. All patients presented within 24h of symptom onset and underwent urgent NCCT. After QC and standard preprocessing, scans were hemisphere-cropped based on lesion side. ResNet18 adapted for 3D NCCT volumes with sex as an auxiliary input was first trained on whole-brain scans from 2730 non-stroke control NCCTs and subsequently distilled into hemisphere-specific models. Age bias was corrected using control validation data. Performance (mean absolute error-MAE, correlation) was evaluated on the bias-corrected test set. In stroke patients, only the contralesional bias-corrected BrainAGE was analyzed. Results: We analyzed 1470 acute ischemic stroke patients (mean age 70.5±13.3 years; range 24-100 years, 44% female). The median baseline NIHSS was 12 (IQR 5-18). Occlusion sites included ICA (19%), MCA-M1 (38%), MCA-M2 (25%), MCA-M3 (9%), and other sites (29%). Treatment allocation was IV thrombolysis in 22%, EVT in 27%, combined therapy in 25%, and no reperfusion in 26%. Stroke etiology was cardioembolic (37%), large-artery atherosclerosis (22%), small-vessel occlusion (2%), other determined causes (7%), and undetermined/ESUS (32%). At 90 days, 57% achieved independence (mRS≤2), and mortality was 21%. On the control test set, the model achieved an MAE of 5.4 and 5.9 years and a correlation with chronological age of 0.88 and 0.85 for the left and right hemispheres, respectively. In stroke patients, contralesional hemispheres showed a mean positive brain age gap of 2.6 years, consistent with stroke-related accelerated brain aging (Figure 1). Qualitative review highlighted individual cases with marked brain age gaps, including younger patients with “older-appearing” brains and older patients with “youthful-appearing” brains (Figure 2). Conclusions: We demonstrate that our model can estimate BrainAGE from urgent NCCT in acute ischemic stroke patients, providing a measurable age prediction that can inform physicians about the potential benefit of acute stroke reperfusion therapies.
Mahajna et al. (2026) studied this question.