An integrated radiomics-clinical nomogram incorporating Radscore, NIHSS score, type 2 diabetes mellitus, and monocytes showed potential for predicting 30-day progression risk in patients with acute ischemic stroke, achieving AUCs of 0.945 and 0.904 in training and validation cohorts respectively.
Eight optimal radiomics features were selected from 300 extracted DWI features using LASSO logistic regression to build a radiomics signature.
Multivariate analysis identified four independent predictors of 30-day stroke progression: Radscore, NIHSS score, type 2 diabetes mellitus, and monocytes.
The combined radiomics-clinical model demonstrated favorable discriminative performance in both training and validation cohorts.
When the training-derived cutoff was applied to the validation cohort, the combined model achieved high sensitivity but moderate specificity.
The combined model outperformed both the clinical-only and radiomics-only models in predicting 30-day AIS progression.
A web-based visualization tool was developed to deploy the integrated nomogram for clinical application.
This research suggests that combining MRI-based imaging analysis (radiomics) with routine clinical information can help predict which stroke patients are likely to get worse within 30 days of their stroke. Researchers analyzed 254 patients from two hospitals and extracted 300 mathematical features from a type of brain MRI scan called diffusion-weighted imaging, ultimately narrowing these down to 8 key features. These imaging features were combined with three clinical variables — stroke severity (measured by the NIHSS score), the presence of type 2 diabetes, and a white blood cell measure called monocytes — to build a predictive model. The combined model was highly accurate, correctly identifying at-risk patients about 94% of the time (sensitivity) while correctly clearing low-risk patients about 74% of the time (specificity) in the validation group of patients from a separate hospital. The research also found that type 2 diabetes was the strongest single clinical predictor, with diabetic patients having more than 9 times higher odds of stroke progression compared to non-diabetic patients. The combined radiomics-plus-clinical model performed better than either the imaging-only or clinical-only model alone, suggesting that bringing both types of information together adds real predictive value. The team also built a free web-based tool so clinicians could enter a patient's information and immediately see the estimated risk of progression. This research suggests a practical way to use advanced imaging analysis alongside standard clinical information to flag high-risk stroke patients early, potentially allowing doctors to intensify monitoring or treatment sooner. However, the study was retrospective (looking back at past cases), included only 254 patients across two centers, and the authors themselves emphasize that larger, prospective studies are needed before this tool should be used routinely in clinical practice. The calibration slope from internal validation also suggests some optimism in the model's performance estimates, so real-world accuracy may be somewhat lower than reported.
Liu X, Li P, Zhao H, Zhang W, Ma H, Li S, et al.. (2026). A web-based radiomics nomogram combining MRI features and clinical data for predicting 30-day progression of acute ischemic stroke: a two-center study.. Frontiers in neurology. https://doi.org/10.3389/fneur.2026.1774830