Multi-parametric MRI radiomics-based machine learning model for predicting clinical outcomes in acute ischemic stroke

Early and accurate prediction of outcome in acute ischemic stroke (AIS) is important for clinical management. We aimed to develop and validate a clinical-radiomics nomogram to predict clinical outcomes in patients with AIS. A total of 214 AIS patients were retrospectively included, which were assigned to a training ( n = 105), validation ( n = 43), internal test ( n = 33), and external test cohorts ( n = 33). All patients underwent baseline multi-parametric MRI (mp-MRI) scans. According to the 90-day modified Rankin Scale (mRS) after discharge, the patients were stratified into good (mRS ≤ 2) and poor (mRS > 2) prognosis subtypes. A total of 1130 radiomic features were extracted from diffusion-weighted imaging, apparent diffusion coefficient, and cerebral blood flow images generated from arterial spin labeling data, respectively. The clinical, radiomics, and clinical-radiomics models were constructed for predicting AIS prognosis. The model discrimination, calibration, and clinical usefulness were evaluated using the area under the curve (AUC), calibration curves, and decision curve analysis (DCA), respectively. The clinical-radiomics nomogram, which integrates the multi-parametric MRI-based (mp-MRI) radiomics signature and diabetes, demonstrated higher AUCs (0.796 and 0.741) compared to the clinical model (0.787 and 0.670) and the mp-MRI radiomics model (0.783 and 0.682) in the internal and external test cohorts, respectively. The nomogram showed strong goodness-of-fit in both training and validation cohorts. DCA further indicated the clinical utility of the integrated model. The clinical-radiomics model is effective in predicting clinical outcomes of AIS patients, which can be used to guide personalized treatment strategies.