Abstract
BACKGROUND: Anticipating early neurological deterioration in patients with ischemic stroke undergoing intravenous thrombolysis poses a considerable challenge in clinical practice. This study aimed to develop and validate a diffusion-weighted imaging (DWI)-based clinical-radiomics nomogram for predicting early neurological deterioration in patients with ischemic stroke without large vessel occlusion or hemorrhagic transformation undergoing intravenous thrombolysis. METHODS: A total of 273 patients with stroke were randomly divided into training (n = 192) and validation (n = 81) cohorts at a ratio of 7:3. DWI images taken within 24 hours post-intravenous thrombolysis were used to extract radiological features. The t test, least absolute shrinkage, and selection operator algorithm were used for feature selection. These features were used to create a radiomics score (radscore) for each patient. Combined with the clinical features, a logistic regression model was used to select independent risk factors that were used to construct a clinical-radiomics nomogram. The performance of the nomogram was evaluated using the area under the curve (AUC), calibration, discrimination, and decision curve analysis. RESULTS: A total of 1307 radiomics features were extracted from each patient's data. A total of 310 radiomics features were found to be stable after being screened by intraclass correlation coefficients. Seven features were included in the construction of the radscore. The AUC of the clinical-radiomics nomogram was 0.89 (95% CI, 0.83-0.95) in the training cohort and 0.95 (95% CI, 0.90-0.99) in the validation cohort. The calibration curve and decision curve analysis indicated favorable calibration and net clinical benefits of the nomogram. CONCLUSION: A DWI-based clinical-radiomics nomogram can effectively predict early neurological deterioration in patients with ischemic stroke in the early phase after intravenous thrombolysis.