Enhancing early stroke diagnosis with radiomics models derived from CT scans: A retrospective analysis

Background Stroke, a leading cause of adult disability and mortality, requires timely diagnosis for effective treatment. This study radiomics models developed from CT scans as a means to enhance early stroke diagnosis. Methods Our retrospective study analyzed CT scans and medical records of 174 patients diagnosed with stroke from January 2023 to December 2024 at the local university. After manual ROI segmentation by radiologists, 2380 high-throughput radiomic features were derived using the PyRadiomics package. Quantitative imaging features were preprocessed, normalized, and filtered through t-tests for statistical significance and Pearson correlation coefficients for collinearity. Feature selection was conducted using the Least Absolute Shrinkage and Selection Operator (LASSO) regression. The predictive model was established based on the logistic regression (LR) algorithm, utilizing cross-validation for performance assessment. Results After rigorous feature selection, 15 radiomic Features were determined to be the strongest predictors of stroke. The LR radiomics model, based on these features, demonstrated high diagnostic accuracy with a specificity of 0.7895. The model's performance was further supported by calibration curves and decision curve analysis, indicating its clinical applicability. Conclusion The radiomics-based model presents a potentially transformative approach for early stroke diagnosis, with high sensitivity and specificity, indicating its utility in facilitating timely clinical intervention and improving patient outcomes.