Abstract
BACKGROUND AND AIM: Symptomatic intracranial arterial stenosis (ICAS) is a leading cause of ischemic stroke, and its progression is associated with an increased risk of stroke recurrence and poor outcomes. Accurate prediction of the risk of progression in ICAS patients is crucial for timely intervention and management. This study aims to develop and validate logistic regression and deep learning models to predict the risk of progression in symptomatic ICAS patients and compare their predictive performance. METHODS: A retrospective study was conducted on 266 symptomatic ICAS patients who were followed for at least 3 years. The dataset was randomly split into a training set (70%) and a validation set (30%). Data preprocessing involved normalization, feature selection, and class balancing techniques to enhance model performance. Logistic regression, and deep learning models were developed to predict the risk of ICAS progression. The models were evaluated using accuracy, sensitivity, specificity, precision, F1-score, and the area under the receiver operating characteristic curve (AUC). RESULTS: The logistic regression model achieved an AUC of 0.771 (training) and 0.767 (validation; 95% CI: 0.702-0.832). The deep learning model demonstrated superior performance with an AUC of 0.898 (training) and 0.863 (validation; 95% CI: 0.801-0.925), showing a statistically significant improvement (p = 0.016, DeLong's test). Feature importance analysis identified hypertension, diabetes, stenosis degree, and prior stroke history as the most influential predictors of ICAS progression. These results highlight the value of early risk stratification to guide timely clinical intervention. CONCLUSION: Compared to logistic regression, the deep learning model exhibited significantly improved predictive accuracy for the risk of progression in symptomatic ICAS patients. The high performance and reliability of the deep learning model highlight its potential clinical utility in predicting ICAS progression, ultimately aiding in risk stratification and personalized treatment strategies.