Machine Learning for Identification of Fast Progressors of Infarct Growth in Early Window Acute Ischemic Strokes Without Perfusion Imaging

Overview

Purpose: Fast progressors (FP) of anterior circulation acute ischemic strokes at 0 -6 hours from onset (early window) are associated with less favorable outcome. FP is determined by using computed tomographic perfusion (CTP) to measure infarct core volume ( ICV), but CTP requires additional time, costs, and radiation, in addition to pitfalls such as patient motion. My aim was to determine if machine learning (ML) can accurately identify FP using only baseline demographic, clinical, laboratory, CT, and CT angiography (CTA) data, to bypass CTP. Methods: Retrospective study of stroke patients arriving 0 -6 hours from symptom onset with anterior circulation occlusion on CTA who had concurrent CTP. Infarct growth rate (IGR) was calculated as ICV/time from onset and FP was defined as IGR ≥10 mL/hour. Four ML algorithms for binary classification were trained on 11 selected features. Data were randomly split 70:30 for training and testing. 10-fold cross validation scores were assessed during training with tuning of hyperparameters. Models were tested using areas under the receiver -operating characteristics curves (AUROC) and predictive values. Results: 147 patients were included with median (IQR) age of 76 years (67-85), ICV of 11 mL (0- 34), and IGR of 3.4 mL/hour (0-12.3). 47 (32.0%) were FP. XGBoost was the best performing ML model with AUROC of 0.900, PPV of 76.5%, NPV of 96.4%, sensitivity o f 92.9%, specificity of 87.1%, and accuracy of 88.9%. Conclusion: ML can accurately identify FP in the early window without CTP.

Competition history

• JSHS 2025 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.