Automated Diffusion-Weighted MRI Analysis for Exploratory Risk Stratification of Malignant Cerebral Edema After Acute Ischemic Stroke.
Authors
Affiliations (11)
Affiliations (11)
- Graduate Institute of Biomedical Electronics and Bioinformatics, National Taiwan University, Taipei 106, Taiwan.
- Department of Medical Imaging, China Medical University Hospital, Taichung 404, Taiwan.
- Department of Medical Imaging, China Medical University Hsinchu Hospital, Hsinchu 302, Taiwan.
- Department of Management Science, National Yang Ming Chiao Tung University, Hsinchu 300, Taiwan.
- Show Chwan Memorial Hospital, Changhua 500, Taiwan.
- China Medical University Hospital, Taichung 404, Taiwan.
- Master's Program of Biomedical Informatics and Biomedical Engineering, Feng Chia University, Taichung 407, Taiwan.
- Department of Electrical Engineering, National Taiwan University, Taipei 106, Taiwan.
- Department of Radiology, School of Medicine, College of Medicine, China Medical University, Taichung 406, Taiwan.
- Department of Biomedical Engineering and Environmental Sciences, National Tsing Hua University, Hsinchu 300, Taiwan.
- Department of Computer Science and Information Engineering, National Taiwan University, Taipei 106, Taiwan.
Abstract
<b>Background/Objectives:</b> Malignant cerebral edema (MCE) is an infrequent but devastating complication of acute ischemic stroke (AIS). Early, reliable identification of patients at risk remains a major clinical challenge. The purpose of this study was to evaluate SCR-U1.8, a single automated imaging biomarker integrating stroke lesion burden and cerebrospinal fluid reserve, for exploratory risk stratification of MCE after AIS. <b>Methods:</b> In this retrospective study, consecutive patients with AIS who underwent diffusion-weighted imaging (DWI) between January 2019 and October 2022 were screened. Stroke lesions were automatically segmented on initial DWI using an ADC threshold of <1.8 × 10<sup>-3</sup> mm<sup>2</sup>/s to derive U1.8 lesion volume. Cerebrospinal fluid volume (CSFV) was automatically estimated within the intracranial compartment, and SCR-U1.8 was calculated as U1.8/CSFV. Four low-complexity prediction strategies were evaluated: U1.8 > 82 mL, U1.8 > 145 mL, single-predictor U1.8 logistic regression, and single-predictor SCR-U1.8 logistic regression. Performance was evaluated across 100 repeated stratified patient-level train-test splits. Random undersampling was restricted to the training data, and no synthetic samples were generated. Model performance was assessed by accuracy, sensitivity, specificity, precision, negative predictive value (NPV), F1 score, receiver operating characteristic area under the curve (ROC-AUC), and average precision (AP) as appropriate. <b>Results:</b> A total of 530 patients were included, comprising 198 women and 332 men (mean age, 68 ± 14 years); a total of 12 patients developed MCE. Across 100 repeated stratified train-test splits, SCR-U1.8-LR achieved the highest accuracy (0.990 ± 0.006), precision (0.759 ± 0.150), specificity (0.992 ± 0.005), F1 score (0.806 ± 0.131), ROC-AUC (0.998 ± 0.002), and average precision (0.935 ± 0.062). Compared with the U1.8 > 82 mL threshold, SCR-U1.8-LR achieved significantly higher accuracy, precision, specificity, and F1 score, although sensitivity decreased from 1.000 to 0.900 and NPV decreased slightly from 1.000 to 0.998. <b>Conclusions:</b> SCR-U1.8 may provide a simple and physiologically interpretable measure of stroke lesion burden relative to CSF reserve. The findings remain exploratory because only 12 independent MCE events were available and require external validation.