Imaging-Based Prediction Model for TOAST Classification in Acute Ischemic Stroke Using Cerebral Edema Progression Patterns

Background The etiology of acute ischemic stroke (AIS) significantly influences therapeutic decision-making, yet rapid TOAST classification in the hyperacute phase remains challenging. The spatiotemporal dynamics of cerebral edema development are associated with TOAST subtypes. Objective This study aimed to construct a predictive model by analyzing early cerebrospinal fluid (CSF) spatiotemporal characteristics in AIS patients, integrating clinical and imaging data, and evaluating its discriminative value for TOAST etiological classification, thereby providing a basis for early precision treatment. Methods A retrospective study was conducted on 271 AIS patients with anterior circulation stroke who underwent baseline CT within 24 hours of onset and follow-up NCCT within 24 hours. Artificial intelligence-based measurements were used to quantify CSF volumes in different regions. Multivariate logistic regression and LASSO were employed to construct clinical, imaging, and integrated models. A linear mixed-effects model was applied to analyze the dynamic progression of CSF changes. Results Patients with cardioembolic stroke exhibited a more rapid reduction in CSF volume and higher baseline net water uptake. The integrated model demonstrated superior discriminative performance (AUC = 0.854, 95% CI: 0.810–0.900) compared to the clinical model (AUC = 0.798) and imaging model (AUC = 0.734). Furthermore, cardioembolic stroke patients showed a faster CSF volume decline within the first 12 hours than non-cardioembolic stroke patients. Conclusion CSF dynamics may serve as an early biomarker for TOAST classification, potentially guiding acute stroke treatment stratification.