Framework for analysis of brain CT perfusion data: investigation in stroke patients

Background & Objective: Current Computed Tomography Perfusion (CTP) software for stroke assessment often relies on “black-box” algorithms and rigid thresholds. This study explores a transparent pipeline using Symbolic Regression (SR) to identify interpretable, multi-parametric equations for ischemic core prediction. Methods: Hemodynamic maps were extracted from an Acute Ischemic Stroke (AIS) cohort using an optimized Tensor Total-Variation deconvolution framework. The PySR evolutionary algorithm was employed to identify new mathematical models, which were then evaluated across various spatial regularization scenarios. Results & Conclusion: The identified SR formulas showed improved alignment with expert-delineated tissue infarction compared to standard clinical benchmarks and optimized single-parameter thresholds. This framework offers a potential alternative to proprietary commercial software, providing a methodological basis for more transparent and explainable tools in neuroradiological research.