Xiaxia Wu Jinfang Yang Xianqun Ji Yingjian Ye Ping Song Lina Song Peng An ^*

• radiology，xiangyang no.1 peoples hospital, Xiangyang, China

Objective: To explore the value of delta radiomics from cerebral CT perfusion (CTP) in predicting hemorrhagic transformation after intravenous thrombolysis for acute cerebral infarction (HT-ACI). Methods: Clinical and imaging data of 419 patients with acute cerebral infarction who underwent CTP after treatment between November 2016 and August 2024 were retrospectively collected. Based on post-thrombolysis cranial CT or MRI results, patients were divided into the HT-ACI group (114 cases) and the non-HT-ACI group (305 cases). The dataset was split into a training set and a test set in a 7:3 ratio based on time nodes. In the training set, regions of interest(ROI) within the cerebral infarction area on CTP images were delineated using 3D slicer software, and delta radiomic features were extracted. Hemodynamic parameters such as cerebral blood volume (CBV), cerebral blood flow (CBF), and time to peak (TTP) were obtained using CTP techniques. These were combined with baseline patient data (e.g., age, sex, NIHSS score, medical history) to establish various models for predicting HT-ACI through multivariable logistic regression analysis. The predictive performance of the models was compared using DeLong curves, clinical net benefit was assessed using decision curves, and model predictions were validated using the XGboost algorithm. These results were then validated in the test set, and a nomogram and calibration curve were constructed for clinical application. Results: In the training set, significant differences were observed between the two groups in NIHSS score, pre-illness usually use of anticoagulants, age, infarction size, ADC difference, CBF, and Delta radscore (P<0.05). The combined model [AUC 0.878, OR 0.0217, 95%CI 0.835-0.913] demonstrated superior predictive performance compared to the clinical model [AUC 0.725, OR 0.0310, 95%CI 0.670-0.775] and the imaging model [AUC 0.818, OR 0.0259, 95%CI 0.769-0.861]. This was confirmed by the XGboost algorithm, and decision curves confirmed the higher clinical net benefit of the combined model. Similar results were validated in the test set, and a novel nomogram was constructed to simplify the prediction process for HT-ACI. Conclusion.