Prediction of Hematoma Expansion in Hypertensive Intracerebral Hemorrhage by a Radiomics Nomogram

Objective: To develop and validate a radiomics-based nomogram model which aimed to predict hematoma expansion Methods: Patients with HICH (n=187) were included from October 2017 to March 2022 in the Yongchuan Affiliated Hospital of Chongqing Medical University. Patients were randomly divided into a training set (n=130) and a validation set (n=57) in a ratio of 7:3. The radiomic features were extracted from the regions of interest (including main hematoma, the surrounding small hematoma(s) and perihematomal edema) in the first CT scan images. The variance threshold, SelectKBest and LASSO (least absolute shrinkage and selection operator), features were selected and the radiomics signature was built. Multivariate logistic regression was used to establish a nomogram based on clinical risk factors and the Rad-score. A receiver operating characteristic (ROC) curve was used to evaluate the generalization of the models' performance. The calibration curve and the Hosmer-Lemeshow test were used to assess the calibration of the predictive nomogram. And decision curve analysis (DCA) was used to evaluate the prediction model. Results: Thirteen radiomics features were selected to construct the radiomics signature, which has a robust association with HE. The radiomics model found that blend sign was a predictive factor of HE. The radiomics model ROC in the training set was 0.89 (95%CI 0.82-0.96) and was 0.82 (95%CI 0.60-0.93) in the validation set. The nomogram model was built using the combined prediction model based on radiomics and blend sign, and worked well in both the training set (ROC: 0.90[95%CI 0.83-0.96]) and the validation set (ROC: 0.88[95%CI 0.71-0.93]). Conclusion: The radiomic signature based on CT of HICH has high accuracy for predicting HE. The combined prediction model of radiomics and blend sign improves the prediction performance.