Prediction of Intracerebral Hemorrhage Based on the Image-Related and Clinical Text Features

Accurate identification of patients with intracerebral hemorrhage (ICH) who present with early hematoma expansion is important to guide clinical decision-making. Typically, physicians determine whether a patient has early hematoma expansion based on clinical text features and non-contrast computed tomography (NCCT). However, medical image processing technology is limited to extracting image features to complete the task of image classification. In recent years, radiomics has been widely used for the study of brain tumor diseases, but rarely for the study of brain hemorrhage diseases. To effectively predict whether patients have early hematoma expansion phenomenon, this study combined both the handcrafted radiomics (HCR) features and clinical text features to construct prediction models and improve the prediction and generalization ability of the models. The experimental results showed that the feature fusion classification model was best with the help of SVM. The AUC in the training cohort, internal validation cohort and external validation cohort were 0.860, 0.818 and 0.833, respectively. In terms of predicting early hematoma expansion in ICH patients, the combined model based on NCCT has better predictive performance than the HCR model.