A Radiomic-based Method for Predicting the Prognosis of Ischemic Stroke from Diffusion-weighted Imaging Images

Accurate prognosis is the keystone of stroke treatment. To provide a reliable reference for clinic decisions, a radiomic-based method for predicting the prognosis of ischemic stroke from diffusion-weighted imaging (DWI) images was proposed in this work. On the acquired DWI images of stroke patients, the features of the ischemic core and the peripheral ischemic penumbra (salvageable tissue) were extracted by radiomics method, and the DWI images of corresponding cases collected after treatment were taken as the gold standard of the infarction progression. 60 cases were randomly split into training and testing set in the ratio of 2:1. Three-fold cross-validation was applied. The sparse representation algorithm was adopted to select the radiomics features and train the prediction classifier. The accuracy of predicting the ischemic core and the salvageable tissue around the core was 82% and 87% respectively. This information has the potential to be utilized for diagnosis, prognosis prediction, and evaluation of therapeutic effectiveness.