Fig. 1

Pipeline of the classification process of artificial intelligence. There are three discrete steps in the pipeline: inputting images (i.e., image acquisition and pre-processing), model development, and model performance validation. Generally, there are two main approaches to developing AI models: the hand-crafted radiomics method and the deep learning radiomics method. For the hand-crafted radiomics model, there are four steps: annotation, feature extraction, feature selection and modeling with traditional machine learning methods. For the deep learning radiomics model, the images can be fed into the end-to-end model to output the risk probability of adverse outcomes. ADC apparent diffusion coefficient, AUC area under the receiver operating characteristic curve, DWI diffusion-weighted imaging, LASSO least absolute shrinkage and selection operator, LR logistic regression, RFE recursive feature elimination, T2WI T2-weighted imaging, SVM support vector machine