Zhou J, Zhang Y, Chang KT, et al. - Via the retrospective study of 133 persons with histologically validated 91 malignant and 62 benign mass lesions for training (74 individuals with 48 malignant and 26 benign lesions for testing), researchers assessed the diagnostic exactitude of mass lesions using region of interest (ROI)-based, radiomics and deep-learning methods, by taking peritumor tissues into account. In the training dataset, the diagnostic exactitude was 76%, 84%, and 86% using three ROI-based parameters, the radiomics model, and the ROI + radiomics model, respectively. In deep learning using the per-slice basis, the area under the receiver operating characteristic was similar, smallest and 1.2 times box, which was significantly greater compared with 1.5 and 2.0 times box. For per-lesion diagnosis, when using the smallest bounding box, the highest exactitude of 91% was attained, and that reduced to 84% for tumor alone and 1.2 times box, and further to 73% and 69% for 1.5 times box and for 2.0 times box, respectively. In the independent testing dataset, when using the smallest bounding box, the per-lesion diagnostic exactness was also the highest, ie, 89%. Hence, deep learning using ResNet50 attained high diagnostic exactitude. Moreover, in comparison with using tumor alone or larger boxes, using the smallest bounding box comprising proximal peritumor tissue as input had greater accuracy.