Table 2 Predictive performance results for intratumoral plus 4 mm peritumoral region for CR and DLB models. Values shown are from the random seed with highest training set accuracy. The DLB pipeline slightly outperformed CR in the validation set of the same resolution as the training set. A larger improvement is seen in the testing set of dissimilar resolution. This indicates the DLB pipeline might be more generalizable and less sensitive to pixel size differences

From: Preoperative prediction of lymph node metastasis using deep learning-based features