Han SS, Moon IJ, Lim W, et al. - Researchers examined if an algorithm using a region-based convolutional neural network can assist in locating suspected areas and predicting the probability of a lesion being malignant. Using a region-based convolutional neural network, they created a total of 924,538 training image-crops including various benign lesions from 182,348 clinical photographs in this diagnostic study. Based on image findings, manual or automatic interpretation of these possible lesions was done; following this, convolutional neural networks were trained with 1,106,886 image crops to locate and diagnose cancer. From 3 hospitals, they obtained validation data sets (2,844 images from 673 patients; mean age, 58.2 [19.9] years; 308 men [45.8%]; 185 patients with malignant tumors, 305 with benign tumors, and 183 free of tumor) between January 1, 2010, and September 30, 2018. A mean of 4.2 (2.4) photographs per patient was analyzed using the algorithm and the malignancy score was reported according to the highest malignancy output. The area under the receiver operating characteristic curve for the validation data set was 0.910, and the algorithm’s F1 score (mean of precision and recall; range, 0.000-1.000) and Youden index score (sensitivity + specificity −1; 0%-100%) were comparable with those of dermatologists and exceeded those of nondermatologists. Based on these findings, they suggest the possible applicability of the algorithm with unprocessed photographs for localizing and diagnosing skin cancer without manual preselection of suspicious lesions by dermatologists.