Pham HHN, et al. - In this study, 349 whole-slide lung cancer lymph node images, including 233 slides for algorithm training, ten for verification, and 106 for assessment, were obtained and a novel two-step deep learning algorithm was formed in order to address the subject of false-positive prognostication while sustaining specific cancer discovery. Using this two-step strategy, errors were, on average, decreased by 36.4% and up to 89% in slides with reactive lymphoid follicles. Moreover, in cases of macrometastases, micrometastases, and isolated tumor cells, 100% sensitivity was attained. Using foci size thresholds of 0.6 mm and 0.7 mm, a receiver-operating characteristic curve was created to decrease the small number of persisting false positives, attaining sensitivity and specificity of 79.6% and 96.5%, and 75.5% and 98.2%, respectively. Thus, a two-step strategy could be employed to efficiently identify lung cancer metastases in lymph node tissue and with few false positives.