The red dot® algorithm is based on deep learning models. The training process teaches the algorithm to both classify a CXR and localize its findings. This localization is rendered through heatmaps and allows to understand and interpret red dot® output.