Unsupervised Deep Anomaly Detection in Chest Radiographs

摘要

Overview of our anomaly detection system. (a) Anomaly detection based on reconstruction error. The anomaly (a lung mass in this figure) disappears after the reconstruction, and the total reconstruction error of an abnormal image is expected to be larger than that of a normal image. (b) Anomaly detection using code norm. Abnormal images will be out of the distribution of the normal images in the latent space (the standard Gaussian distribution ideally) and farther from the origin than normal ones. The 128-dimensional latent space is drawn as two-dimensional for the explanation