Figure 1.
Overview of the feature disentanglement modeling approach. We propose to learn a model that simultaneously predicts the class label and domain label for a given CXR image. The parameters of the model are updated to extract representations that contain information about the class label but not about domain label.