Unsupervised Domain Adaptation Using Feature Disentanglement and GCNs for Medical Image Classification

The success of deep learning has set new benchmarks for many medical image analysis tasks. However, models often fail to generalize in the presence distribution shifts between training (source) data and test (target) data. One method commonly employed counter is domain adaptation: using samples from target learn account shifted distributions. In this work we propose an unsupervised adaptation approach that uses graph neural networks and, disentangled semantic invariant structural features, allowing better performance across shifts. We extension swapped autoencoders obtain more discriminative features. proposed classification on two challenging datasets with - multi center chest Xray images histopathology images. Experiments show our achieves state-of-the-art results compared other methods.