Unsupervised Representation Learning from Pathology Images with Multi-directional Contrastive Predictive Coding

Digital pathology tasks have benefited greatly from modern deep learning\nalgorithms. However, their need for large quantities of annotated data has been\nidentified as a key challenge. This need for data can be countered by using\nunsupervised learning in situations where data are abundant but access to\nannotations is limited. Feature representations learned from unannotated data\nusing contrastive predictive coding (CPC) have been shown to enable classifiers\nto obtain state of the art performance from relatively small amounts of\nannotated computer vision data. We present a modification to the CPC framework\nfor use with digital pathology patches. This is achieved by introducing an\nalternative mask for building the latent context and using a multi-directional\nPixelCNN autoregressor. To demonstrate our proposed method we learn feature\nrepresentations from the Patch Camelyon histology dataset. We show that our\nproposed modification can yield improved deep classification of histology\npatches.\n