Semi-Supervised Histology Classification using Deep Multiple Instance Learning and Contrastive Predictive Coding

Convolutional neural networks can be trained to perform histology slide using weak annotations with multiple instance learning (MIL). , given the paucity of labeled histology data, direct application of MIL easily suffer from overfitting and the network is unable to learn rich representations due to the weak supervisory signal. We propose to such limitations with a two-stage semi-supervised approach that the power of data-efficient self-supervised feature learning via predictive coding (CPC) and the interpretability and flexibility of attention-based MIL. We apply our two-stage CPC + MIL-supervised pipeline to the binary classification of breast cancer images. Across five random splits, we report state-of-the-art with a mean validation accuracy of 95% and an area under the ROC of 0. 968. We further evaluate the quality of features learned via CPC to simple transfer learning and show that strong classification using CPC features can be efficiently leveraged under the MIL even with the feature encoder frozen.