Self-rule to multi-adapt: Generalized multi-source feature learning using unsupervised domain adaptation for colorectal cancer tissue detection

•A label-efficient framework called SRA for tissue type recognition in histological images is proposed.•The architecture takes advantage of the large quantity of unlabeled whole side images using self-supervised learning.•The presented model uses an entropy-based approach that progressively learns domain invariant features.•It generalizes to multi-source domain adaptation.