Data Efficient and Weakly Supervised Computational Pathology on Whole Slide Images

Ming Y. Lu, Drew F. K. Williamson, Tiffany Y. Chen, Richard J. Chen, Matteo Barrberi and Faisal Mahmood*

arXiv | GitHub

TL;DR: CLAM is a high-throughput and interpretable method for data efficient whole slide image (WSI) classification using slide-level labels without any ROI extraction or patch-level annotations, and is capable of handling multi-class subtyping problems. Tested on three different WSI datasets, trained models adapt to independent test cohorts of WSI resections and biopsies as well as cellphone microscopy data.

Understanding Attention - This demo showcases the model's ranking of different regions in the slide in terms of their importance to the final slide-level diagnostic prediction. High attention regions (shown in red) correspond to regions considered by the model to be important towards the classification determination and the regions of highest attention should correspond to morphological patterns that are characteristic of the predicted class. On the other hand, low attention regions (shown in blue) should generally correspond to regions deemed by the model as lacking in diagnostic relevance to the underlying prediction.