Shapley Values-enabled Progressive Pseudo Bag Augmentation for Whole Slide Image Classification
Published in IEEE Transactions on Medical Imaging, 2024
To solve the problem that the attention score fails to accurately locate positive instances, we introduce a metric termed instance importance score (IIS), which can be estimated by the Shapley value, thereby guiding the progressive assignment of pseudo bags for multiple instance training. We reached the SOTA performance on the WSI classification task for CAMELYON-16, BRACS, and TCGA-LUNG datasets. The Shapley value also provides class-wise interpretations.
Recommended citation: Yan, Renao, et al. "Shapley Values-enabled Progressive Pseudo Bag Augmentation for Whole-Slide Image Classification." IEEE Transactions on Medical Imaging (2024). https://arxiv.org/pdf/2312.05490.pdf