Reproduce “STARC-9: A Large-scale Dataset for Multi-Class Tissue Classification for CRC Histopathology” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Barathi Subramanian, Rathinaraja Jeyaraj, Mitchell Nevin Peterson, Terry Guo, Nigam Shah, Curtis Langlotz, Andrew Y. Ng, Jeanne Shen
Researchers from Stanford University and DeepLearning.AI developed STARC-9, a large-scale, high-quality dataset for multi-class tissue classification in colorectal cancer (CRC) histopathology, alongside DeepCluster++, a semi-automated framework for its construction. Models trained on STARC-9 achieved up to 99% accuracy in tissue classification and significantly improved tumor segmentation Dice scores by 14-35% compared to existing public datasets.
Key insights
- A custom autoencoder trained with SSIM loss can effectively learn fine-grained, morphology-sensitive embeddings from histopathology images, proving more suitable for clustering diverse tissue patterns than general-purpose foundation models.
- Combining K-means clustering with an equal-frequency binning strategy enables systematic sampling of both prototypical and atypical representations within each cluster, ensuring morphologic diversity and balanced class representation.
- A semi-automated cluster labeling process, followed by WSI-level pathologist verification, substantially reduces the manual effort and subjectivity in dataset curation, enabling efficient creation of large-scale, high-quality datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.