Reproduce “Slideflow: Deep Learning for Digital Histopathology with Real-Time Whole-Slide Visualization” 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
James M. Dolezal, Sara Kochanny, Emma Dyer, Andrew Srisuwananukorn, Matteo Sacco, Frederick M. Howard, Anran Li, Prajval Mohan and 1 more
The University of Chicago Medical Center and collaborators developed Slideflow, an open-source Python package for deep learning in digital histopathology, offering efficient data processing, comprehensive model training, and real-time whole-slide visualization. It achieved an AUROC of 0.87 for HPV status prediction on an external TCGA dataset and demonstrated tile extraction speeds up to 3.3 times faster than existing backends.
Key insights
- Significant gains in computational efficiency for WSI processing and dataset iteration can be achieved through optimized tile extraction strategies and data storage formats, making deep learning more accessible.
- Integrating a suite of explainable AI (XAI) tools and uncertainty quantification methods within a single platform is vital for enhancing model interpretability and fostering clinician trust in pathology AI.
- Developing a unified, framework-agnostic platform streamlines the entire deep learning workflow for digital histopathology, enabling rapid experimentation and broad deployment across various hardware environments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.