Repository: https://github.com/JhuangLab/BMSGC
Principal Investigator: Jhuanglab
Contact: hiekeen
Accurate diagnosis of bone marrow diseases from hematoxylin-eosin (HE)-stained whole-slide images (WSIs) remains challenging due to diffuse growth patterns, complex spatial heterogeneity, and overlapping morphological features. This repository implements an interpretable two-stage deep learning framework that closely mirrors the routine pathological diagnostic workflow, enabling automated, robust, and clinically aligned classification of bone marrow disorders.
- 🔀 Two-Stage Diagnostic Pipeline: Mimics real-world clinical workflow Stage 1: Normal vs. Abnormal screening → Stage 2: Disease subtyping
- 🌐 Region-Aware Graph Modeling: Captures spatial heterogeneity and long-range tissue topology for superior feature aggregation
- 📉 Distribution-Robust: Minimal performance degradation across temporal and cross-institutional shifts
- 🔍 Built-in Interpretability: Region-level attention maps highlight diagnostically relevant morphological patterns
- 🏥 Clinical-Ready Design: Optimized for integration into digital pathology pipelines
git clone https://github.com/JhuangLab/BMSGC.git
cd BMSGC
Install via pip:
pip install -r requirements.txt
requirements.txt includes:
torch==1.13.1 torchvision==0.14.1 numpy==1.24.3 pandas==2.0.2 matplotlib==3.7.1
scikit-learn==1.2.2 opencv-python==4.7.0.72 pillow==9.5.0 tqdm==4.65.0
lime==0.2.0.1 grad-cam==1.4.6 onnx==1.14.0 onnxruntime==1.15.1 seaborn==0.12.2
This project is licensed under the MIT License - see the LICENSE file for details.
Contact
For questions, issues, or collaboration requests, please contact:
-
Project Maintainer: JhuangLab
-
GitHub Issues: https://github.com/JhuangLab/BMSGC/issues
-
We thank our collaborator for providing clinical data.
-
We acknowledge the open-source community for the foundation models and tools used in this project.
