Join the RAM‑10000 Collaboration
We are building a multi-institutional effort to advance medical imaging AI in rheumatoid arthritis. Partners can contribute data, methods, and clinical expertise to accelerate an open-source dataset, trustworthy models, and real‑world impact. Co‑authorship and joint funding proposals are welcome.
About the Collaboration
RAM‑10000 is an open, multi‑institution effort to advance clinically useful AI for rheumatoid arthritis and related medical imaging tasks (detection, segmentation, diagnostics, report generation, VQA) in radiographs.
Our near‑term goal is a standardized corpus of 10k+ cases to enable robust cross‑center evaluation while respecting patient privacy and institutional policies.
We welcome de‑identified clinical data contributions.
You’ll gain
- Shared authorship on papers, benchmarks, and releases.
- Reproducible codebases and MLOps templates for rapid experimentation.
- Access to baselines, evaluation servers, and cross‑site validation.
You’ll commit
- Share de‑identified clinical data to help build the multi‑region, multi‑center RA dataset.
- Open and responsible research practices.
- Follow collaboration agreement, DUA/IRB, and attribution policies.
Authorship
Collaborators will retain authorship positions on all publications in peer-reviewed journals that will result from this project. The project leads will retain first and second authorship positions for the main articles that will result from this project. Core team members will then be listed according to their contributions. All collaborators will then be listed in alphabetical order according to their surname.
RA Multi‑Region, Multi‑Center Dataset
Why this dataset?
- RA currently lacks large, openly accessible datasets spanning diverse centers and devices.
- There is no widely adopted, unified evaluation platform, fragmented metrics hinder fair comparison and progress.
- A curated, multi‑center resource will accelerate method development, external validation, and clinical translation.
Focus Areas
Rheumatoid Arthritis Imaging
JSW/JSN quantification, bone erosion scoring, and prognosis from multi‑site hand/wrist radiographs and MRI.
Multimodal & VLMs
Vision‑language models for image–report learning, retrieval‑augmented VQA, and report generation.
Fairness & Robustness
Cross‑center generalization, bias checks, uncertainty quantification, and external validation protocols.
Evaluation & Benchmarks
Transparent leaderboards with clinically meaningful metrics and human‑in‑the‑loop audits.
Open Tooling
De‑ID, DICOM tooling, labeling UIs, and privacy‑preserving analytics (federated / DP options).
Education
Seminars, shared reading groups, and student exchanges across institutions.
Selected Publications
-
RAM-W600: A Multi-Task Wrist Dataset and Benchmark for Rheumatoid Arthritis
-
Layer Separation: Adjustable Joint Space Width Images Synthesis in Conventional Radiography
-
BLS-GAN: A Deep Layer Separation Framework for Eliminating Bone Overlap in Conventional Radiographs
-
A Sub-Pixel Accurate Quantification of Joint Space Narrowing Progression in Rheumatoid Arthritis