The Emory Knee Radiograph (MRKR) Dataset

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Orthopedics and Musculoskeletal Research

Background:

• Large-scale datasets are crucial for developing robust deep learning models in medical imaging.
• Existing knee radiograph datasets often lack comprehensive clinical information and demographic diversity.
• The Emory Knee Radiograph (MRKR) dataset addresses these limitations by providing extensive data.

Purpose of the Study:

• To introduce and describe the MRKR dataset, a large, demographically diverse collection of knee radiographs.
• To highlight the unique features of the MRKR dataset, including clinical data and imaging metadata.
• To facilitate research in osteoarthritis, pain management, and the development of equitable AI models.

Main Methods:

• Compilation of 503,261 knee radiographs from 83,011 patients.
• Inclusion of de-identified DICOM imaging data.
• Integration of detailed clinical information: patient-reported pain scores, diagnostic codes, procedural codes, image laterality, view type, and arthroplasty status.

Main Results:

• The MRKR dataset comprises 503,261 knee radiographs, with 40% from African American patients, ensuring demographic diversity.
• The dataset includes comprehensive clinical data and imaging metadata not commonly found in public repositories.
• Data are publicly available under a CC-BY-SA license, promoting open research.

Conclusions:

• The MRKR dataset is a valuable, openly accessible resource for advancing research in knee osteoarthritis and pain.
• Its rich data facilitates the development of equitable deep learning models for improved patient outcomes.
• Researchers can leverage this dataset to enhance diagnostic accuracy and treatment strategies for musculoskeletal conditions.