MRKR: 503,261 Knee Radiographs with Clinical Data from 83,011 Patients

503,261 knee radiographs in DICOM format from 83,011 patients, with 40% of patients identified as African American. The dataset includes detailed clinical information such as patient-reported pain scores, diagnostic codes, and procedural codes, along with imaging metadata like laterality and view type. It is hosted on AWS Open Data and licensed under CC-BY-SA-4.0.

Use Cases

• Training computer vision models for knee pathology detection based on DICOM radiographs.
• Studying correlations between imaging findings and patient-reported pain scores.
• Analyzing demographic disparities in osteoarthritis diagnosis and treatment based on diagnostic and procedural codes.
• Developing multimodal models that combine image data with structured clinical metadata.

Strengths

• Large scale with 503,261 images from 83,011 patients.
• Demographically diverse sample with 40% African American patients.
• Includes clinical data not commonly available, such as pain scores and diagnostic codes.
• Provides detailed imaging metadata including laterality, view type, and hardware presence.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

Emory University, with acknowledgment to MD.ai for image de-identification.

Collection Method

Collection of clinical knee radiographs and associated patient data.

Time Range

null

Freshness

Last updated date is unknown.

Geography

null