Emory Knee Radiograph (MRKR) Dataset

Data description of the Emory Knee Radiograph (MRKR) Dataset hosted by Open Data on AWS.

An example notebook is included.

Summary

The Emory Knee Radiograph (MRKR) dataset is a large, demographically diverse collection of 503,261 knee radiographs from 83,011 patients, 40% of which are African American. This dataset provides imaging data in DICOM format along with detailed clinical information, including patient- reported pain scores, diagnostic codes, and procedural codes, which are not commonly available in similar datasets. The MRKR dataset also features imaging metadata such as image laterality, view type, and presence of hardware, enhancing its value for research and model development. MRKR addresses significant gaps in existing datasets by offering a more representative sample for studying osteoarthritis and related outcomes, particularly among minority populations, thereby providing a valuable resource for clinicians and researchers.

Dataset Size

The total dataset is 2.3 TB and includes DICOMs (2.3 TB) and seven CSV files (2.7 GB) each containing clinical and metadata.

The dataset will be hosted with Open Data on AWS.

Publication

Will be available soon

License

CC-BY-SA

MRKR Dataset Patient Summary Statistics

Total Patients	83,011 (100%)
Gender
Female	51,175 (61.6%)
Male	31,836 (38.4%)
Age, years
Mean	56.6 (std. +/- 16.6)
Median	58
Race demographics
White	36,927 (44.5%)
Black	33,503 (40.4%)
Asian	2,893 (3.5%)
Unknown/Unreported	8,751 (10.5%)
Other	937 (1.1%)
Ethnicity
Hispanic	2,501 (3.0%)
Non-Hispanic	66,378 (80.0%)
Unknown/Unreported	14,132 (17.0%)
Clinical outcomes
Arthroplasty	14,843 (17.9%)

Dataset Structure

Filename: MRKR_CPT.csv File size: 178 MB Total rows: 6,216,190 Description: This table contains information regarding all CPT codes for a patient and corresponding dates.
Field Name	Data Type	Description
empi_anon	Integer (8 digits)	Unique patient identification number (83,011 patients)
cpt_code	String (5 characters)	Current Procedural Terminology code used in coding of medical services and procedures for billing (7,166 CPT codes)
cpt_group_modifier	String	Used to provide further information regarding service or procedure. Most CPT codes do not include modifier data. If there is modifier data, it is often used to indicate the laterality of a procedure (left or right). There can be multiple modifiers for a single CPT code entry.
date_anon	Date	Date of when the associated procedure or service occurred.
age_at_procedure	Integer	Age when the procedure was performed.

Filename: MRKR_CPT_dictionary.csv File size: 754 KB Total rows: 7,166 Description: A lookup table between CPT codes and corresponding descriptions.
Field Name	Data Type	Description
cpt_code	String (5 characters)	Current Procedural Terminology code used in the coding of medical services and procedures for billing.
cpt_description	String	Description of the procedure. There are some unique CPT codes that share the same description.

Filename: MRKR_ICD.csv File size: 1.7 GB Total rows: 21,956,056 Description: ICD9 and ICD10 diagnoses for patients with corresponding dates. Certain diseases of interest are indicated by binary flags to ease data cleaning.
Field Name	Data Type	Description
empi_anon	Integer (8 digits)	Unique patient identification number (83,011 unique patients)
ICD9	String	International Classification of Diseases - 9 (12,418 unique codes)
ICD10	String	International Classification of Diseases - 10 (26,963 unique codes)
date_anon	Date	Date of when the diagnosis code was entered.
age_at_dx	Integer	Age when the diagnosis was recorded.
DX_LINE	String	Primary, Secondary, Active, Not Recorded, Resolved, Canceled, Inactive.
DX_ICD_SCOPE	String	Billing Diagnosis, Discharge Diagnosis, Admitting Diagnosis, Referring Diagnosis, Not Recorded, Reason For Visit, Problem List, Working Diagnosis, Other Diagnosis, Final, Pre-Op Diagnosis, Post-Op Diagnosis, Principal Diagnosis, Suggested Billing.
autoimmune	Binary	If ICD code corresponds to auto-immune disease such as rheumatoid arthritis, juvenile arthritis, gout, etc.
diabetes	Binary	If ICD code corresponds to type I or type II diabetes.
hypertension	Binary	If ICD code corresponds to hypertension.
joint_infection	Binary	If ICD code corresponds to a knee joint infection.
knee_osteoarthritis	Binary	If ICD code corresponds to knee osteoarthritis.
knee_osteomyelitis	Binary	If ICD code corresponds to knee osteomyelitis.
obesity	Binary	If ICD code corresponds to obesity.
nicotine_use	Binary	If ICD code corresponds to nicotine dependence.
trauma_lower_extremity	Binary	If ICD code corresponds to lower extremity trauma.

Filename: MRKR_ICD_dictionary.csv File size: 1.9 MB Total rows: 25,209 Description: Lookup table for ICD9 (International Classification of Diseases) and ICD10 codes and corresponding descriptions.
Field Name	Data Type	Description
ICD9	String	ICD9 code.
ICD10	String	ICD10 code.
DX_NAME	String	Diagnosis name or description.

Filename: MRKR_pain.csv File size: 137 MB Total rows: 4,975,933 Description: Contains information on self-reported pain scores by patients during any encounter, including outpatient, emergency, and perioperative. Pain scores related to knees are curated.
Field Name	Data Type	Description
empi_anon	Integer (8 digits)	Unique patient identification number (83,011 unique patients)
pain_location	String	Raw, uncurated strings of pain locations entered by staff. Approximately 75% of entries are blank.
knee_pain	Binary	Curated using regular expressions to identify if the pain_location is definitely knee related.
pain_score	Integer	0 - 10 pain score.

Filename: MRKR_demographics.csv File size: 4.5 MB Total rows: 83,011 Description: Patient demographics, indexed at the patient level.
Field Name	Data Type	Description
empi_anon	Integer (8 digits)	Unique patient identification number.
sex	Nominal string	[male, female] - Patient sex.
race	Nominal string	[African American or Black, American Indian or Alaskan Native, Asian, Caucasian or White, Multiple, Native Hawaiian or Other Pacific Islander, Unknown] - Patient self-reported race.
ethnicity	Nominal string	[Hispanic patients, Non-Hispanic patients, Unknown, Unreported] - Patient reported ethnicity.

Filename: MRKR_image_metadata.csv File size: 210 MB Total rows: 503,261 Description: Contains relevant public DICOM metadata tags that may be helpful for identifying images. Patient and exam identifiers are replaced with de-identified versions in this table and within DICOM files. Other Non-PHI containing metadata tags that are not in this table are left intact within DICOM files. Fields containing PHI such as patient name, addresses, or referring physician are removed from this table and DICOM files. For data curation, the below fields were modified or added.
Field Name	Data Type	Description
empi_anon	Integer (8 digits)	De-identified patient identification number.
StudyInstanceUID_anon	String	De-identified Study UID, shared between all images in the same study.
SeriesInstanceUID_anon	String	De-identified Series UID, shared between all images in the same series.
SOPInstanceUID_anon	String	De-identified SOP Instance UID which corresponds to a single DICOM image.
img_height	Integer	Image pixel height.
img_width	Integer	Image pixel width.
laterality	Nominal string	[R: Right, L: Left, B: Bilateral, -1: Unknown or not present] - Laterality of the image, as inferred by DL model.
view_position	Nominal string	[F: Frontal, L: Lateral, S: Sunrise, I: Internal Oblique, E: External Oblique] - Anatomical projection of radiograph, as inferred by DL model.
horizontal_flip	Binary	Indicates if the patient’s left side was oriented to the left side of the image, which is opposite of typical radiographic orientation, as inferred by DL model.
weight_bearing	Binary	Indicates if the radiograph was weight-bearing as indicated by a marker and derived by DL model. Not all images in a given exam will be weight-bearing or non-weightbearing.
inverted	Binary	Indicates whether pixel intensity values are inverted from typical radiographic convention, as inferred by DL model.
arthroplasty	Nominal string	[R: right, L: left, B: bilateral, NL: unknown (no laterality marker), NaN: no arthroplasty] - Indicates if image contains a knee arthroplasty and its laterality, as derived by DL model.
L_KLG_inference	Integer [0,1,2,3,4,NaN]	KLG score of left knee in a bilateral knee radiograph, inferred by DL model.
R_KLG_inference	Integer [0,1,2,3,4,NaN]	KLG score of right knee in a bilateral knee radiograph, inferred by DL model.
SeriesDescription	String	DICOM Metadata describing the series.
StudyDescription	String	DICOM metadata describing the study.
StudyDate_anon	Date	De-identified date of radiograph.
age_at_exam	Integer	Age of the patient when the radiograph was performed.
dicom_path	String	Path to DICOM file.