COVID-19 Image Repository
This project aims to create an anonymized data set of COVID-19 cases with a focus on radiological imaging. This includes images with extensive metadata, such as admission-, ICU-, laboratory-, and patient master-data.
This repository contains image data from the Institute for Diagnostic and Interventional Radiology, Hannover Medical School, Hannover, Germany.
Feature Set
| id | label | unit | comment / reference interval (precision) |
|---|---|---|---|
| 1 | patient_id | randomly generated patient id | |
| 2 | image_id | randomly generated image id (filename) | |
| 3 | sex | m/w | |
| 4 | age | years | currently redacted (see below) |
| 5 | size | cm | currently redacted (see below) |
| 6 | weight | kg | currently redacted (see below) |
| 7 | admission offset | days | days since admission (begin of symptoms) |
| 8 | icu admission offset | days | |
| 9 | death offset | days | days until death or null |
| 10 | modality | currently all images are chest radiographs | |
| 11 | projection | ap, pa ... | |
| 12 | lactate dehydrogenase | U/l | < 248 (5) |
| 13 | c-reactive protein | mg/l | <= 5 (1) |
| 14 | d-dimer | mg/l | 0 - 0.5 (0.1) |
| 15 | coagulation factor XIII | % | 70 - 140 (5) |
| 16 | neutrophils | Tsd/µl | 1.5 - 7.7 (0.1) |
| 17 | lymphocytes | Tsd/µl | 1.1 - 4.0 (0.1) |
| 18 | pO2 | mmHg | (5) |
| 19 | pCO2 | mmHg | (5) |
| 20 | corona test type |
Age, size, and weight are currently redacted. We will publish this data when there are enough patients, that meaningful intervals can be chosen according to the concept of k-anonymity and l-diversity.
Offsets, i.e. admission offset and icu admission offset, are given as relative times in regard to the exam. Please consult the feature set table for units. I.e. an admission offset == -4 and icu admission offset == 6 would encode, that the patient was admitted to the hospital four days ago and was transferred to the ICU six days after the image was taken. Especially the admission offset can be noisy; please see FAQ #1.
All lab values (12 - 19) are given in intervals to protect patient identity. A value below the detection limit is denoted by -inf, above the detection limit by inf.
Download
We provide the raw, unprocessed, gray value image data as Nifti files. This is done to protect patient identity, as Dicom files are hard to anonymize. However, the files are too large to host on Github.
Additionally, we included downscaled versions of the Nifti images in the png folder.
Space and bandwith are kindly provided by the Open Telekom Cloud (Deutschen Telekom AG).
Download Links: https://doi.org/10.25835/0090041
-
Do not use this version (see erratum 1). The data is provided for reproducibility reasons.
Observations (FAQ)
- admission offset > icu admission offset
Some of the cases suggest that the admission to the clinic was after admission to the ICU. This is an artifact due to using the specific (COVID) case data for determining the admission date (and therefore offset). The patient may have already been at the ICU before being diagnosed with COVID.
Errata
- In Version 1.0 there is an issue with the date calculation (issue #6). This might lead to incorrect date offset calculations. We strongly advise to upgrade to a newer version.
License and Attribution
Master-, image- and laboratory-data of this repository are licensed under the Creative Commons Attribution 3.0 Unported (CC BY 3.0).
If you use this data, you must attribute the authors in any publication (DOI: 10.6084/m9.figshare.12275009). You may include the specific release or commit hash for reproducibility.
Contact Information
Please use the ticketing system where applicable. Otherwise, please use the following EMail address: [email protected]