Clinical Ultrasound Image Repository

This repository containes generic ultrasound data from random subjects, acquired for various clinical reasons from three different disciplines:

Abdomnial imaging (patients with prefix "A")
Cardiac imaging (patients with prefix "C")
Obstetrics and Gynaecology (patients with prefix "O")

All data is in DICOM format and was de-identified to be in accordance with HIPAA regulations:

Protected personal information was removed from DICOM elements and burnt-in image-pixel data.
Absolute dates are reduced to the year, with "19xx" or "20xx" used for patients age 90 or older.
An age information of "90+" is used for patients 90 and over at the time of imaging.
No data from patients younger than 18 years was used.
Private groups and Overlay groups were removed from the DICOM data.

Contact Information: shuver@nvidia.com
Managed by: MONAI Development Team

Content of the Repository

Studies		Abdominal	Cardiac	OB/Gyn	Overall
Number of Patients:		667	667	666	2000
Number of Studies:		667	667	666	2000
Number of Image Files:		24283	37356	25001	86640
Years of Imaging:		2012-2021	2014-2021	2018-2021	2012-2021
Age at Exam:	median	55yrs	58yrs	30yrs	39yrs
Age at Exam:	[5%, 95%]	[22, 80]	[20, 83]	[21, 39]	[21, 79]
Deceased:		12.29%	17.84%	0.150%	10.1%
Female:		54.27%	55.17%	100.0%
Trimester:		1st: 2nd		48.94% 51.05%
Manufacturers
Philips Medical Systems:		434	667	396	1497
GE Healthcare:		132		269	401
Siemens Healthineers:		82			82
Samsung Medison:		10		1	11
B-K Medical:		8			8
Acuson:		1			1

Access and Download

Using your browser from the download page
Following the instructions by Amazon as a starting point
Mounting the S3 bucket as an s3fs file system, e.g., by s3fs-fuse for Linux
Using the Python code (or an equivalent) below for bulk downloading

Example Python Code:

To download a JSON file with the cumulative meta information and all studies as per-study ZIP files:

host = "https://clinical-ultrasound-image-repository.s3.amazonaws.com"
httpGET(host+"/archives/meta-only/all-meta.json")

with open("all-meta.json","r") as metafile:
   studies = json.load(metafile)
   for study in studies:
      httpGET(host+"/archives/per-study/"+studies[study]["_ref"]+".zip")

where httpGET() wraps around your favorite HTTP library or tool; e.g., on Linux a simple

def httpGET(url):
    subprocess.call(["curl", url, "-o", os.path.basename(url)])

will do to utilize curl for the actual HTTP download.

File-Tree Structure

<patient>_<study-description>

images

dicom

<patient>-<series>-<scan>.dcm: DICOM Image file with meta group

meta

json

<patient>-meta.json: Hierarchical-dictionary JSON file

csv

<patient>-meta.csv: alternate Comma-Separated-Value file

archives

per-study

<patient>_<study-description>.zip: ZIP file with image and meta data following above structure

meta-only

all-meta.json: Combined JSON file as a single dictionary
all-meta.csv: Combined CSV file as a single sheet
all-meta.zip: ZIP file with all individual JSON and CSV files

JSON-Dictionary Structure

<patient>

_ref

<patient>_<study-description> (directory reference)

demo

<key>: <value> (sub-dictionary with demographic information)

exam

<key>: <value> (sub-dictionary with specifics on procedures)

labs

<type>

<value> (not tied to a specific year)

<year>

<type>: results for the given year, single or multiple entries (unit may be empty)
[<value>,<unit>]
[ [<value>,<unit>], [<value>,<unit>], ... ]

Comma-Separated-Value Structure

Description of Columns:

<patient>
Category: demo,exam,labs
<key> or <type> (from JSON dictionary)
<value>
<unit> (optional)
<year> (optional)