KneeMRI dataset was gathered retrospectively from exam records made on a Siemens Avanto 1.5T MR scanner, and obtained by proton density weighted fat suppression technique at the Clinical Hospital Centre Rijeka, Croatia, from 2006 until 2014. The dataset consists of 917 12-bit grayscale volumes of either left or right knees. Each volume record was assigned a diagnosis concerning the condition of the anterior cruciate ligament in a double-blind fashion, i.e. each volume record was labelled according to the ligament condition: (1) healthy, (2) partially injured, or (3) completely ruptured. A wider rectangular region of interest (ROI) was manually extracted from the original volumes, and is also annotated. For more details regarding the dataset, the reader is referred to the paper stated under the "acknowledging source" section of this webpage.
This dataset was built with the intention of providing scientists, involved with machine vision and/or machine learning, an easy way of working with the data.
An animation of sagittal-plane view T1-weighted MR sequence of a knee
This dataset is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND 4.0).
If you are using this dataset in your work, please acknowledge the source (Clinical Hospital Centre Rijeka, Croatia) and reference this paper (preprint pdf):
I. Štajduhar, M. Mamula, D. Miletić, G. Unal, Semi-automated detection of anterior cruciate ligament injury from MRI, Computer Methods and Programs in Biomedicine, Volume 140, 2017, Pages 151–164.
The knee MR volumes are saved as Python pickle objects. See example.py for details regarding accessing individual volumes.
File contents of this archive (approx. 3.2GB of data):
READ.ME - file contains some basic information regarding this archive
example.py - Python script used to demonstrate how to access the files
metadata.csv - comma-delimited (csv) data containing descriptions of distinct volume (header attribute information included)
volumetric_data - directory containing all knee MR volumes, archived using 7-zip losless file compression
|-example.pck - an example Python .pck file (just to inspect whether one wants to be bothered downloading the archive)
|-vol01.7z - compressed independent archive (1/10), containing 92 cases
|-vol02.7z - compressed independent archive (2/10), containing 92 cases
|-vol03.7z - compressed independent archive (3/10), containing 92 cases
|-vol04.7z - compressed independent archive (4/10), containing 92 cases
|-vol05.7z - compressed independent archive (5/10), containing 92 cases
|-vol06.7z - compressed independent archive (6/10), containing 92 cases
|-vol07.7z - compressed independent archive (7/10), containing 92 cases
|-vol08.7z - compressed independent archive (8/10), containing 92 cases
|-vol09.7z - compressed independent archive (9/10), containing 92 cases
|-vol10.7z - compressed independent archive (10/10), containing 89 cases
If you have any questions, please contact Ivan Štajduhar:
University of Rijeka - Faculty of Engineering