Overview
Accurate lesion segmentation is critical for stroke rehabilitation research for both quantification of lesion burden and accurate image processing. Current automated lesion segmentation methods for T1-weighted MRIs lack the accuracy to be used reliably in research. Manual segmentation remains the gold standard, but it is time-consuming and requires significant neuroanatomical expertise. We previously released a large, open-source dataset of stroke T1-weighted MRIs and manually segmented lesion masks (ATLAS v1.2, N=304) to encourage the development of better segmentation algorithms. However, most methods developed with ATLAS v1.2 are not publicly accessible or have been overfitted to the data, resulting in algorithms with poor performance on new data. Here we present ATLAS v2.0 (n=955), a larger dataset of stroke T1-weighted MRIs and lesion masks that includes both training (public) and test (hidden) data. Algorithm development using this larger sample should lead to more robust solutions, and the hidden test data allows for unbiased performance evaluation via web-based challenges. We anticipate that ATLAS v2.0 will lead to the development of improved lesion segmentation algorithms, facilitating large-scale stroke research.
The dataset includes:
- 955 T1-weighted MRI scans, divided into a training dataset (n=655 T1w MRIs with manually-segmented lesion masks) and a test dataset (n=300 T1w MRIs only; lesion masks not released)
- MNI152 standard-space T1-weighted average structural template image
- Two .csv files containing lesion and scanner metadata
Release Download
Note: In order to obtain the password needed to access the data, you must agree to the Terms of Use specified on this page and complete a brief Google form. After completing the Google Form, your completed application will be reviewed, and an encryption code to access the data will then be emailed to you. Please note that this process should be completed for each individual accessing the data, and the encryption code should not be shared with others.
Release Notes
Release R1.2: N=304
Release R2.0:- N=955. Dataset now includes training and testing data. Data has been reorganized to BIDS derivatives format.
Data Descriptor Manuscript for R2.0
XXX
Data Descriptor Manuscript for R1.2
Click here to access the bioRxiv preprint for ATLAS R1.2.
doi: https://doi.org/10.1101/179614
Click here to access the printed version of the data descriptor in Scientific Data.
Authors
Sook-Lei Liew1†*, Julia M. Anglin1*, Nick W. Banks1, Matt Sondag1, Kaori L. Ito1, Hosung Kim1, Jennifer Chan1, Joyce Ito1, Connie Jung1, Stephanie Lefebvre1, William Nakamura1, David Saldana1, Allie Schmiesing1, Cathy Tran1, Danny Vo1, Tyler Ard1, Panthea Heydari1, Bokkyu Kim1, Lisa Aziz-Zadeh1, Steven C. Cramer2, Jingchun Liu3, Surjo Soekadar4, Jan-Egil Nordvik5, Lars T. Westlye6,7, Junping Wang3, Carolee Winstein1, Chunshui Yu3, Lei Ai8, Bonhwang Koo8, R. Cameron Craddock8,9, Michael Milham8,9, Matthew Lakich10, Amy Pienta11, Alison Stroud11
Corresponding author: Sook-Lei Liew (sliew@usc.edu)
*Denotes equal contributions
Affiliations
- University of Southern California, Los Angeles, California, USA
- University of California, Irvine, Irvine, California, USA
- Tianjin Medical University General Hospital, Tianjin, China
- University of Tübingen, Tübingen, Germany
- Sunnaas Rehabilitation Hospital HT, Nesodden, Norway
- NORMENT and KG Jebsen Centre for Psychosis Research, Division of Mental Health and Addiction, Oslo University Hospital, Oslo, Norway
- Department of Psychology, University of Oslo, Oslo, Norway
- Child Mind Institute, New York, New York, USA
- Nathan S. Kline Institute for Psychiatric Research, Orangeburg, New York, USA
- University of Texas Medical Branch, Galveston, Texas, USA
- University of Michigan, Ann Arbor, Michigan, USA
Acknowledgements !!NEED TO UPDATE !!!
We would like to acknowledge the following people for their assistance on this effort: Anthony Benitez, Xiaoyu Chen, Cristi Magracia, Ryan Mori, Dhanashree Potdar, Sandyha Prathap.
The archiving of this dataset was specifically supported by the NIH-funded Center for Large Data Research and Data Sharing in Rehabilitation (CLDR) under a Category 2 Pilot Grant (P2CHD06570) and this work was also funded by an NIH K01 award (1K01HD091283).