A large, curated, open-source stroke neuroimaging dataset to improve lesion segmentation algorithms

• Type of Material: Article
• Authors: Sook-Lei Liew, University of Southern California; Bethany P Lo, University of Southern California; Miranda R Donnelly, University of Southern California; Artemis Zavaliangos-Petropulu, University of Southern California; Jessica N Jeong, University of Southern California; Giuseppe Barisano, University of Southern California; Alexandre Hutton, University of Southern California; Julia P Simon, University of Southern California; Julia M Juliano, University of Southern California; Anisha Suri, University of Pittsburgh; Zhizhuo Wang, University of Southern California; Aisha Abdullah, University of Southern California; Jun Kim, University of Southern California; Tyler Ard, University of Southern California; Nerisa Banaj, IRCCS Santa Lucia Foundation; Michael Borich, Emory University; Lara A Boyd, University of British Columbia; Amy Brodtmann, University of Melbourne; Cathrin Buetefisch, Emory University; Lei Cao, Child Mind Institute, New York; Jessica M Cassidy, University of North Carolina Chapel Hill; Valentina Ciullo, IRCCS Santa Lucia Foundation; Adriana B Conforto, São Paulo University; Steven C Cramer, University of California Los Angeles; Rosalia Dacosta-Aguayo, University of Barcelona; Ezequiel de la Rosa, icometrix, Leuven; Martin Domin, University of Greifswald; Adrienne N Dula, University of Texas Austin; Wuwei Feng, Duke University; Alexandre R Franco, Child Mind Institute, New York; Fatemeh Geranmayeh, Imperial College London; Alexandre Gramfort, Université Paris-Saclay; Chris M Gregory, Medical University of South Carolina; Colleen A Hanlon, Wake Forest School of Medicine; Brenton G Hordacre, University of South Australia; Steven A Kautz, Medical University of South Carolina; Mohamed Salah Khlif, The Florey Institute of Neuroscience and Mental Health; Hosung Kim, University of Southern California; Jan S Kirschke, Technical University of Munich; Jingchun Liu, Tianjin Medical University General Hospital; Martin Lotze, University of Greifswald; Bradley J MacIntosh, University of Toronto; Maria Mataró, University of Barcelona; Feroze B Mohamed, Jefferson Magnetic Resonance Imaging Center; Jan E Nordvik, CatoSenteret Rehabilitation Cente; Gilsoon Park, University of Southern California; Amy Pienta, University of Michigan; Fabrizio Piras, IRCCS Santa Lucia Foundation; Shane M Redman, University of Michigan; Kate P Revill, Emory University; Mauricio Reyes, University of Bern; Andrew D Robertson, Schlegel-University of Waterloo Research Institute for Aging; Na Jin Seo, Medical University of South Carolina; Surjo R Soekadar, Charité - Universitätsmedizin Berlin; Gianfranco Spalletta, IRCCS Santa Lucia Foundation; Alison Sweet, University of Michigan; Maria Telenczuk, Université Paris-Saclay; Gregory Thielman, St. Joseph’s University; Lars T Westlye, University of Oslo; Carolee J Winstein, University of Southern California; George F Wittenberg, Geriatrics Research, Education and Clinical Center; Kristin A Wong, University of Texas Austin; Chunshui Yu, Tianjin Medical University General Hospital
• Language: English
• Date: 2022-06-16
• Publisher: NATURE PORTFOLIO
• Publication Version: Final Published Version
• Copyright Statement: © The Author(s) 2022
• License: Creative Commons Attribution 4.0 International
• Final Published Version (URL): http://dx.doi.org/10.1038/s41597-022-01401-7
• Title of Journal or Parent Work: SCIENTIFIC DATA
• Volume: 9
• Issue: 1
• Start Page: 320
• End Page: 320
• Keywords: Science & Technology; NET; Multidisciplinary Sciences; BIOMARKERS; FUSION NETWORK; Science & Technology - Other Topics; RECOVERY
• Research Categories: Computer Science; Health Sciences, Mental Health; Health Sciences, Rehabilitation and Therapy

Abstract

Accurate lesion segmentation is critical in stroke rehabilitation research for the quantification of lesion burden and accurate image processing. Current automated lesion segmentation methods for T1-weighted (T1w) MRIs, commonly used in stroke research, lack accuracy and reliability. Manual segmentation remains the gold standard, but it is time-consuming, subjective, and requires neuroanatomical expertise. We previously released an open-source dataset of stroke T1w MRIs and manually-segmented lesion masks (ATLAS v1.2, N = 304) to encourage the development of better algorithms. However, many methods developed with ATLAS v1.2 report low accuracy, are not publicly accessible or are improperly validated, limiting their utility to the field. Here we present ATLAS v2.0 (N = 1271), a larger dataset of T1w MRIs and manually segmented lesion masks that includes training (n = 655), test (hidden masks, n = 300), and generalizability (hidden MRIs and masks, n = 316) datasets. Algorithm development using this larger sample should lead to more robust solutions; the hidden datasets allow for unbiased performance evaluation via segmentation challenges. We anticipate that ATLAS v2.0 will lead to improved algorithms, facilitating large-scale stroke research.