Noelle Toumey Reetz
Office of the Vice President for Research and Economic Development
ATLANTA—Three research teams are the winners of an international competition involving more than 1,000 teams of researchers in a brain-mapping data challenge hosted by Georgia State University and the Center for Translational Research in Neuroimaging and Data Science (TReNDS).
The first-place team was Team Neo Cortex & N. Gin (Nikita Churkin, Dmitry Simakov) from Russia, which received $12,000. Team No Brainer (Kazuki Onodera) from Japan came in second place and was awarded $8,000. The third-place team, Team No Magic (Naoto Shimakoshi) from Japan, won $5,000 for its work.
Researchers and data science practitioners from around the world analyzed a large dataset of brain images and built models that could assist with predicting the course of disorders like schizophrenia and depression, as well as patient outcomes.
The competition provided teams with access to the largest normative dataset of its kind, nearly 12,000 magnetic resonance imaging (MRI) scans from thousands of subjects from the UK Biobank, a large, long-term health study in the United Kingdom. Using this set of images without any known disease history, the teams developed models that predict a subject’s age and other attributes. The dataset will be used as a basis for comparison with patients in a clinical setting, helping doctors and researchers better understand and diagnose mental illness.
Each team worked to ensure their predictive models are unbiased by differences in the scanners used to capture the images. Brain researchers face a number of challenges due to differences between MRI scanners that can affect results, so identifying strategies to overcome this issue would represent a major breakthrough.
The competition was led by Georgia State research scientist Rogers Silva, who spent more than a year in preparation.
“The mission of the TReNDS center is to translate our findings to clinical settings,” said Silva. “The first step towards that is to build a strong baseline from which you can extrapolate information about the brain and patient outcomes, and then compare those to the brains and outcomes of patients with specific diseases.”
The competition was hosted with Kaggle, home to one of the world’s largest data science communities, and was co-sponsored by the Organization for Human Brain Mapping (OHBM), the Institute of Electrical and Electronics Engineers (IEEE) Signal Processing Society Data Science Initiative, and the IEEE Challenges and Data Collections Program. The OHBM seeks to advance understanding of the anatomical and functional organization of the human brain and promote its medical and societal applications. The IEEE is dedicated to advancing technology for the benefit of humanity.
Once the teams submitted their predictions, the judging committee at TReNDS used the university’s newly built high-performance computing cluster to review and process the data. The technology allows massive amounts of information to be analyzed and processed quickly.
“We were able to leverage the high-performance computing and analytics capabilities at TReNDS to download and prepare multimodal brain imaging data from over 10,000 individuals for use in the first UK biobank-based competition ever, and move the needle forward on brain imaging based predictive modeling,” said Vince Calhoun, Distinguished University Professor of Psychology at Georgia State and founding director of TReNDS.
The organizing team at TReNDS plans to evaluate the winning submission and those from other interested teams. Silva and Calhoun believe the separate evaluation will be crucial to further assess the generalization of these models to data from multiple sites, which is important for clinical applications.
TReNDS is a tri-institutional center supported by Georgia State, the Georgia Institute of Technology and Emory University. The center is focused on making better use of complex brain imaging data through improved analysis, with a goal of identifying biomarkers that can help address brain health and disease.