GREMLIN: Graph Estimation From MR Images Leading to Inference in Neuroscience

Connectome;MRI;DTI;pipeline;brain graph;neuroscience;not listed

Kiar, GregoryVogelstein, Joshua T. ;

Johns Hopkins University

关键词: Connectome;    MRI;    DTI;    pipeline;    brain graph;    neuroscience;    not listed;   

Others  :  https://jscholarship.library.jhu.edu/bitstream/handle/1774.2/39487/KIAR-THESIS-2016.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: JOHNS HOPKINS DSpace Repository

【 摘 要 】

In recent years there has been a growing desire to understand the structure and function of the human brain. Approximately 1 in 5 adults suffers from mental illness, and many of these illnesses, including Alzheimer;;s Disease, Autism Spectrum Disorders, ADHD, and Schizophrenia could be described as connectopathies and may appear when observing the connectome (structural map of the brain). To this end, an abundance of MRI datasets have been collected around the globe. Of particular interest when seeking a connectome are the diffusion weighted (DTI) and structural (MPRAGE) sequences. Tools have been developed to process these brain images and enable quantitative analysis of brain structure. However, these tools often require computational expertise, and there exist few options to perform end-to-end analysis of MR images easily. Previous iterations of end-to-end connectome estimation pipelines have been limited in their ability to run at scale in parallel and have complex dependencies and setup routines.We have developed a one-click open-source pipeline which allows for the reliable estimation of connectomes from MR data across multiple scales. The pipeline produced, ndmg, has been engineered to optimize the discriminability of resulting graphs across many datasets, effectively optimizing the lower bound of predictive accuracy for any downstream inference task. The ndmg pipeline has been used to generate connectomes from all known redistributableDTI and MPRAGE datasets to date, resulting in over 5,000 subjects processed and over 100,000 estimated connectomes across multiple scales. All of the connectomes we produced are made available through our graph database, MR-GRUTEDB. The code for this open-source pipeline is available at http://m2g.io.A web service, C4, also exists in which users can upload their MRI data and receive an estimated connectome in return at no cost. These tools lower the barrier for entry to connectomics by removing significant computational duress from researchers. This pipeline empowers reproducible science by abstracting hyper-parameter selection and over-fitting opportunities from researchers when processing their data, and enables mega-analysis of MR data across sites and studies, further opening the door for interesting and powerful scientific discovery.