TL;DR
Neuroimaging research is hampered by the high cost and time required for data collection, leading to data silos and duplicated efforts. The COINS Data Exchange (DX) addresses this by providing an open-source platform that simplifies the compilation, curation, and dissemination of neuroimaging data, fostering collaboration and accelerating scientific discovery.
What Problem Does This Paper Address?
Neuroimaging studies are incredibly expensive and time-consuming to conduct. This high barrier to entry means that valuable data, once collected, is often kept private by researchers until they have fully analyzed and published their findings. This practice creates several problems:
- Missed Opportunities: Other researchers in the field might be able to make significant breakthroughs by analyzing this data, but they can't access it.
- Duplicated Effort: Without access to existing datasets, researchers may unnecessarily collect similar data, wasting resources and time.
- Technical Hurdles: Sharing data across different institutions and formats is technically challenging, further discouraging data sharing.
The paper argues that these technical and cultural barriers need to be overcome to advance neuroscience.
Methods
The COINS Data Exchange (DX) tackles the data sharing problem with a three-pronged approach:
- Data Compilation: DX acts as a central repository, bringing together data collected from various sites worldwide, often in different formats. It integrates with the Collaborative Informatics Neuroimaging Suite (COINS) for seamless data ingestion from studies already using COINS. For data not managed by COINS, an "ad-hoc" upload tool allows researchers to upload their data after providing basic study information. DX supports a wide variety of data formats, including raw and pre-processed DICOM, behavioral data, and theoretically any other data type.
- Data Curation: Once data is compiled, DX ensures its quality and consistency. It organizes data into a meaningful structure and allows for fine-grained control over data sharing. This includes the ability to exclude data that is later found to be of poor quality or disqualified for other reasons. Automated processes anonymize sensitive DICOM headers to protect participant privacy.
- Data Dissemination: DX provides an intuitive graphical user interface (GUI) for data requesters to easily search, query, and download specific datasets. This targeted approach aims to prevent the "shotgun method" of downloading massive amounts of data, saving bandwidth and storage costs for both providers and consumers. Data Usage Agreements (DUAs) are managed through the platform, allowing studies to define their own terms of use. Study administrators can approve or deny requests, or make data "pre-approved" for easier access.
Datasets
The paper describes the COINS Data Exchange as a platform that hosts data, rather than a platform that introduces new datasets for evaluation. It mentions that COINS currently hosts over 550 studies, 400,000 cognitive assessments, and 350,000 functional MRI images from 30,000 participants. DX itself is currently sharing 55,000 cognitive assessments and 16,000 functional MRI images from 4,100 participants.
The paper uses examples of queries to illustrate the platform's capabilities, referencing the MRN COBRE study and the Autism Brain Imaging Data Exchange (ABIDE) dataset.
Key Insights and Ideas
- The "Three-Pronged" Approach is Key: The combination of compilation, curation, and dissemination is crucial for making data sharing effective and sustainable.
- Flexibility in Data Formats: DX's ability to handle a wide array of data formats is a significant advantage, accommodating the diverse nature of neuroimaging data.
- Granular Control for Data Providers: Empowering data providers with fine-grained controls over who can access their data and under what conditions is essential for encouraging sharing.
- Targeted Queries for Efficient Access: The emphasis on precise querying and downloading reduces the burden on data consumers and promotes more efficient use of resources.
- Building Trust Through Transparency: The DUA system and the ability for study administrators to communicate directly with requesters build trust and transparency in the data sharing process.
Key Results
- As of the paper's writing, DX has over 600 users who have queried the database.
- DX has packaged over 31 TB of data from nearly 1200 unique data requests.
- The platform supports a wide range of data formats, including
.bval,.bvec,.csv,.dat,.dcm,.edat,.gz,.hdr,.img,.info,.mat,.nhdr,.nii,.nrrd,.raw,.txt, and.zip. - The platform is designed to be accessible via modern web browsers.
Limitations and Future Work
- Current Centralization: DX is currently implicitly linked to the COINS architecture and hosted solely by MRN. Data must be uploaded to MRN's local servers.
- Future Standalone Package: MRN is developing a standalone DX package that would allow researchers to host their own data sharing platforms, mimicking the flexibility of COINS' data collection functions. This would allow data providers to keep their data onsite and maintain closer control.
- Cost of Data Provision: While data storage and serving costs have decreased, there are still costs associated with being a data provider. The paper notes that these costs are spread across multiple funding sources.
Summary updated