This case study should resonate with anyone frustrated by trying to use a DICOM PACS database to manage their research.Says David Gutman, "Instead of having to conform to the PACS data, I can make the data conform to me."
Introduction to Using XNAT for Clinical Research
The wealth of imaging research that is becoming available has the ability to directly influence clinical care; likewise, patients receiving clinical care may elect to participate in ongoing research into their condition, even if they cannot directly benefit.
Clinical imaging research includes both retrospective studies, which rely on batches of historic patient scans meeting some diagnostic criteria, and prospective studies, which rely on individual patients scans to be imported and often processed in real-time, sometimes during a surgical procedure. From an imaging informatics perspective, support for clinical research requires integration with clinical devices and information systems, careful compliance with regulatory requirements; and agility in moving between clinical and research data formats and protocols.
XNAT's existing DICOM workflow, pipeline service, and a variety of standard clinical forms (e.g. Radiological reads, NIH Stroke scale) provide support for clinical imaging research.
XNAT's overall security infrastructure is well-suited to clinical research, particularly in adhering to strict patient data privacy regulations such as HIPAA and CITI. Additionally, XNAT's built in ability to share subsets of data, and control sharing on a granular level, make it possible to quickly translate patient-specific clinical data into anonymized research data that can be grouped, analyzed, shared and published on.
Why install XNAT?
David Gutman, M.D. Ph.D., works at the Center for Comprehensive Informatics at Emory University, where he has spent time studying brain connectivity and other forays into neuroimaging-based research. As one example, exploring how the structure and anatomy of the brain might correlate to clinically diagnosed neuropsychological conditions such as depression or PTSD. Another field of research might involve isolating and comparing brain tumor tissues such as glioblastomas across multiple patients and correlating tissue composition with known clinical outcomes.
Like many others in the field, Gutman's research depends on data mining imaging studies involving large groups of people, and making probabilistic comparisons using data processing pipelines such as the LONI pipeline. Dr. Gutman's pipelines generate multiple statistical maps that need to be stored and associated with his imaging data and this functionality just doesn't exist with a traditional PACS. The alternative, storing millions of files on his local file system, was equally unpalatable.
For Gutman, XNAT provides:
- Storage & Data Support: Manage and organize his group's imaging data right alongside the custom data generated by his research
- Data Quality Control: Provide a platform to help clean up the source data, which are often inconsistently named, and cleanly separate usable from unusable image data.
- Programmatic Script Access: Create and support automated data processing workflows to assist in such tasks as tumor segmentation or analysis of anatomical features (e.g. cortical thicnkess).
Additionally, XNAT serves as a data backbone for advanced informatics exploration tools that Gutman is currently developing.
Who are your primary users?
Currently, only Gutman, his co-PI, and their group of assistants are using XNAT. However, future projects on deck include a large study being coordinated through his department that will involve multiple investigators and neuroradiologists tracking a number of subjects through multiple visits. XNAT will also prove valuable when it comes time to share and synchronize his completed data with other investigators.
What inherent features of XNAT do your users find most valuable?
Says Gutman: "A year ago, I didn't know what REST was. Now, I love REST." XNAT's REST-based data access provides incredible flexibility in working with his data, pulling from the database for processing and pushing resulting data sets back to the DB.
How have you customized XNAT to meet your needs?
Only minor customizations so far, adding custom variables to support specific data captured as part of the research process. Planning to experiment with extending the XNAT schema for subjects and tumor studies.
How could XNAT improve for the future?
Would love more scripts and more supported processes. And would appreciate more documentation [Ed: we're working on it!], providing concrete examples and best practices for using XNAT.