Abstracts

RETURN TO ABSTRACT LISTING


E1399. Open-source Informatics Platform for Successful Navigation of Large-Scale Radiology Research
Authors
  1. Ajai Nelson; University of Cincinnati
  2. Vivek Khandwala; University of Cincinnati
  3. Lily Wang; University of Cincinnati
  4. Achala Vagal; University of Cincinnati
  5. Matt Stegman; UC Health, IS&T
  6. Jacob Day; UC Health, IS&T
  7. Brady Williamson; University of Cincinnati
Objective:
In the era of big data, it is becoming increasingly important to successfully navigate large-scale, multimodal, cross-disciplinary, multi-site clinical research projects. The current solutions for imaging platforms for such projects have challenges that make them difficult to generalize to different imaging core labs. These include lack of robust data mining capabilities, difficulty of use for central readers, and expense. The aim of this project was to lay the foundation for an extensible, user-friendly open-source imaging informatics platform that significantly reduces the time and effort needed by users along with seamless data mining.

Materials and Methods:
Deidentified data from a NIH funded multicenter clinical research project were used. Open-source technologies were used to create the platform as a Web application. The back-end application programming interface uses Python with a Flask server. PostgreSQL was used for storing data and the Orthanc DICOM server for uploading and serving images. Each of these components were created as a portable Docker container. The front-end user interface uses the TypeScript programming language and the VueJS framework. The open source Open Health Imaging Foundation viewer with our own custom extensions was used. A custom JavaScript Object Notation format for creating imaging case report forms (CRF) was specified. Two central readers who have extensive expertise in imaging for large clinical trials provided feedback about the system.

Results:
A web application platform was created that includes a detailed table on the main page, to view subject images and easily filter based on variables of interest. Analytic charts can also be easily generated. Users can click on a subject ID to open their studies, along with study completion status and importantly side-by-side view of an image viewer and electronic imaging CRF for the study. -When the central reader completes the reading, the data are added to the database and appears in the table on the main screen. Notable, innovative features of our platform include dynamic integration of data with a back-end database, simultaneous viewing the eCRFs and image viewer for increased efficiency, and the reliance on open-source frameworks that allow virtually endless customizability. Both central readers agreed that the platform is more time efficient and user friendly than the existing commercial imaging platforms. There was a consensus for feasibility, and reproducibility of large-scale clinical imaging research.

Conclusion:
The platform we have built can streamline all the steps of large-scale clinical radiology research, seamlessly integrating with front-end, site-facing platforms and significantly improving the efficiency of data mining that is easily scalable. Future directions include adding machine learning-guided semi-automatic segmentation, optimizing rendering speed, handling different imaging formats, and customizing the user interface to address requests from the central readers.