Kim, Michael E.; Kanakaraj, Praitayini; Liu, Yihao; Schwartz, Trent; Li, Zhiyuan; Newlin, Nancy R.; Gao, Chenyu; Schilling, Kurt G.; Bao, Shunxing; Rogers, Baxter P.; Landman, Bennett A.; Ramadass, Karthik. (2026).听.听Brain Imaging and Behavior, 20(5), Article 123.听
Reproducibility鈥攖he ability of other researchers to repeat a study and obtain consistent results鈥攊s essential for confirming scientific findings, advancing knowledge, and maintaining trust in research. In neuroimaging, researchers commonly use containerization, a method that packages software code together with the tools and dependencies needed to run it, to make analyses easier to reproduce. However, there is no universal standard for sharing these containers, and the wide variety of research practices makes it difficult to enforce a single approach. We propose that effective container sharing should meet three key requirements: containers should be publicly available, include clear and self-contained documentation, and provide procedures for checking that the container and its expected outputs remain consistent. We demonstrate this approach by publicly releasing three separate containers through Zenodo, an online repository for research materials. Our method meets these requirements while requiring little additional effort from researchers. Because several existing tools can also be used to distribute and maintain containers, we additionally discuss alternative approaches and resources available for container sharing.
Fig.听1
Containerization is a practice meant to promote reproducible and replicable science; however, simply giving another researcher a container cannot ensure this, nor do any standards exist for sharing containers. There are three considerations that must be taken into account for any container release: 1.) The container should be publicly accessible on an archival service that can outlast the utility of the container itself. 2.) Sufficient documentation should be included along with the container that can allow users to understand how to properly run the container, such as the command(s) and structure/format of inputs. 3.) There must be some aspect of consistency or verification so that the container algorithm will result in reproducible outputs