In the spring of 2018, Red Hat, together with the Massachusetts Open Cloud (MOC) and the Fetal-Neonatal Neuroimaging and Developmental Science Center (FNNDSC) at Boston Children’s Hospital, announced a collaboration to further develop and deploy the ChRIS Research Integration System. ChRIS was originally developed by the FNNDSC’s Advanced Computing Group to bring sophisticated (but often complex and hard-to-use) medical imaging, such as MRI and CT scans, analysis into the front lines to better inform clinical care. It is built on Red Hat OpenShift and Red Hat OpenStack. ChRIS has since evolved into a powerful general purpose, open source distributed data and computation platform. 

First, who or what is ChRIS?

“ChRIS” is the name of a platform comprising a constellation of services that manage data and containerized computations on that data. At the core of ChRIS is a Python Django-based heart (called CUBE)  that provides a REST API, manages users, tracks data, and interacts with other services. Several ancillary microservices provide overall coordination, data IO, and management of compute. ChRIS and its parts are containerized and can be deployed in many environments where containers are supported. The central premise of ChRIS is to run some compute on some data anywhere and provide a very low barrier of entry to application developers.

The functional units of ChRIS are “apps” (or plugins) that perform the actual work. And yes, there is also a prototypical ChRIS Store where developers can publish their plugins to be deployed to any ChRIS instance. We have focused heavily on reducing the barrier-to-entry for new plugin developers. In some cases just half a “screen page” of python wrapper will transform a stand alone application into a fully functioning ChRIS plugin.

While most of the current batch of apps that work with ChRIS strongly reflect its original neuro-radiological roots - neuro MRI analysis, brain image reconstruction, tractography, AI/ML and segmentation - ChRIS is actually quite agnostic and can deploy and manage anything from genomics to electronic medical record text processing to even crunching through your next shopping list (provided there’s a ChRIS app for that!). 

ChRIS plugins are standalone Linux containers, and the platform allows for the easy concatenation of these plugins into a powerful execution stack. In this manner ChRIS adheres to the UNIX philosophy of many small modules that can be easily combined and reused in many contexts. ChRIS schedules these chains of compute and their associated data on a variety of computing platforms and backends, even within the same workflow. 

Parts of the same workflow can be run on a local computer, other parts on an in-premise cluster, while other parts of the same execution can be run on a cloud provider. It is a true hybrid computation platform allowing users the freedom to isolate parts of computation and data on local resources while allowing other computation and data to execute on public clouds.

ChRIS has a maturing web-based user-friendly PatternFly interface that is intended to be used by non-technical folks, such as clinicians, radiologists and life science researchers, in addition to technical-minded users including algorithm developers. The reference UI can visualize medical images, using both novel and industry standard JavaScript libraries, and has a drag-n-drop metaphor for building out execution workflows on any data. 

ChRIS is designed to be a collaborative, easy-to-use platform for the exchange of knowledge for clinicians and researchers on a variety of medical related computation and topics, especially the construction of processing workflows.

What’s new with the work with Red Hat and others? 

Since the collaboration between Red Hat, the MOC, and the FNNDSC in 2018, ChRIS has progressed rapidly and new use cases have steadily grown. One of the most timely, and arguably important ones, has been in playing a part in helping diagnose COVID infections on lung images.

DarwinAI, a startup located in Waterloo, ON, Canada and the University of Waterloo built the COVID-Net Open Source Initiative -- a set of neural networks models and Python programs that can detect possible COVID infection on lung CT images. In its published form, COVID-Net requires fairly sophisticated skills to use -- GitHub access, Python familiarity, Jupyter notebooks, and command line chops -- skills that arguably a doctor on the front lines might not have, nor should they be expected to.

This is where Red Hat and Boston Children’s with ChRIS come into the equation. The ChRIS platform is tailor-made to help deploy such projects, and an initiative was formed to use the ChRIS platform to bring COVID-Net front and center. Work started early summer 2020 and within a few days the backend pipeline for COVID-Net was fully containerized as several ChRIS plugins -- testament to how easy it is to fold existing complex compute into a ChRIS plugin,or series of plugins.

Work progressed steadily over the summer to further develop a focused UI experience for analyzing lung CT images, and reporting COVID infection probabilities, an experience suitable for clinical users who simply want to analyze an image and get results without caring about how it all happens. 

Red Hat and the FNNDSC worked closely with DarwinAI to develop this more focused user interface, again using Patternfly to allow  for quickly designing, building, and deploying a modern and clean UI experience. The COVID-Net UI is focused on the COVID workflow and is separate but not mutually exclusive from the main ChRIS UI that is more workflow construction focused. The ChRIS COVID-Net UI has a doctor as its clear end user in mind.

It is hoped that once production ready versions of this ChRIS COVID initiative are deployed, the tool may be used to rapidly help triage large volumes of cases quickly, reliably, and easily. Fundamentally important too, is the open source and open access nature of this whole initiative. 

By being open, any clinician, scientist, researcher, or even patient, can study and examine exactly each step of the workflow and test how the computation is performed and verify model results independently. In theory, any physician anywhere could upload suitable images to a ChRIS instance, run the workflow, and get inference results -- however it is critically important to note that this workflow is intended to only help inform and supplement established clinical protocols and guidelines. It is not a diagnostic tool nor should it be used as such.

Red Hat, Boston Children’s and DarwinAI all recognize how important it is to team up and use open source technology to help the fight against the greatest pandemic of our time. The work the teams are doing together has always been incredibly important, and it is even more so now. Open technologies are crucial to better understanding and coming up with solutions for complex problems, and we have long understood that openness and collaboration can create the best solutions.

For more information on the work that DarwinAI and the University of Waterloo are leading, check out the recent press release.

For more information on how you can get involved with the ChRIS project and the project itself, see here.


About the author

Dr. Rudolph Pienaar completed a Bachelors and Masters in Electrical, Electronic, and Computer Engineering at the University of Pretoria in South Africa. He also holds a Doctorate in Biomedical Engineering from Cleveland State University/Cleveland Clinic Foundation, where he conducted research in Reinforcement Learning applied to musculo-skeletal bio-mechanical control systems. He completed postdoctoral work at the Massachusetts General Hospital, where he was an assistant in Medical Imaging. He is currently faculty in Radiology at Boston Children's Hospital and an assistant professor in Radiology at Harvard Medical School. 

Read full bio