Although some key challenges remain, cloud computing can facilitate new clinical and research applications in MRI, according to a May 16 presentation at the annual meeting of the International Society for Magnetic Resonance in Medicine (ISMRM).
In MRI, for example, cloud computing can be utilized to support applications such as real-time image reconstruction, automated quantitative perfusion mapping in MRI cardiovascular stress testing, and training of artificial intelligence (AI) algorithms, according to presenter Michael Hansen, PhD, of Microsoft.
"You can do a lot if you free yourself up from computational restraints," Hansen said.
Cloud computing has many definitions and can be utilized on-premises or via a public cloud, according to Hansen. Service types including infrastructure-as-a-service, platform-as-a-service, and software-as-a-service.
Transmitting images via a gateway from an MRI scanner to a cloud environment -- either a local cloud or a public cloud that can scale infinitely -- can open up new opportunities in MRI, Hansen said.
"That means we can now think about bringing applications to the clinic that would otherwise be hard to work [with]," he said. "This is not science fiction at all."
For example, MR image reconstruction can be performed using cloud computing technology either in real-time or offline, he said. One approach, the Gadgetron open-source framework co-developed by Hansen, provides for interactive, real-time image reconstruction using the public cloud.
Another option is the Yarra open-source toolbox for clinical-translational MRI research developed by the New York University School of Medicine. Like Gadgetron, Yarra is a reconstruction framework that can be connected to an MR scanner and work with data in a remote environment, Hansen said. However, Yarra is more of an offline tool than Gadgetron, Hansen said.
"Depending on what your specific application is and what your workflow is, this may be the right model for you," he said. "It has a number of specific advantages."
What's more, cloud resources can also be used to accelerate AI training platforms, Hansen said.
However, one of the big challenges to the use of cloud computing in MRI is the proprietary data formats used by different vendors. The ability to effectively share data and have common platforms would facilitate progress, Hansen believes. As a result, he encourages participation in the ongoing ISMRM Raw Data (ISMRMRD) project, which is developing a vendor-neutral, raw MRI data format standard in hopes of enabling generic image reconstruction.
Bandwidth is also a concern when considering implementation of cloud computing technology.
"If you have a gigabit connection from your institutions you can use cloud environments if it's done properly, but this is also something that we need to work on," Hansen said. "Compression is potentially an option; there are some papers in that area."
What's more, security is an important topic. Better understanding is needed for when raw data is identifiable, Hansen said. In addition, well-defined standards for security controls are necessary.
"And that is critical to change the culture in a lot of hospitals, where IT organizations are rightly concerned [about security in cloud computing]," Hansen said. "So the more best practices and patterns we develop, the better off we are in terms of making this actually happen."
