AI Insights: AI on Trial — Startup Developing Machine Learning for Clinical Trials
By Keith Loria
Radiology Today
Vol. 22 No. 4 P. 26

One of the most notable advances for clinical trials in recent years is how AI and machine learning are enabling trial researchers to fill in the gaps when real world evidence is incomplete or inconsistent. Experts have proven that data-driven protocols and strategies powered by advanced AI algorithms have the potential to reduce trial costs by increasing patient compliance and retention, improving data quality, and identifying treatment efficacy more reliably. Flywheel, a research data platform based in Minneapolis, recently received $15 million in Series B funding for AI technology to help researchers bring drugs and medical devices to market faster. Jim Olson, CEO of Flywheel, notes that, thanks to the company’s data management platform, researchers who are working on treatments for complex diseases can now accomplish in hours and days what used to take weeks and months.

“Our platform speeds up scientific research, and that includes clinical trials,” Olson says. “That’s managing data, receiving clinical trial data, and automating processes. Clinical trials are very data dependent and, even with digital tools, can take an enormous amount of time—sometimes months or years.”

The platform can be used in the academic research space, clinics, and hospitals and assist life sciences and pharma companies in gathering important data, getting them organized, and further speeding up clinical trials. Because of COVID-19, speeding up processes is more important than ever, which has changed the game for therapeutic developments over the last year. To address this need, complex data are being collected from medical images for MR, CT, PET, X-rays, and virtually all modalities. For example, Flywheel is collaborating with the University of Wisconsin at Madison, and Richard Bruce, MD, an assistant professor of radiology and medical director of radiology informatics, on a COVID-related clinical trial.

“Dr. Bruce has a team that is collaborating with five regional hospitals,” Olson says. “They are using X-rays for COVID research instead of CT scans of lungs and applying machine learning and AI against that data to predict the presence of COVID.”

Olson calls the use of X-rays over CTs a novel approach and one that can be done better because of AI. “By using X-ray images and machine learning algorithms to identify the existence of COVID on those images and apply those data to new, untagged, and unidentified X-rays coming in, that speeds up the action,” he says. “With the machine learning capabilities, it automatically identifies the likelihood of the presence of COVID.”

Big Data
Olson notes that a great deal of health care innovation is being driven by Big Data, which is tasked with managing the data and getting value out of them. “Enhancing existing therapies in drugs is very data dependent, and machine learning and AI on that Big Data is where the innovations are coming from,” he says. “We’ve certainly seen that with COVID, but we’re also seeing it across the board. The time to market is speeding up, and it also drives drug discovery.”

For example, the University of California (UC) Irvine Yassa Lab and Flywheel are accelerating scientific collaboration around the globe by sharing data, analytic processes, and passion for understanding brain diseases. Under the direction of Michael Yassa, PhD, Flywheel secures the data and streamlines computational processing for diverse neuroimaging projects at UC Irvine. The system is also being used as a cost-effective research data and collaboration platform to enable a multicenter study on Alzheimer’s disease and Down syndrome with research partners around the world.

Another Flywheel collaboration is with Carnegie Mellon University (CMU) and the University of Pittsburgh, who together are accelerating discovery at the CMU-Pitt Brain Imaging Data Generation & Education Center by implementing Flywheel to standardize data acquisition, quality, and computation using leading neuroimaging standards, allowing users to easily share and collaborate with the international research community at large. “It’s all about gathering up meaningful data, helping to automate those, and using the analytical tools to operate on those data,” he says. “That speeds up the therapies and development of treatments for those conditions.”

Although much of the development is proprietary, Olson notes Flywheel is working with a large pharma company that is doing work in ophthalmology and neural science as well as others that are working with brain injuries and multiple sclerosis.

“The [return on investment] for our clients is in the automating of the data, which saves them time and money right out of the box, because that process no longer needs to be done manually,” Olson says. “As data are in the platform, they’re well organized and ready to support product development and research. Then we have the capabilities to support secure collaboration—so it’s HIPAA compliant—and we work within our clients’ security framework so they can work across teams and large enterprises.”

Flywheel’s recently earned funding will be used to deepen its presence in the life sciences and AI markets and accelerate product development. It’s Olson’s goal to have AI widely utilized in clinical trials over the next few years. “We want to bring this to market more quickly, and we will continue to expand the capabilities of the platform,” he says.

Looking ahead, Olson sees the market going toward federated AI, which means the movement and sharing of data given the compliance restrictions. Flywheel, he notes, can keep data resident where they are but allow a pharma company to collaborate with an institution, for example. AI allows that to happen.

“This idea of federation is what our clients are needing and what we’re focused on providing,” he says. “It’s a game changer.”

— Keith Loria is a freelance writer based in Oakton, Virginia. He is a frequent contributor to Radiology Today.