Throughout the healthcare industry, CIOs are implementing technologies that allow precision diagnostics, reduce clinician workload, and automate back-office functions, from ambient documentation to machine learning-based scheduling. A lot of data is available in Penn Medicine BioBank, an institution run by the University of Pennsylvania Health System. A team led by Michael Restuccia's SVP and Chief Information Officer saw the opportunity to use this data for the benefit of patients at the research hospital.
As a physician, professor, and vice chair of radiology at the University of Pennsylvania Perelman School of Medicine, Charles Kahn says that understanding the characteristics of a population and how a particular individual differs from the rest allows the person to intervene earlier in the condition in question. This is a group of innovative healthcare companies that are pushing the envelope in the digitization of healthcare that has earned the CIO100 award over the past few years. Penn is just one example. The Stanford Medicine Children’s Health, the University of Miami Health System, as well as Atlantic Health have all begun working on precision medicine, machine learning, ambient documentation and other projects.
From a clinical point of view, Bill Fera, MD, the principal who leads Deloitte Consulting’s AI practice, says that we’re witnessing a growing number of advances in radiology, diagnostic services, and pathology. It is very noteworthy that the AI-powered CT scan analysis system is one of the first systems to be implemented in clinical practice, partly because academic medical practices that conduct research can build and operate their own tools without the burden of obtaining FDA approval, which is what healthcare product manufacturers have to deal with.
Although the system did not appear overnight, it took some time for it to come together. According to Donovan Reid, associate director of information services applications at Penn Medicine, it took at least two years for the algorithm to be ready for real-time deployment, and four years before the system finally became operational last year. "It took us hopefully two years to get it ready for actual deployment," he says.
Due to the large amount of processing resources required, the team decided to host the algorithm in the cloud.
As a result, the data was encrypted before it was sent to the cloud for processing, and the results were returned to the radiology report after the processing was completed. This was coordinated by the IT team, who developed an AI orchestrator that will be made available to other healthcare providers as a free software package. According to Penn professor Walter Witschey, the availability of this will be a great help for community service hospitals.
A couple of challenges were faced by the team before the system was up and running. There was concern among IT regarding the impact of imaging data flows on infrastructure, and the amount of computing resources needed at any given time had to be matched to the amount of imaging studies being required. Additionally, the system would have to be able to provide results as soon as possible. It has been incredibly surprising to find out that the direct cost, outside of labor, is only about $700 per month. “Doctors want interpretation right away, not at 4 a.m.,” she says.
Over 6,000 scans have already been processed through the system, and the team now plans to expand the application to accommodate more of the 1.5 million imaging scans that the hospital system performs on an annual basis.
