Following on the heels of a successful pilot that brought artificial intelligence to bear on radiology diagnoses, the Department of Veterans Affairs is looking for new opportunities to enhance its health care with AI. The VA’s National Artificial Intelligence Institute (NAII) will soon be announcing the launch of a new AI Tech Sprint focused on cardiothoracic medicine.
“This is particularly important for veterans’ health issues from lung cancer screening, which is the most common cancer to kill a veteran, and heart conditions, one of the most common causes of death. The top two, cancer and heart disease, so it has significant implications for health of veterans,” Michael Kim, NAII’s chief of staff, told Federal News Network.
The AI Tech Sprints are competitions involving industry and academia, designed to help federal program office leaders see which business product will be most beneficial, without having different medical centers across the VA independently running their own assessments and potentially choosing different products. That can lead to differing standards of care, and a lack of IT standardization across Veterans Integrated Service Networks (VISNs), VA’s 18 regional care systems.
The AI Tech sprints allow a panel of program office leaders to evaluate different AI tools on a variety of criteria, including the performance of the AI model itself, the potential value it would bring to VA medical centers, how close the tool is to being ready to launch, and whether the vendor already has approvals to work within VA. This competition then allows the panel to make recommendations at a regional or national level to promote standardization in the care offered.
Kim said NAII is currently wrapping up an AI Tech Sprint on data science, with a ceremony planned for Feb. 22. The next AI Tech Sprint is anticipated to be announced sometime in the near future. It will provide vendors with expertise in radiology AI with VA data in a secure fashion to test the efficacy of their models.
Radiology pilot and other programs
But that’s not the only AI-related development coming from the VA. Amanda Lienau, director of Data and Analytics Innovation at the VA Office of Healthcare Innovation and Learning, said her office is currently working on a full business case analysis of the successful radiology pilot that recently wrapped up within the radiology department of the VA St. Louis Health Care System. The pilot used an AI model to aid radiologists in prioritizing scans so that the most serious cases would be read first.
“As that formal business case analysis is complete, can then present that to medical centers, many of whom are interested in locally, then, adopting this technology based on the very interesting results of the initial pilot, and now that the requirements for information security and privacy have been completed for national use,” Lienau said. “But it will be up to individual medical centers to adopt the technology.”
That business case analysis will include information about a variety of different factors that medical centers can use to make that decision. For example, what other competitor products are out there besides the AI model used in the pilot? What is the value add in terms of build time? What are the technical requirements, including information security, privacy and other standards? What difference does it make in clinical terms, both for the patient and the radiologist?
Lienau said radiologists involved in the pilot universally approved of the AI tool, finding it user-friendly and valuable. In fact, Lienau said, there was one specific instance in which using the AI tool directly averted a potentially serious, adverse health outcome for a patient by prioritizing their scan, allowing radiologists to more quickly report that issue to the clinical care team.
Patrick C. Malloy, the executive director for the Veterans Health Administration’s National Radiology Program, told Federal News Network that other AI pilots are currently in the works as well. One will focus on improving oversight and quality assurance on remote mammography screenings by providing real time feedback. That program is expected to go live sometime in May or June.
The other will focus on lung cancer screenings. In that program, the AI model will assist radiologists in identifying lung nodules by comparing them to prior examinations and making the lung nodules look more conspicuous for easier identification.
“We currently have close to 100 facilities who are now certified to do lung cancer screening and are looking to expand this to all of our facilities across the country. This will be a significant aid to the radiologists and in scaling up those efforts to meet the needs of the program,” Malloy said.
Building the infrastructure
Lienau said the pilot required the assistance of local IT and biomedical engineering experts to help set up local servers and cloud resources for testing, development and validation of the AI utilities and tools. This all happened in collaboration with the VA’s Office of Information and Technology to help meet the field needs. Lienau said the Office of Healthcare Innovation and Learning is currently working on user guides, self-service tools and registries to help make VA users aware of these resources.
Malloy also said there are efforts underway to improve the connectivity of imaging databases. All 18 VISNs currently have their own picture archiving and communication (PAC) system, all of which function similarly, but have slight disparities, often due to using different vendors. That creates challenges querying across the various systems, all of which archive into the Veterans Health Information Systems and Technology Architecture (VistA).
“We have up until the recent past did not have very good connectivity between our individual VISN databases through our our individual PAC systems,” Malloy said. “And we’ve most recently developed the technology to connect up through our VistA imaging and our PAC system, the ability to query and retrieve images from multiple different VISN sources so that we’ll have that national connectivity of all of our imaging databases. This we’ve enacted in the current [electronic health record], VistA CPRS. And then we’ll be continuing the effort as we move to Cerner EHR.”
Daisy Thornton is Federal News Network’s digital managing editor. In addition to her editing responsibilities, she covers federal management, workforce and technology issues. She is also the commentary editor; email her your letters to the editor and pitches for contributed bylines.