Charting a Course for AI-based Solution Adoption – 2022 SIIM Q&A

During the 2022 Society for Imaging Informatics in Medicine (SIIM) meeting, Blackford’s Founder and CEO Ben Panter, PhD participated in the Ask Industry Pa

post image
Blackford Analysis

During the 2022 Society for Imaging Informatics in Medicine (SIIM) meeting, Blackford’s Founder and CEO Ben Panter, PhD participated in the Ask Industry Panel Discussion, moderated by Dr Katherine P. Andriole, PhD, Director of Imaging Informatics at Brigham and Women’s Hospital.

The panel was asked 8 key AI adoption questions - we’ve condensed Ben’s insight and answers to these questions below to help provide you with an insightful overview of Ben's perspectives:    Q: We have stand-alone AI-based applications, but what are some of the ways solutions can be integrated, and into which systems? A: It is very rare that we see a single point of integration with AI applications because they need to be integrated throughout the workflow. Most facilities have different providers for worklist, viewing, and reporting, but AI is only useful if there’s seamless integration with all those different vendor’s products. PACS is certainly a key element of AI integration, but it’s also critical to integrate beyond the walls of the radiology department so we can demonstrate the downstream clinical value and upstream quality improvement that unlocks the true value of AI.   Q: What are the pluses and minuses of different approaches to integration? A: Procurement through a single channel certainly makes the paperwork easier, but you still need to have a full range of solutions available to ensure you have the right solutions to meet your needs. Various vendors claim best-in-class applications, but our experience with deploying more 3 rd party AI solutions than anyone else, is that different facilities have different clinical and business drivers, different patient demographics and different radiologists and referrers. So, a best-in-class solution that may work for one institution can completely miss the mark for another.   Q: What is the role of the cloud in clinical implementation? What are the concerns? A: The cloud certainly has a role in the clinical implementation of AI, but how much of a role depends on the preferences of the institution. Flexibility is the key. Today, the vast majority of infrastructure is on-premises. But soon, we’ll see a lot of infrastructure transferring to the cloud. When it’s time to transfer from on-premises to the cloud, the ease of facilitating the transfer should be a concern. That’s why it’s important to find a partner who has demonstrated delivery of both on-premises and cloud solutions, with the ability to enable a smooth shift between them without any forklift upgrades.   Q: Who pays for AI? How do we get sustainable ROI from AI in healthcare? A: There are four routes for AI to be financially successful and sustainable.
  1. One route is for AI that has a strong enough value proposition that it can deliver value without any external reimbursement. For example, stroke care coordination, modality throughput enhancement, or image registration.
  2. Another route is for AI that becomes the standard of care. I’m not sure if we should refer to mammography Computer Aided Detection as AI, but the use of CAD is included in some screening reimbursement codes and this is a potential route for some AI applications.
  3. A third route is for AI that provides real clinical value to patients but doesn’t fit in the current paradigm. Examples of this are Heartflow and Perspectum. This route requires reimbursement, but that takes time.
  4. The fourth route is where visionary providers and clinicians adopt AI and show the value. Low Dose CT lung cancer screening programs are an example. This route requires linkage with downstream disciplines – in this case pulmonology and oncology. It takes effort, but it can happen.
  Q: What integration standards do you see as important for clinical implementation? A: As a platform provider that integrates across and between vendors, we’re very keen on standards. They make our life easier. But to be useful, they have to be adopted. That’s not in the hands of vendors like Blackford, that’s in the hands of the large OEMs. The truth is, PACS implementation of even modest enhancements to the DICOM standard often lags years, even decades behind the standard. My frustration is that I’ve spent my entire career finding ways to work around these inadequacies. As an industry, I think we have to focus on adoption of the standards that are already out there before we create new ones.   Q: When and how should organizations adopt AI clinically? A: We had an interesting experience with an intracerebral brain hemorrhage (ICH) product last year. We deployed the application within two different organizations, and they had polar opposite experiences. One loved it, the other wanted it removed immediately. Up until that point, I’d been laboring under the assumption that we would find a ‘best-in-class’ application, but that turns out to be absolutely wrong. The patient demographics, modalities, and business environment differences between these two organizations meant that they needed very different solutions. And crucially, this can be difficult to tell in advance. That’s why we now recommend that organizations trial solutions and measure the potential benefit prior to implementation. To do this efficiently requires a platform, and often, a team to help deliver that trial. If you go direct with a single vendor, you have all your eggs in one basket. So if the VC funding runs out for that vendor, you’re in trouble. If, however, you’ve set things up through a platform, you can manage change gracefully. For example, when MaxQ decided to end their product recently, we were able to transfer all our sites that were using that product to an alternative product with zero disruption and zero additional cost to those sites. The AI market is in its infancy, and nobody knows which solutions are going to be best in the future. Organizations need to find a way to insulate from that reality and a platform approach is the way to address that need.   Q: How do you envision the impact of AI on resident training and programs and other personnel in radiology? A: We’re already hearing from resident training programs that applicants want to know what AI is in use and how they’ll gain experience during their program. So, I think it’s going to be essential that AI in some form is in use to attract candidates. From the perspective of impact on other personnel in radiology, there is lots of potential: automatic quality improvement to assist technologists in patient positioning, intelligent automation in scheduling to optimise patient attendance and streamline modality configuration, etc.   Q: What will the AI diagnostic or AI therapeutic device company of the future look like? A: Currently, the focus is on radiology benefits. But the real potential is downstream of radiology. We need to also focus on what benefits AI can bring to the activities of the referrer, and how it can influence the healthcare system’s economics.