The potential uses of artificial intelligence range from diagnostic imaging to patient education, according to four healthcare technology leaders in a panel discussion at Becker's 4th Annual Health IT + Revenue Cycle Conference, Sept. 21 in Chicago. Robert Mittendorff, a partner at Norwest Venture Partners, moderated the discussion.
The four panelists included:
- Sam King, president of HFMA Southern California
- Raymond Aller, MD, department of pathology physician at Los Angeles-based Keck Hospital of USC, USC Norris Comprehensive Cancer Center and Hospital
- Michael Clark, general manager and senior vice president of provider solutions at Nuance
- Viraj Patwardhan, vice president of digital design and consumer experience, information systems and technology at Philadelphia-based Thomas Jefferson University Hospital
Here's where they think the biggest opportunities for AI in healthcare are:
Note: Responses have been edited for style, length and clarity.
Mr. Clark: "The hospitals and health systems that we work with, there's really two things they're looking to do. One, unburden. Use the technology to unburden the physician and the caregivers. What are those things in their workflow that behind the scenes AI can help [with] and deliver? Saving time, saving money, those types of things. That's definitely No. 1. The second is where you start to merge clinical and financial data together in massive amounts. How can AI help deliver meaningful information to the various different stakeholders to better inform care, which leads to better informed documentation, which leads to better quality care and ultimately appropriate reimbursement?"
Mr. Patwardhan: "In our opinion, the two areas where there's definitely been a lot of scope and a lot of need, one is definitely being predictive about certain outcomes. So, for example, if a patient is going to come to your hospital, what are certain things you can identify beforehand about this patient based on certain data because you have already learned from hundreds of previous patients that come to your hospital? That's the one area where we're really focusing. The other area that will be very helpful is delivering better patient experience and removing some of the workload that [burdens] our nurses, our physicians and staff members for a simple question like, 'What are the visiting hours?'"
Mr. King: "We buy Windex because [it cleans] the windows. Otherwise, we're not going to buy it. When we look at healthcare, same thing: What is value? That's one thing you look at as a beachhead. The other thing is [asking] what is the top priority your organization has and looking at diagnostic tools to help you: radiology, cancer diagnosis, looking at sensitivity, looking at different issues. How can you help clinicians, improve the care, improve the experience? Whatever you choose, take a look at what's the best value you can get out of that, and that's [where] AI or deep learning or machine learning technology can apply."
Dr. Aller: "We're seeing great strides in deep learning and other AI techniques being applied against images, whether they be radiology images or pathology images or a gastroenterology image. For example, the situation in pathology is it's going to be many years before a computerized microscope can actually look at a slide and make a diagnosis of cancer or not cancer or other findings. But specific questions can be addressed. What's the rate of mitoses in this section? Or is this marker of malignancy present? What proportion is it present? Those sorts of questions are being addressed now with AI-related tools and having great success. There are specific areas in medicine where we can apply these tools and have good outcome and good use."
Mr. Mittendorff: "Google's brain team, over the last couple years, did some research in dermatology. They collected, through a consortium of health systems, hundreds of thousands of dermatologic images, and then these were labeled to train a neural network to recognize the patterns of these lesions, and from those patterns recognize the diagnosis based on the label. And then they tested in a validation study that machine against [nearly two dozen] dermatologists on new images that the machine had not seen and the clinicians had not seen. … It turned out that the average of all the dermatologists was [not as] good, in terms of precision and recall, as the machine."