Can Artificial Intelligence Help Diagnose Dementia?

This transcript has been edited for clarity.

Kathrin LaFaver, MD: Hello, and welcome to the session today. I am Dr LaFaver. I'm a movement specialist in Saratoga Springs, New York. I have the pleasure to talk with Dr David Jones today. Dr Jones is a consultant at the Mayo Clinic and the Department of Neurology, and is the director of artificial intelligence (AI) in the neurology department at Mayo Clinic.

We will be talking today because David was part of a session of hot topics in neurology at the recent annual meeting of the American Academy of Neurology in Boston, speaking on the topic of AI. Welcome, David.

David T. Jones, MD: Thanks for having me.

LaFaver: I think anyone who attended that meeting in Boston was taking away that AI is indeed a hot topic. If you're talking about all the tools, there were five or six sessions dedicated specifically to the use of AI tools, and there were over 30 abstracts on that topic.

To ease us into it a little bit, why don't you tell us what got you interested and involved with AI as it pertains to the practice of neurology?

ChatGPT Is Not Intelligent

Jones: I think that mainly came out of my interest and background in quantitative neuroimaging in the field of behavioral neurology. One of the things that AI is very good at is pattern recognition and turning unstructured data into structured data. That's where quantitative neuroimaging has a lot in parallel with those aspects of AI.

I've been part of the Alzheimer's Disease Neuroimaging Initiative and other large projects, turning brain images into numbers for my whole career. When the Department of Neurology responded to an institutional effort to lead the transformation of healthcare using AI, I was happy to be asked to step up to lead the Department of Neurology at Mayo Clinic's efforts in AI to transform the practice of neurology.

LaFaver: That's wonderful. You make it sound like a natural next step in what you were already doing, specifically with neuroimaging.

Before we dig a little deeper, the question always is, what is actually the definition of "artificial intelligence"? I wrote down what was mentioned in the Hot Topic sessions and I wonder if you agree with that.

This was the definition taken from an AMA statement in 2018: Artificial intelligence encompasses a host of computational methods that produce systems that perform tasks normally requiring human intelligence. What do you say to that?

Jones: That's a pretty common definition. I think the problem with most of these definitions is that they kick the can down the road in terms of defining what intelligence is by just saying that it's human-like. Whatever humans do, those are the types of tasks we're talking about. I don't really like that definition of intelligence because there are obviously intelligent things that humans do not do.

There are certainly other definitions. We actually put forward our own definition that we use in the Department of Neurology here, at Mayo Clinic, in an article last year in the green journal. This definition takes into account exactly what we mean by intelligence, first. How we define that in the article is the collection and processing of data and information about environments that an agent can use to model, to make, to do beneficial things within that environment or related environments. That's the definition of intelligence that we use.

Why do I think that's important? It calls out the different elements of intelligence, breaks it out into a tripartite definition of the process, the product in an agent. Intelligence actually requires all those things.

That helps us answer questions about large language models like ChatGPT. Is it intelligent? It doesn't meet our definition of intelligence, but it meets the definition of a tool that can be involved in an intelligent process. It can interact with an agent, a human who has goals, and be part of their process of modeling the environment. We're able to do intelligent things using the assistance of ChatGPT, but ChatGPT is not intelligent.

Diagnosing Dementia With AI

LaFaver: I think those are important distinctions. It goes back to a fear that some people might have that we're going to be taken over by robots.

Let's talk about what you covered in your talk, which is really the pipeline that you are developing at Mayo Clinic to specifically help with diagnosing and managing dementia and behavioral disorders. Why don't you give a bit of an overview of that pipeline?

Jones: We are developing what we call the neurology technology development ecosystem, which tries to wrangle data in this way to digitize it, model it, and organize it so that we can use things like machine learning and AI algorithms to model and extract the most relevant information from the data we're getting into clinics, and then bring that to the point of care to help clinicians make medical decisions.

The thing that's furthest along within this technology development pipeline is FDG-PET scans of the brain, which are extremely helpful in evaluating people with cognitive problems, but they take a lot of expertise to interpret. That expertise, I think, is quite limited.

What we have been trying to do is put algorithms around the reading of the data and do it in a way that's very natural for physicians, and in a way that I currently read scans, which is pretty simple. If somebody asked me to look at a brain PET scan and use my expertise to interpret it to help them with medical decision-making, what I'm typically doing in that setting is thinking about all the other cases that I've seen that look like that. I'll usually ask questions about the context of the particular patient and see if it fits into those other cases I've seen, and then give some advice based on my experience with scans that have looked like that in the past in those patients' context.

That's limited to the fact that I have seen a bunch of scans and have experience with them. Many people may not, and they may not have experienced those scans in this particular context. AI is really good at solving all those problems. It can take patterns and images and put those in numbers in a way that we can then match to other cases that look like that, summarize what happened to those cases, and present that to the clinician and the patients in a way that they can then make a highly informed decision about what the data mean for their patient.

LaFaver: You're basically having access to all the rich databases that have been collected in studies of aging and so on over many years. You mentioned during your talk that you might give clinicians in the future the opportunity to do virtual consults with that database. Can you expand a bit on that?

Jones: I think about how we triage many cases for referral throughout the world here at Mayo Clinic. We can't accept every case, but we try to give the best advice we can when possible. If we could give people a way to access the knowledge and experience that we have with these rare or complex conditions, we think you'd be able to do virtual versions of those types of consultations or accessing this deep knowledge with a variety of rare or complex neurologic conditions, which should help people anywhere access that level of knowledge and expertise. I think this technology will really facilitate that type of thing.

The Next 5 Years

LaFaver: That sounds really helpful. It's difficult for many people to travel for a second or third opinion, so accessing it in different ways is going to be really helpful. Where do you see this work going? What are your plans over the next 5 years or so?

Jones: Our plans are to make sure that we are collecting, processing, and modeling information about neurologic status in a way that we can link all aspects of the neurologic exam and evaluation to these types of technologies, so that we can really scale what we build inside of Mayo Clinic outside of our walls to benefit as many people as possible.

LaFaver: Wonderful! I think this is really important work and really fascinating to move away from having the expertise of a single neurologist that you happen to see at that particular time and circumstance to accessing much richer datasets, as you're describing. I think that was very insightful.

Do you have any parting words or takeaway messages for neurologists listening?

Jones: One of the things I try to emphasize is that we all fear change, and there's a large amount of anxiety, I think, around the fast pace of change that's going on. I think the appropriate response to that is to actually have neurologists step up and lead this transformation, or the change that's occurring, so that we can do it shaped by our traditional values, the needs of the patient and rigorous research, and the benefits we want to see for healthcare.

We can make sure that the technology is doing those things, whereas if we let these things happen and change without our input and involvement, it may not do the things that we want them to be doing. I certainly encourage everyone in healthcare and in neurology, in particular, to step up and lead this transformation, and not let it be led by a sector outside of healthcare.

LaFaver: That's a very wise perspective, and you're certainly a leader in this emerging field. Thank you so much, David, for talking to me. This is Kathrin LaFaver on behalf of Medscape. Have a good rest of your day.

Jones: Thank you very much.

Follow David Jones on Twitter @DavidJonesBrain

Follow Kathrin LaFaver on Twitter @LaFaverMD

Follow Medscape on Facebook, Twitter, Instagram, and YouTube

Comments

Commenting is limited to medical professionals. To comment please Log-in.

Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.