Click Here for podcast episode details and listening links.

[Colin Miller] 

Hello, you're tuned in to SNiPS, a reoccurring special segment from our ongoing series, Fractals: Life Science Conversations. Bracken is the professional services firm for life sciences and digital health organizations. Our intelligence ecosystem fulfils consulting, regulatory, marketing and analytics, an integrated and strategic approach. 

As you know, AI has become this buzzword and jargon word that everyone's using at the moment. So how do you think AI and machine learning will intersect with molecular imaging in the coming years? What's your opinion of this in the future?

[Ben Larimer] 

So, there's the low-hanging fruit, and that is the analysis of images that come out of every CT, MRI, PET scan, these giant collections of pixels. And of course, radiologists are experts at being able to interpret these. But I think what we've seen with, and I'm no computer scientist, but what we've seen with AI and machine learning is the ability to digest these pixels and interpret them. 

And so, I think whether that's assistance, real-time assistance for radiologists or charts or picking up on things that are patterns we don't even see, and also just making radiologists more efficient. I'm not in the camp that I think AI is going to eliminate radiology, but I do think it'll make them more efficient, which is good for them and is good for the entire field. I think that a PET diagnostic is a drug with radioactivity attached to it. 

One of the limitations of our field right now is that it's very hard to make these drugs, and it takes a long time. And the ones we have were developed over decades. AI has the potential to take that timeframe of development from decades to months. 

So, I think that AI will speed up and revolutionize the breadth and quality of the diagnostics that we have and allow us to be able to really tailor imaging agents for whatever question that we may have. 

[Colin Miller] 

That's an interesting one on the actual drug development side itself. I was going to take you down into a bit more of an imaging one, but it's fascinating to think of how we can shorten timeframes just by putting AI in the middle of a molecular development program. 

[Ben Larimer] 

Yeah, yeah. I mean, it's still very, very early stages. The technology is already at a pretty advanced stage. 

You look at what DeepMind is doing out of Google, David Baker, the University of Washington. Some of these AI-designed drugs are already in clinical trials. There was a COVID vaccine that was approved that was completely AI-designed. 

And AI is, you know, they're doing it for antibodies. Even the recent FDA announcement that they're going to move from needing antibodies to be tested in mice to potentially doing it through AI, not just touching the actual design of the drug itself, but the process of moving it into patients safely. I think actually AI will improve medicine, will improve accessibility, will improve the cost and add value. 

[Colin Miller] 

I'm optimistic with you as well. Just to sort of go into the PET part of that, the PET imaging part of that. I wonder what your thoughts are. 

If I go back a little bit, when Radiomics first came out, I suddenly realized that we had the ability to start really characterizing lesions. And then, you know, go back 10 years, that's when it sort of, or probably at least that. And I actually wondered whether you could create databases at that time that were of all the lesions and you could basically ask machine learning to go and look for that characterization in future patients and say, well, this patient will respond to X, Y, and Z. 

I think we've got way more sophisticated than that. And now with the PET imaging probes that we have overlaying and knowing that there's such rich data in a PET scan and a CT scan, I mean, what's your thoughts on that and being able to do much better diagnosis and prognosis? 

[Ben Larimer] 

So, cancer metabolism is measured with a PET probe colloquially called FDG, which is a radioactive sugar analog that gets trapped inside of cancer. Cancer needs sugar because it's growing rapidly. And so, you know, it's the workhorse for a large number of cancers in terms of diagnosis, staging, and even treatment monitoring. 

And so, I think there's something like 2 million FDG PET scans done each year in the United States. So, it's on that order of magnitude, right? There's a huge wealth of cancers that have been imaged, you know, that we've looked at the sugar uptake. 

What we're clinically using it for is to say, okay, there is a tumor here or something, you know, that we suspect there's a tumor here, you know, or the tumor that was here has shrunk after we treated it. But there are other components. One of the beauties of PET is that it's quantitative. 

So not only are we saying, okay, there is high sugar uptake here, but the number is 11. But we can get even more granular than that and we can break the tumor down into a bunch of pixels. And we can say, actually, the tumor has, on average, the signal is 11, but in the upper right quadrant, the signal is 27. 

And in the lower right, lower left quadrant, it's four. There could be aspects of that, that radiologist doesn't have time or the context to be able to expand upon that. But a machine learning algorithm fed 2 million of these may identify certain patterns that are consistent with any number of clinical features, prognoses, and things like that. 

So I think people have been looking at that in terms of FDG for a while. And, you know, I think that some interesting trends have come out of that. I'm the most excited for us to move beyond just sugar metabolism, amino acid metabolism, receptor expression, immune function. 

I think the beauty of AI machine learning is that the more data we feed into it, the better. I'm not an expert by any means in FDG radiomics or CT radiomics. I think we could all agree that feeding more data into it, so if we have, you know, the CT, you know, the shape and the size of the tumor, plus it's how much sugar it's eating, plus how much granzyme is inside of the tumor, you know, the more we can feed into these models, I think the better predictions or the newer information we'll have out of it. 

[Colin Miller] 

Fractals: SNiPs is brought to you by Bracken and available wherever you get your podcasts. Visit us at thebrackengroup.com or reach out directly on LinkedIn. We'll be delighted to speak with you. 

I'm Colin Miller, wishing you sound business and good health. Thanks for listening.