Why AI-based Computational Pathology Detects More Cancers

Chances are you or someone you love has had a biopsy to check for cancer. Doctors got a tissue sample and they sent it into a pathology lab, and at some point you got a result back. If you were lucky, it was negative and there was no cancer. But have you ever wondered exactly what happens in between those steps? Until recently, it’s been a meticulous but imperfect manual process where a pathologist would put a thin slice of tissue under a high-powered microscope and examine the cells by eye, looking for patterns that indicate malignancy. But now the process is going digital—and growing more accurate.

Harry's guest this week is Leo Grady, CEO of, Paige AI, which makes an AI-driven test called Paige Prostate. Grady says that in a clinical study, pathologists who had help from the Paige system accurately diagnosed prostate cancer almost 97 percent of the time, up from 90 percent without the tool. That translates into a 70 percent reduction in false negatives—nice odds if your own health is on the line. This week on the show, Grady explains explain how the Paige test works, how the company trained its software to be more accurate than a human pathologist, how it won FDA approval for the test, and what it could all mean for the future of cancer diagnosis and treatment.

Please rate and review The Harry Glorikian Show on Apple Podcasts! Here's how to do that from an iPhone, iPad, or iPod touch:

1. Open the Podcasts app on your iPhone, iPad, or Mac. 

2. Navigate to The Harry Glorikian Show podcast. You can find it by searching for it or selecting it from your library. Just note that you'll have to go to the series page which shows all the episodes, not just the page for a single episode.

3. Scroll down to find the subhead titled "Ratings & Reviews."

4. Under one of the highlighted reviews, select "Write a Review."

5. Next, select a star rating at the top — you have the option of choosing between one and five stars. 

6. Using the text box at the top, write a title for your review. Then, in the lower text box, write your review. Your review can be up to 300 words long.

7. Once you've finished, select "Send" or "Save" in the top-right corner. 

8. If you've never left a podcast review before, enter a nickname. Your nickname will be displayed next to any reviews you leave from here on out. 

9. After selecting a nickname, tap OK. Your review may not be immediately visible.

That's it! Thanks so much.

Full Transcript

Harry Glorikian: Hello. I’m Harry Glorikian. Welcome to The Harry Glorikian Show, the interview podcast that explores how technology is changing everything we know about healthcare.

Artificial intelligence. Big data. Predictive analytics. In fields like these, breakthroughs are happening way faster than most people realize. 

If you want to be proactive about your own health and the health of your loved ones, you’ll need to learn everything you can about how medicine is changing and how you can take advantage of all the new options.

Explaining this approaching world is the mission of my new book, The Future You. And it’s also our theme here on the show, where we bring you conversations with the innovators, caregivers, and patient advocates who are transforming the healthcare system and working to push it in positive directions.

Chances are you or someone you love has had a biopsy to check for cancer. 

Doctors got a tissue sample and they sent it into a pathology lab, and at some point you got a result back. If you were lucky it was negative and there was no cancer.

But have you ever wondered exactly what happens in between those steps?

Well, until recently, it’s been an extremely meticulous manual process. 

A pathologist would create a very thin slice of your tissue, put it under a high-powered microscope, and examine the cells by eye, looking for patterns that indicate malignancy. 

But recently the process has started to go digital. 

For one thing, the technology to make a digital scan of a pathology slide has been getting cheaper. That’s a no-brainer, since it makes it way easier for a pathologist to share an image if they want a second opinion.

But once the data is available digitally, it opens up a bunch of additional possibilities. 

Including letting computers try their hand at pathology. 

That’s what’s happening at a company called Paige AI, which makes a newly FDA-approved test for prostate cancer called Paige Prostate.

The test uses computer vision and machine learning to find spots on prostate biopsy slides that look suspicious, so a human pathologist can take a closer look.

So why should you care?

Well, in a clinical study that Paige submitted to the FDA, pathologists who had help from the Paige system accurately diagnosed cancer almost 97 percent of the time, up from 90 percent without the tool.

That translates into a 70 percent reduction in false negatives. At the same time there was a 24 percent reduction in false positives. 

I gotta tell you, if I were getting a prostate biopsy, I’d really like those improved odds. 

And it’s a great example of the kinds of AI-driven medical technologies that I write about in The Future You, which is now available from Amazon in Kindle ebook format.

So I asked Paige’s CEO, Leo Grady, to come on the show to explain how the test works, how Paige trained its software to be more accurate than a human pathologist, how the company got the FDA to give its first ever approval for an AI-based pathology product, and what it could all mean for the future of cancer diagnosis and treatment.

Here’s our conversation.

Harry Glorikian: Leo, welcome to the show.

Leo Grady: Hi, Harry. Glad to be here.

Harry Glorikian: Yeah. You know, I've been watching the company for some time now, and the big story here seems to be that we're really entering the area of digital pathology, also known as sort of computational pathology, and it's funny because I've been talking about digital pathology since I think I started my career back when I was 25, which seems like a long time ago at this point. But for a lot of laboratory tests that we use, like it's usually done by eye, and now we can get a lot from sort of AI being assistive in this way. So keeping in mind that some of the listeners are professionals, but we have a bunch of sort of non-experts, could you start off explaining the term maybe computational pathology and summarize where the state of the art is, which I assume you guys are right at the cutting edge of it?

Leo Grady: Yeah, so I think it actually might help just to jump back a level and talk about what is pathology and how is it done today? So today, so pathology is the branch of medicine where a doctor is taking tissue out of a patient through a biopsy, through surgery and making glass slides out of that tissue, looking at it under a microscope in order to make a diagnosis. And today, all of that process of taking the tissue out, cutting it, staining it, mounting it on slides. Then gets looked at under a microscope by a pathologist to make a diagnosis, and that diagnosis the pathologist makes is the definitive diagnosis that then drives all of the rest of the downstream management and care of that patient. When pathologists are looking through a microscope, sometimes they see something that they're not quite sure what it is. And so they may want to do another test. They may want to do another stain. They may want to cut more out of the tissue, make a second slide. Sometimes they want to ask a colleague for their opinion, or if they really feel like they need an expert opinion, they may want to send that case out for a consultation, in which case the glass slides or are put in a, you know, FedEx and basically shipped out to another lab somewhere. All of those different scenarios can be improved with digital pathology and particularly computational pathology and the sort of technology that we build at Paige. So in a digital world, what happens instead is that the slides don't go to the pathologist as glass. They go into a digital slide scanner, and those slide scanners produce a very high resolution picture of these slides.

Leo Grady: So these are quarter-micron resolution images that get produced of each slide. And then the pathologist has a work list on their monitor. They look through those those cases, they open them up and then that digital workflow, they can see the sides digitally. When they have those slides digitally, if they want to send them out to a second opinion or or show them to a colleague, it's much easier to then send those cases electronically than it is to actually ship the glass from one location to another. Once those slides are digital it, it opens up a whole other set of possibilities for how information can come to the pathologist. So if they want additional information about something they see in those slides, rather than doing another stain, doing another cut, sending for a second opinion, what we can do and what we do at Paige is we we identify all the tissue patterns in that piece of tissue, match those against a large database where we have known diagnoses and say, OK, this case, this pattern here has a high match toward to something that's in this database. And by providing that information to the pathologists on request that pathologists can then leverage that information, integrate