US drug watchdog green-lights first prostate-cancer-predicting AI software

The US Food and Drug Administration this month authorized the first AI-powered tool designed to help healthcare physicians diagnose prostate cancer. Paige, a startup spun out of the Memorial Sloan Kettering Cancer Center (MSKCC), a research institution in New York City, was granted approval to officially tout its computer- …

[NATTtrash]

What about..?

Diagnosis of cancer ultimately is left up to the medics. Paige Prostate only helps them better visualize potentially cancerous cells.

Well, that is all very good and very nice, especially since "increased efficiency" isn't fazing out the final check of flagged positive samples by a human. But...

What about the false negatives? Might be nice to hear something about that, right? I mean, I look at this from a clinical side, but I assume that you techies over here might have something to say about training this based on 30.000 images. Which to me looks like... pretty limited for a super duper automated system. And what kind of images? Like with "facial recognition", that also seems to show a "particular bias"?

I mean, image recognition in medicine isn't that new (± 20 years), nor is image classification. So it all comes down to how "smart" the AI is? Or how smart people are to put their faith in it?

[JetSetJim]

Re: What about..?

> training this based on 30.000 images

Depends on the image, the article mentions zooming & annotating - if each image has many thousand cells on it, some of which are/may be cancerous, then that would help, but even then the number of labelled positives to train on might be a bit light for my liking.

Saying that, how many slides are used to train a medic? IANAMedic, so no idea. How many different features can a cancerous cell/cell group have? How big a grouping of cells do you need to look at to say "here be cancer"?

[Kev99]

From experience, the best first test is a female doctor with long, skinny fingers. The next best is a physical examination of the prostate. Unfortunately, this can only be done after a radical prostatectomy. In the former case, two doctors said if it wasn't for her physical exam the cancer would not have been detected. In the case of the latter, when the prostate was opened to do the biopsy, the surgeon said my prostate was about to burst.

Anything that can help give an earlier detection is worth every cent.

[Snake]

Cancer

I'm glad to hear that you are doing well!

Prostate cancer has a good successful treatment rate: 98% in the United States.

Pancreatic cancer, however...5%.

We need more focus on the truly deadly ones.

[Sam Therapy]

I was on regular monitoring because I'd had an enlarged prostate for many years, and according to the docs, at greater risk. So, regular blood tests for PSA level, the odd finger up the jacksie, and eventually biopsies, scans and whatnot.

Found the cancer very early on and was told that, due to the very small incidence of cancerous cells, they'd prefer to wait and see for a while. Fair enough, so I waited but, due to the pandemic, everything was delayed for over a year, except PSA tests. PSA kept climbing and after waiting and waiting, so was I, as in climbing the walls. After more tests and scans, I'm pleased to say the cancer didn't increase hugely and it's still highly localized.

Anyhow, time has come to get the thing out, so I'm waiting to hear from the team about my treatment.

In my case, I don't think the new system would have made any difference, since I was already on the watch list, as it were. That said, anything that makes early discovery easier has to be a good thing.