Prostate algorithm may be ready for routine clinical practice

Researchers have demonstrated the use of an AI tool that can accurately identify or rule out prostate cancer on digitized pathology slides from core needle biopsies.

And they’ve shown the model’s promising utility in routine clinical practice.

The work was led by investigators at UPMC in Pittsburgh and is running in The Lancet Digital Health.

Testing an AI solution developed by Ibex Medical Analytics, which sponsored the study, pathologists Liron Pantanowitz, MD, Rajiv Dhir, MD, and colleagues found the tool achieved 98% sensitivity and 97% specificity at detecting prostate cancer.

That’s significantly better performance than previous attempts have brought back, the authors note.

The AI had been trained on more than 1 million portions of stained tissue slides as labeled by senior-level pathologists.

It was tested on its ability to tell normal from abnormal tissue on a separate set of 1,600 slides prepared from biopsies of 100 consecutive patients with possible prostate cancer.

In routine practice, the algorithm was used to assess more than 11,000 slides from more than 900 patient cases. The AI caught at least one cancer that had been missed by the experts.

In a news release sent by UPMC, Dhir says algorithms like this can be particularly helpful when it comes to dealing with lesions that have unusual characteristics.

“A nonspecialized person may not be able to make the correct assessment,” he says. “That’s a major advantage of this kind of system.”

The journal has posted the study in full for free.