Artificial Intelligence, Diagnostics

AI algorithm for detecting prostate cancer shows more than 98% sensitivity, 97% specificity in study

Tel Aviv-based Ibex Medical Analytics published data on what it called the first algorithm to go beyond detection and into areas like tumor grading and sizing.

AI, machine learning

An Israeli startup developing a digital pathology system based around artificial intelligence has published what it calls “outstanding outcomes” in a clinical validation study.

Tel Aviv-based Ibex Medical Analysis said Tuesday that it had published data on Galen Prostate, its AI-based system for use by pathologists to detect and measure prostate cancer, in The Lancet Digital Health. The company called it the first and only AI-based system used by pathologists in routine clinical practice. The study took place at the University of Pittsburgh Medical Center, led by Drs. Liron Pantanowitz and Rajiv Dhir, both pathologists at UPMC’s Shadyside Hospital.

According to the data, sensitivity measured for prostate cancer was 98.46%, and specificity was 97.33%, while the operating characteristic curve was 0.991. The company said all of these were higher than previously reported metrics for AI algorithms in pathology, and that this was the first such algorithm to go beyond detection and extend into tumor grading, detection of perineural invasion and tumor sizing.

“This is the first time in a peer-reviewed publication that we’re reporting on deployment of an actual clinical use of a solution that is AI-based,” said Daphna Laifenfeld, Ibex’s chief scientific officer, in a phone interview.

To conduct the study, the team led by Pantanowitz and Dhir used blinded whole-slide images of prostate core needle biopsies and fed them into the Galen system. They then assessed the output from the algorithm against their clinical pathology reports, followed by blinded discrepancy resolution.

CEO Joseph Mossel said in the same phone interview that work by companies like Ibex, as well as NewYork-based Page.ai and Stockholm-based ContextVision – both of which he cited as Ibex’s nearest competitors, despite different approaches – is particularly important due to the severe shortage of pathologists. With fewer new doctors going into the field, he said, pathologists often find themselves overworked and in need of tools akin to the autopilot used on airplanes.

“It’s a nascent field, so while we’re competing, we’re helping each other build a new market and a new field,” Mossel said. “As much as we want to compete, we’re also working together in a way, educating the market.”

Photo: Andrzej Wojcicki, Getty Images

Shares1
Shares1