Can Artificial Intelligence help Cancer Detection? Studies show promise

By Rahul Vaimal, Associate Editor
  • Follow author on
Use of Artificial Intelligence
Representational Image

In the future, artificial intelligence (AI) is likely to take the pressure off oncologists, with a recent study showing that it can identify prostate cancer with near-perfect accuracy.

The study, published by UPMC Shadyside and the University of Pittsburgh today in The Lancet Digital Health, showed the highest precision to date in identifying and characterizing prostate cancer using an AI system.

[expander_maker id=”1″]

The technology demonstrated 98% sensitivity and 97% precision in the identification of the disease during testing, significantly higher compared to previously documented studies using similar AI techniques.

The scientists presented images of more than one million parts of stained tissue slides taken from patient biopsies to train the AI for recognizing prostate cancer. Expert pathologists marked each image to teach the AI how to differentiate between healthy and anomalous tissue.

The algorithm was then used to check for potential prostate cancer on a different sample of 1,600 slides sourced from a total of 100 consecutive patients seen at UPMC.

“Humans are good at recognizing anomalies but they have their own biases or past experience,” said senior author Rajiv Dhir, chief pathologist at UPMC Shadyside and professor of biomedical informatics at the University of Pittsburgh.

“Machines are detached from the whole story. There’s definitely an element of standardizing care.”

The study also marked the first time that AI was able to reach beyond cancer detection by documenting high tumor grading performance, sizing, and assessing surrounding nerves invasion, all needed as part of a pathology report.

However, Mr. Dhir cautioned that these findings did not automatically suggest that the machines could be better detectors of cancer than humans.

Experienced pathologists, for instance, may have found enough signs of malignancy elsewhere in samples of that patient to suggest treatment.

The algorithm might serve as a fail-safe to capture cases that would otherwise be overlooked for less experienced pathologists though.

“A non-specialized person may not be able to make the correct assessment. That’s a major advantage of this kind of system.”

Though these findings are encouraging, Mr. Dhir stressed on the need to train new algorithms to detect various forms of cancer. The markers for pathology are not common for all types of tissues. But he did not see why it should not be achieved, for example, to adapt this technique to deal with breast cancer.