AI identifies prostate cancer with near perfect accuracy, study finds

Scientists found machines were able to spot some malignancies missed by experienced pathologists

AI may be better at spotting prostate cancer than humans. AFP
Powered by automated translation

Machines may take the burden off oncologists in the future, with a new study finding that artificial intelligence can identify prostate cancer with near-perfect accuracy.

The research, published on Tuesday in The Lancet Digital Health by UPMC Shadyside and the University of Pittsburgh, shows the highest accuracy to date in recognising and characterising prostate cancer with an AI program.

During testing, the technology showed 98 per cent sensitivity and 97 per cent specificity at detecting the illness, significantly higher than previously reported with similar AI techniques.

To train the AI to recognise prostate cancer, the scientists provided images from more than a million parts of stained tissue slides taken from patient biopsies.

Each image was labelled by expert pathologists to teach the AI how to discriminate between healthy and abnormal tissue.

The algorithm was then tested on a separate set of 1,600 slides from 100 consecutive patients seen at UPMC for suspected prostate cancer.

"Humans are good at recognising 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 standardising care."

The study also marked the first time AI was able extend beyond cancer detection, reporting high performance for tumour grading, sizing and invasion of the surrounding nerves, all required as part of a pathology report.

Mr Dhir said these results did not necessarily indicate the machines can be better cancer detectors than humans.

For example, experienced pathologists may have seen enough evidence of malignancy elsewhere in that patient's samples to recommend treatment.

For less experienced pathologists, though, the algorithm could act as a fail-safe to catch cases that might otherwise be missed.

AI also flagged six slides that were not seen by the expert pathologists.

"Algorithms like this are especially useful in lesions that are atypical," Mr Dhir said.

"A non-specialised person may not be able to make the correct assessment. That's a major advantage of this kind of system."

While new algorithms will have to be trained to detect different types of cancer, scientists believe that it is possible adapt the technology to work with other types.