Artificial Intelligence Bests Clinicians in Flagging Prostate Cancer

August 10, 2020 - An artificial intelligence algorithm was able to identify prostate cancer from tissue slides significantly more accurately than expert pathologists and previous algorithms, according to a study published in The Lancet Digital Health.

Adenocarcinoma of the prostate is the second most common cancer diagnosed in men, with more than one million new cases diagnosed annually, researchers noted. Currently, the main way to diagnose prostate cancer is assessment of biopsy tissue, which includes core needle biopsy (CNB).

Over the past decade, there has been a dramatic increase in the number of CNBs reviewed per case, as well as an increase in overall cancer incidence and a growing shortage of pathologists. Thus, there is a heightened need to develop automated artificial intelligence-based tools to support pathologists, the group wrote.

Additionally, reliably diagnosing prostate cancer is often a challenging task for pathologists, further emphasizing the potential for AI tools to improve this care process.

“Humans are good at recognizing anomalies, but they have their own biases or past experience,” said senior author Rajiv Dhir, MD, MBA, chief pathologist and vice chair of pathology at UPMC Shadyside and professor of biomedical informatics at Pitt. “Machines are detached from the whole story. There’s definitely an element of standardizing care.” 

Researchers set out to develop an algorithm that could more accurately identify prostate cancer. The team trained the tool on images from more than a million parts of stained tissue slides taken from patient biopsies. Each image was labeled by expert pathologists to teach the AI how to distinguish between healthy and abnormal tissue.

The algorithm was then tested on a separate set of 1,600 slides taken from 100 consecutive patients seen at University of Pittsburgh Medical Center (UPMC) for suspected prostate cancer.

During testing, the AI algorithm achieved 98 percent specificity and 97 percent sensitivity at detecting prostate cancer, which is significantly higher than previously reported for algorithms working from tissue slides.

This is also the first algorithm to extend beyond cancer detection, reporting high performance for tumor grading, sizing, and invasion of the surrounding nerves. These are all clinically important features required as part of the pathology report.

The AI model also flagged six slides that were not noted by the expert pathologists.

However, researchers noted that these findings do not mean the AI tool is an adequate substitute for human clinicians. For example, in the course of evaluating these cases, the pathologist could have simply seen enough evidence of malignancy elsewhere in that patient’s samples to recommend treatment.

But for less experienced pathologists, the algorithm could serve as diagnostic support to catch cases that may otherwise go unnoticed.

“Algorithms like this are especially useful in lesions that are atypical,” Dhir said. “A nonspecialized person may not be able to make the correct assessment. That’s a major advantage of this kind of system.”

Although these findings are promising, the team stated that new algorithms will have to be trained to detect different types of cancer, because the pathology markers are not the same across all tissue types.

The results demonstrate the potential for implementing AI algorithms in the cancer diagnosis process, and the benefits of providers using these tools for enhanced clinical decision support.

“Deployment of such an AI tool in clinical practice is timely, not only because prostate adenocarcinoma is one of the most common cancers seen in men, but also due to the substantial increase in pathologists' workload as cancer cases rise, combined with increased complexity of histopathological assessment with changes in guideline recommendations,” researchers concluded.

“These data suggest that this AI-based algorithm could be used as a tool to automate screening of prostate CNBs for primary diagnosis, assess signed-out cases for quality control purposes, and standardize reporting to improve patient management. Studies reporting on deployment in additional laboratories and associated clinical utility are underway.”