New studies showcase how researchers around the world are leveraging machine learning to aid pathologists in detecting prostate cancer.

Artificial intelligence (AI) refers to the use of “smart” machines to complete tasks that usually require human intelligence. Although it has applications in many fields, its use in health care has tremendous potential to revolutionize the diagnosis and treatment of numerous illnesses, including cancer. In two separate studies published earlier this year, researchers demonstrated how AI-based algorithms matched (and, in some cases, exceeded) a human pathologist’s ability to detect prostate cancer in both imaging results and tissue samples.

Let’s take a closer look at the results of these studies and their potential implications for patients.

Study #1: Radboud University Medical Center

Published in January, the study conducted by researchers at Radboud University Medical Center in the Netherlands looked at how a new AI-based “deep learning” system developed by the researchers compared to human pathologists when identifying the aggressiveness of prostate tumors. When researchers compared the AI-based system’s performance to the performance of 15 human pathologists, they found that the system performed better than 10 of the pathologists, noting that its ability to evaluate the aggressiveness of prostate tumors was comparable to that of a highly experienced pathologist. The researchers are now working with scientists at the Karolinska Institute in Sweden and Kaggle, a subsidiary of Google that specializes in data science competitions, to further improve the algorithm on which the system is based as part of a major international competition.

Study #2: University of Pittsburgh

Researchers from the University of Pittsburgh and the University of Pittsburgh Medical Center (UPMC) published a study in The Lancet in July that demonstrated extraordinary accuracy in the recognition and characterization of prostate cancer using an AI-based program. The program, which researchers “trained” using more than 1,000,000 parts of stained tissue slides that were labeled by expert pathologists, demonstrated 98 percent sensitivity and 97 percent specificity at detecting prostate cancer among a separate set of 1,600 slides from 100 patients who were evaluated at UPMC for suspected prostate cancer. Furthermore, the AI program also performed well in the areas of tumor grading, sizing, and invasion of surrounding nerves, and identified the presence of abnormal cells in six slides that were previously overlooked by pathologists.

Innovation that Benefits Providers and Patients

While AI-based algorithms and programs will not soon replace human pathologists as the gold standard in detecting prostate cancer in patients, the results of these studies demonstrate the value that such innovations could bring to health care, particularly as a form of quality control for less experienced pathologists. The results also offer hope that similar algorithms could be developed to aid in the detection and diagnosis of other types of cancer, potentially helping patients get diagnosed and begin treatment earlier in the course of the disease.