AI distinguishes between low, high-risk prostate cancer on MRI

A multi-institutional team of researchers has developed a new AI learning algorithm that can distinguish between low- and high-risk prostate cancer from multiparametric (mpMRI) scans with higher sensitivity and predictive value than current risk assessment approaches, according to research published online Feb. 7 in the journal Scientific Reports.  

After testing various types of machine-learning classifiers, the researchers found the best performing classifier could potentially outperform assessments produced using Prostate Imaging Reporting and Data System (PI-RADS) version 2.  

The team of experts from the Icahn School of Medicine at Mount Sinai in New York City and the Keck School of Medicine at the University of Southern California in Los Angeles, hope the AI technology—which combines machine learning and radiomics—can help radiologists more accurately identify prostate cancer treatment for patients.  

“By rigorously and systematically combining machine learning with radiomics, our goal is to provide radiologists and clinical personnel with a sound prediction tool that can eventually translate to more effective and personalized patient care,” senior author Gaurav Pandey, PhD, assistant professor of genetics and genomic sciences at the Icahn School of Medicine, said in a prepared statement from Mount Sinai. “The pathway to predicting prostate cancer progression with high accuracy is ever improving, and we believe our objective framework is a much-needed advancement.” 

Currently, the standard methods used to assess prostate cancer risk are multiparametric magnetic resonance imaging (mpMRI)—which detects prostate lesions—and the Prostate Imaging Reporting and Data System, version 2 (PI-RADS v2)—a five-point scoring system that classifies lesions found on mpMRI. This can lead to variability in radiologists’ interpretations.  

Although both tools are intended to accurately predict the likelihood of prostate cancer risk, PI-RADS v2 scoring is subjective and does not clearly distinguish between intermediate and malignant cancer levels (scores 3, 4, and 5), according to the researchers.  

Because of this, the researchers developed a predictive framework that systematically assessed different kinds of machine-learning methods using 110 radiomic features.  

Quadratic kernel-based support vector machine (QSVM), the best classifier, was then tested with a testing set of 54 prostate cancer patients. The results were then compared with PI-RADS v2 classification results, according to the researchers.  

The PI-RADS v2 classification had a slightly higher overall area under the curve (AUC) than the machine-learning classifier (0.73 versus 0.71). However, the classifier produced better overall results when its performance was assessed with measures more appropriate for unbalanced cohorts. In other words, when there were more low-risk cases than high-risk cases.

“The goal of incorporating machine learning into radiomics is not to compete with the radiologist, but to rather provide the radiologist and physician team taking care of the patient with objective prediction tools that can aid personalized decision-making regarding individual disease course and treatment outcome," the team wrote.  

From their findings, the researchers believe the use of decision support systems will enhance the quality of radiologist's work and, down the road, patient care.