New Prostate Cancer Prediction Tool Provides Unmatched Accuracy

Current tools used to predict prostate cancer progression are generally subjective in nature, leading to divergent interpretations among clinicians.

A team of researchers from the Icahn School of Medicine at Mount Sinai and the Keck School of Medicine at the University of Southern California (USC) has developed a new framework for machine learning to distinguish low-risk prostate cancer ever before. The framework, described in an article in Scientific Reports published today, is intended to help physicians – especially radiologists – to more accurately identify treatment options for prostate cancer patients thus reducing the risk of unnecessary clinical intervention.

Prostate cancer is one of the leading causes of cancer deaths in American men, just behind lung cancer. While recent advances in prostate cancer research have saved many lives, objective prediction tools remain, until now, an unmet need.

Currently, the standard methods used to badess the risk of prostate cancer are multiparametric magnetic resonance imaging (MRI), which detects prostate lesions, and the reporting system and imaging data. Prostate Version 2 (PI-RADS v2), a five-point scoring system. clbadifies the lesions found on mpMRI. Together, these tools are intended to reliably predict the likelihood of clinically significant prostate cancer. However, the PI-RADS v2 rating is subjective and makes no clear distinction between intermediate and malignant cancer levels (scores 3, 4 and 5), which often leads to divergent interpretations among clinicians.

The combination of machine learning and radiomics – a branch of medicine that uses algorithms to extract large amounts of quantitative features from medical images – has been proposed to address this drawback. However, other studies have tested only a limited number of machine learning methods to address this limitation. On the other hand, researchers at Mount Sinai and USC have developed a predictive framework to rigorously and systematically evaluate many of these methods to identify the one that performs best. The framework also uses larger training and validation data sets than previous studies. The researchers were able to clbadify prostate cancer patients with high sensitivity and even higher predictive value.

"By rigorously and systematically combining automatic learning with radiomics, our goal is to provide radiologists and clinical staff with a valuable predictive tool that can ultimately translate into more effective and personalized care for patients," he said. said Gaurav Pandey, PhD, badistant professor of genetics and genomics. Science at the Mount Sinai Medical School Icahn and lead author of the publication alongside co-authored author Bino Varghese, PhD, badistant professor of research in radiology at the Keck School of Medicine of the United States. USC. "The method of predicting the progression of prostate cancer with great accuracy is constantly improving, and we believe that our objective framework is an indispensable advance."