This article is more than 5 years old. Information may no longer be current.
Machine learning framework ‘much-needed advancement’ in prostate cancer risk assessment
Click Here to Manage Email Alerts
A new machine-learning framework enabled researchers to assess prostate cancer risk with more accuracy than the current standard methods, according to results of a single-center, retrospective study published in Scientific Reports.
The framework aims to help radiologists make personalized treatment decisions for men with prostate cancer without unnecessary clinical intervention.
“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,” Gaurav Pandey, PhD, assistant professor of genetics and genomic sciences at Icahn School of Medicine at Mount Sinai, said in a press release. “The pathway to predicting prostate cancer progression with high accuracy is ever improving, and we believe our objective framework is a much-needed advancement.”
The American Cancer Society estimates there were 164,690 new cases of prostate cancer and 29,430 deaths related to the disease in the United States last year, according to study background.
Current methods for assessing prostate cancer risk include multiparametric MRI, which identifies lesions, and the Prostate Imaging Reporting and Data System version 2 (PI-RADS v2) which classifies the lesions based on a five-point scoring system. Although these tools, used in combination, have shown accuracy in predicting the likelihood of intermediate- and high-grade cancers, interpretation is subjective, whereas the machine-learning framework is based on algorithms and can be more precise, Pandey and colleagues wrote.
Researchers retrospectively analyzed 68 men with prostate cancer who underwent multiparametric MRI followed within 2 months by transrectal ultrasound-MRI fusion guided biopsy of the prostate and organized them as high risk (n = 14) or lower risk (n = 54) based on National Comprehensive Cancer Network guidelines.
The researchers applied the machine learning-based framework and its components on radiomics features derived from study cohorts. The framework, consisting of classification, cross-validation and statistical analyses, identified the Quadratic kernel-based SVM as the best-performing classifier for prostate cancer risk stratification.
In an independent validation set, the classifier equaled PI-RADS v2 in area under the receiver operating characteristic curve — routinely used to gauge classification model performance — but performed substantially better in class-specific measures including F-measure, precision and recall.
The study’s retrospective nature, as well as its small sample size, served as limitations.
“Identifying patients with significant [prostate cancer] remains a challenging problem,” Pandey and colleagues wrote. “Although [multiparametric MRI] is a useful tool for this purpose, the reported inter-observer agreement among its interpretations has been variable ... depending on the study and reader experience.”– by John DeRosier
Disclosures: The authors report no relevant financial disclosures.
Read more about
Click Here to Manage Email Alerts