Machine learning model accurately predicts who would benefit most from mpMRIs

A novel machine learning model could accurately predict which men might benefit most from additional imaging before a prostate biopsy, saving patients both money and discomfort, a new study states.

The study, which is part of the American Roentgen Ray Society’s annual symposium this spring, comes less than a year after results of the PROMIS trial—a study that proved the efficacy of multiparametric MRI (mpMRI) scans prior to biopsies in prostate cancer patients—were published in the Lancet.

PROMIS results were “groundbreaking,” the Society of Radiographers wrote, since implementing mpMRI imaging before a biopsy could save at least a quarter of patients thousands of dollars and the trouble of undergoing an invasive surgery that requires inserting between 10 and 18 needles into the prostate. According to the Society, prostate biopsies are not only painful, but lead to infection in 7 percent of cases.

A team at the University of Rochester took the diagnostic process a step further, developing a machine learning model that can accurately predict which patients are most likely to benefit from additional mpMRI screening. A release from ARRS stated the model will aid in patient selection “to optimize resource utilization and reduce unnecessary costs.”

Researchers on the project reviewed 811 prostate mpMRI exams from four tertiary care centers and used the information to develop a support vector machine model that could predict PI-RADS category 4 or 5 lesions on the basis of a patient’s age, prostate specific antigen and prostate volume, according to the release. The model was created using the Microsoft Azure Machine Learning platform and was prospectively tested—with success—on 42 patients.

The model showed 73 percent accuracy for predicting PI-RADS category 4 or 5 lesions, the researchers reported. Prospective validation of the model demonstrated a sensitivity of 75 percent and specificity of 82 percent for a cutoff threshold of 43 percent for predicting category 4 or 5 lesions.

The Society of Radiographers said the PROMIS trial paved the way for a “diagnostic breakthrough” in the field, but noted physicians are often asked to implement complicated new technologies like mpMRI without formal introductions to the tools.

“We...recognize that mpMRI before biopsy is a complex technique and that if it is to produce strong outcomes for men, it must be conducted consistently everywhere, to a clear set of standards,” the Society wrote. “Radiographers are a critical part of this new prostate cancer diagnostic workforce and can be well-placed to make the case for prostate MRI to be adopted at their places of work.”

The University of Rochester Medical Center’s Zachary Nuffer, MD, is set to present his team’s findings and successful machine learning model at the ARRS’s annual meeting this April in Washington, D.C.