Artificial Intelligence in Medicine: Mayo Clinic Algorithm Supports MRI Evaluation of Prostate Cancer

Few medical centers are as venerable as the Mayo Clinic. It began when William Worrall Mayo, a physician and chemist, settled his family on the Minnesota prairie in 1864, and opened a small practice. His two doctor sons joined the practice. As other partners came on board, it grew into the Mayo Clinic. Today, its three major centers in Minnesota, Florida and Arizona are world-renowned for their integration of clinical practice, education and research.

It’s no surprise, then, that their Department of Radiology would turn to leading-edge Artificial Intelligence (AI) to improve evaluation of prostate cancer (PCa) using multiparametric MRI. At the December, 2021 gathering of international radiologists at the Radiology Society of North America (RSNA) annual meeting in Chicago, radiologist Dr. Jason Cai presented his team’s study of a Deep Learning algorithm they developed to assist radiologic readers with consistently identifying PCa.

The goal of their work was to help overcome the problem of reader variance. The ever-growing demand for multiparametric MRI (mpMRI) springs from its ability “to more accurately classify men into those with harmful prostate cancer needing treatment and those with indolent cancer that can be monitored.”^[i] Thus, hundreds—possibly thousands—of medical centers across the U.S. provide mpMRI for patients performed either on 1.5T magnets or (preferably) more powerful 3T magnets such as ours. This means at least as many readers, ranging in experience from residents in training to experts are interpreting patient scans. In turn, the quality of interpretation is inconsistent, and is correlated with experience level.^[ii]

Dr. Cai’s team hypothesized that AI could resolve variance among readers. They created a Deep Learning algorithm to predict and evaluate PCa based on the four imaging sequences (parameters) available for each patient’s scans included in the study: T2 weighted; apparent diffusion coefficient (ADC); diffusion weighted imaging (DWI); and dynamic contrast-enhanced (DCE). According to a news report of Cai’s presentation:

Cai and colleagues developed a deep-learning pipeline that included segmentation of the prostate, preprocessing of the images, and classification of findings. The algorithm was trained on internal data gathered from all three Mayo Clinic sites … [based on] a cohort of 6,137 patients … (Gleason grade groups 2 or higher) who underwent MRI between 2017 and 2019.

When applied to the task of evaluating each patient’s MRI findings, the algorithm’s performance was comparable to that of readers. Furthermore, when readers added the algorithm’s predictions to their own, the combined performance was more accurate than either interpretation alone.

AI is not intended to replace the human brain, but rather to be integrated into the workflow of busy radiologists at all experience levels. In so doing, AI can add efficiency to the job of reading numerous prostate MRI’s during a typical workday, while minimizing variance among readers. As the vision is quickly becoming the reality, it is a radiological dream come true.

NOTE: This content is solely for purposes of information and does not substitute for diagnostic or medical advice. Talk to your doctor if you are experiencing pelvic pain, or have any other health concerns or questions of a personal medical nature.

[i] Macleod LC, Yabes JG, Fam MM, et al. Multiparametric Magnetic Resonance Imaging Is Associated with Increased Medicare Spending in Prostate Cancer Active Surveillance. Eur Urol Focus. 2020;6(2):242-248.
[ii] Ruprecht O., Weisser P., Bodelle B., Ackermann H., Vogl T.J. MRI of the prostate: Interobserver agreement compared with histopathologic outcome after radical prostatectomy. Eur. J. Radiol. 2012;81:456–460. doi: 10.1016/j.ejrad.2010.12.076.