AI model matches radiologists at detecting prostate cancer on MRIs

An artificial intelligence model can match radiologists at detecting clinically significant prostate cancer on MR images, according to new research published Tuesday in RSNA’s flagship journal.

Prostate cancer is the second most common form of the disease among men, typically diagnosed using multiparametric MRI, experts note. However, interpretation of these exams can be difficult, with higher performance among more experienced radiologists.

Utilizing AI has demonstrated potential for reducing inconsistencies from one human reader to the next. But a “major drawback” of traditional artificial intelligence approaches is the need for radiologists or pathologists to annotate each lesion. This “resource intensive” process is time consuming and can limit the size of datasets.

“This applies not only at the time of initial model development but also at the time of model reevaluation and retraining after clinical implementation,” lead author Jason C. Cai, MD, with the Department of Radiology at the Mayo Clinic in Minnesota, and colleagues wrote Aug. 6 in Radiology [1]. “The aim of this study was to develop a [deep learning] model to predict the presence of [clinically significant prostate cancer] using patient-level labels (the presence or absence of csPCa) without information about tumor location and compare its performance with that of radiologists.”

For the study, Cai and co-authors retrospectively reviewed data from patients without known cancer who underwent MRI at the noted institution between 2017 and 2019. Across 5,735 examinations, 1,514 showed clinically significant prostate cancer. Using a test set of 400 exams, and an external dataset of 200 more, the AI model’s performance matched that of experienced abdominal radiologists. Combining deep learning with physicians’ findings resulted in better performance than AI alone, the authors noted.

Experts view the system as a potential assistant to radiologists, helping them bolster detection rates while reducing false positives.

“I do not think we can use this model as a standalone diagnostic tool," study co-author Naoki Takahashi, MD, also with Mayo, said in an announcement from the Radiological Society of North America. “Instead, the model's prediction can be used as an adjunct in our decision-making process."

Mayo researchers are continuing to expand the dataset, which is now double in size, RSNA reported. They hope to eventually conduct a prospective study and see how physicians interact with AI in real world scenarios.

"We'd like to present the model's output to radiologists and assess how they use it for interpretation and compare the combined performance of radiologist and model to the radiologist alone in predicting clinically significant prostate cancer," Takahashi said in the announcement.

The study utilized multiparametric MRI. But currently the field is shifting toward biparametric prostate MRI, motivated by the need to limit scan durations, eliminate costs and reduce risks associated with contrast, experts wrote in a corresponding editorial [2]. Extending this AI approach to faster bpMRI could serve to “substantially amplify its utility and impact.”

“Reducing the cost and increasing the accessibility of MRI for [clinically significant prostate cancer] is critical for the viability of MRI screening programs. The study by Cai et al. represents a step toward MRI screening for csPCa by potentially decreasing the interpretation burden,” Patricia Johnson, PhD, and Hersh Chandarana, MD, MBA, with the Department of Radiology at the NYU Grossman School of Medicine, wrote Tuesday. “As we continue to refine these technologies and methods, the goal of providing effective, accessible and cost-efficient screening tools moves closer to reality. This progress promises to improve patient outcomes through earlier and more accurate diagnosis of csPCa,” they added later.

Marty Stempniak

Marty Stempniak has covered healthcare since 2012, with his byline appearing in the American Hospital Association's member magazine, Modern Healthcare and McKnight's. Prior to that, he wrote about village government and local business for his hometown newspaper in Oak Park, Illinois. He won a Peter Lisagor and Gold EXCEL awards in 2017 for his coverage of the opioid epidemic. 

Around the web

After reviewing years of data from its clinic, one institution discovered that issues with implant data integrity frequently put patients at risk. 

Prior to the final proposal’s release, the American College of Radiology reached out to CMS to offer its recommendations on payment rates for five out of the six the new codes.

“Before these CPT codes there was no real acknowledgment of the additional burden borne by the providers who accepted these patients."

Trimed Popup
Trimed Popup