AI use in MRI-based prostate cancer screening remains limited
Artificial intelligence has proven its utility in numerous radiology settings, but new data suggest its potential may be limited in prostate MRI applications.
Prostate MRIs are routinely administered after prostate specific antigen (PSA) screenings to guide biopsy decisions. These scans have yielded promising results in reducing unnecessary biopsies, but they also have increased radiologists’ workloads. Artificial intelligence has been proposed as a way to ease these burdens, similar to its proliferation in breast cancer screening.
“AI supported workflows have shown valuable evidence in assisting radiologists in MRI interpretations and decision-making in diagnostic settings,” Deependra Singh, with the Early Detection, Prevention, and Infections Branch, International Agency for Research on Cancer in France, and colleagues explained. “MRI-based cancer screening presents opportunities to enhance performance and streamline clinical workflows, particularly in risk-stratified approaches."
Published in the European Journal of Radiology, the new findings question the use of AI in these settings due to numerous shortcomings, including issues with over detection, low specificity and variable agreement. The issues uncovered were primarily based in prostate screening settings, prompting experts to question its utility.
For their work, the group conducted an overview of all published literature on AI in prostate cancer MRI screenings, which included more than 200 papers. This was further narrowed down to 47, but only two met the final inclusion criteria. Multiple use cases of AI in these settings were reviewed, from detection responsibilities to biopsy decision support and agreement with radiologist interpretations.
A commercially available AI tool was used in both studies. The papers highlighted issues with agreement between the tool and expert radiologists, with concurrence judged as poor to moderate. The team also observed significant rates of over detection and low specificity, which could prompt patients to undergo unnecessary, often invasive tests to rule out cancer.
These findings could be owed to the fact that there is little real-world clinical data on the use of AI in prostate cancer MRI applications, the authors noted. However, they believe these findings could be used to improve algorithm development in the future.
“Most AI algorithms have been trained and tested on diagnostic datasets or retrospective cohorts. Thus, deploying AI to improve quality of care in a screening setting is a challenge to prove,” the authors noted. “The right balance between sensitivity and specificity with adjustment for AI’s threshold for interpreting MRI scans is crucial. Training of radiologists on using AI output to provide MRI results, their trust in AI judgement, and adaptation in every day clinical decision are a few of the several key aspects to consider.”
Read more here.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.