AI prediction reduces radiologists' prostate MRI-related workloads by 20%
By adjusting the parameters of a previously developed artificial intelligence algorithm aimed at spotting prostate cancer on MRI scans, experts have demonstrated the tool’s ability to significantly reduce prostate MRI workloads for radiologists.
MRI is being increasingly utilized for the diagnosis and management of clinically significant prostate cancer (csPCa), as improvements in technology have enabled the modality to improve detection rates and reduce the need for invasive biopsies. With many providers opting for a watch-and-wait approach based on serial MR exams over a prolonged period, the demand for radiologists with prostate-specific expertise is growing. Though beneficial for patients, this method has the potential to become burdensome for readers.
Experts believe this is an area where AI can be especially beneficial.
“Recent advancements in AI models have demonstrated near-expert-level accuracy in diagnosing csPCa. Such AI models are primarily studied to provide radiologists with decision support.” Thomas C. Kwee, MD, PhD, with the faculty of medical sciences at University of Groningen in the Netherlands, and colleagues noted. “Although this decision support could provide radiologists with an easy-to-assess second opinion, it does not reduce the number of cases a radiologist has to read. To improve workflow efficiency and meet rising demands, a degree of AI automation is necessary.”
To better automate the process, experts adjusted an already existing algorithm developed for csPCa detection on MRI so that it incorporates a rule out threshold on uncertainty quantification (UQ). Using the UQ threshold, the algorithm determines which scans can be automatically reported, thereby reducing the number of scans radiologists need to review.
“AI-based UQ mimics human uncertainty,” the authors explained. “The performance of AI in cases with a high UQ (ie, uncertain cases) has been shown to significantly decrease. Thresholding AI UQ allows the automation of the selection of uncertain cases for referral to human experts, while other cases could follow an autonomous workflow. This workflow is referred to as rule out.”
The algorithm was retrospectively tested on a group of more than 1,600 prostate MRIs, with its accuracy and efficacy being compared against human readers. In the rule-out pathway, AI submodels (trained on other csPCa cases) that had also been adjusted to issue UQ scores were used to interpret the scans and send any they deemed uncertain to a radiologist for a second look. Based on the algorithms’ correct predictions, the team estimated how their use would impact patient care and radiologist workflow.
Though the performances were institute-specific, the UQ scores showed promise in AI-based rule-out csPCa detection; the researchers suggested the differing performances could be owed to varying image quality at each institution. The group estimated that the UQ method could reduce the number of scans radiologists have to read by up to 20%. However, the varying performance of the UQ rule-out may signal the need for institute-specific UQ thresholds, the authors cautioned.
“Semi-autonomous AI model readings bring the risk of more biopsies, but a threshold customizable by the preferred user could enable centers to set the AI model's threshold at their level of workload reduction and diagnostic performance (eg, centers with a shortage of radiologists might want more autonomous AI readings and allow more biopsies compared to centers without sustainability issues in the current radiologists' csPCa detection pathway),” the authors explained. “Our results indicate that the current AI quality of csPCa detection shows the potential to help reduce the radiologist's workload and offer a more sustainable csPCa detection pathway solution.”
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.