Artificial intelligence can automatically determine if patients are best suited to undergo prostate MRI scans with or without contrast. Incorporating the tool into clinical workflows may save radiology departments time and money.
As urology guidelines continue to recommend MRI scans for prostate cancer care, demand for such exams has skyrocketed, researchers explained in Insights into Imaging.
Lately, biparametic MRI has gained favor for its shorter exam times and diagnostic similarities to complete multiparametric prostate scans, which include contrast. Rads ideally could weigh the pros and cons for each protocol—no contrast side effects and optimized workflow versus fewer indeterminate lesions and less experience required with mpMRI—but time is typically at a premium.
So Swiss physicians developed a neural network to automate the process. Based on thorough initial testing, the tool would have significantly helped busy clinicians, Andreas M. Hötker, with the University Hospital Zurich’s Institute of Diagnostic and Interventional Radiology, and co-authors explained.
“The CNN would have correctly assigned 78% of patients to a biparametric or multiparametric protocol, with only 2% of all patients requiring reexamination to add dynamic-contrast enhanced sequences,” the group wrote Aug. 9. “Hence, integration of AI into quality assessment and decision making could allow for shorter examination times and a more streamlined clinical workflow, while maintaining diagnostic accuracy by including DCE only when truly needed,” they added later.
The researchers developed their platform using 300 prostate MRIs, validating it on a separate set of 100 scans from the same vendor and 31 from a different organization.
Overall, the CNN reached a sensitivity of 94.4% and specificity of 68.8% for determining if patients required contrast. It also beat out a radiology technician, who reached 63.9% sensitivity and 89.1% specificity, the authors noted.
“The AI could automatically detect a scan that might require the acquisition of DCE sequences with high accuracy and alert the attending radiologist (who still has to supervise the application of contrast agent[s] due to legal reasons and the possibility for adverse reactions),” Hötker and colleagues wrote.
The neural network did stumble when tested on scans from a different vendor but including more images from this organization in the training cohort would likely bolster the AI’s accuracy.
Read the full story here.
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.