How should AI-based decision support systems be integrated into radiologist workflows?
Decision support (DS) systems based on artificial intelligence (AI) can improve the diagnostic performance of radiologists, but what’s the best way to integrate those DS systems into a reader’s workflow? Researchers tested two different reading methodologies, sharing their findings in the Journal of Digital Imaging.
The authors decided to focus on how DS systems can improve ultrasound (US) due to its vital role in breast imaging.
“US is a valuable imaging modality used to detect primary breast malignancy,” wrote author Lev Barinov of Princeton University in Princeton, New Jersey, and Rutgers University Robert Wood Johnson Medical School in New Brunswick, New Jersey, and colleagues. “However, radiologists have a limited ability to distinguish between benign and malignant lesions on US, leading to false-positive and false-negative results, which limit the positive predictive value of lesions sent for biopsy (PPV3) and specificity.”
The authors compared DS implementation with two reading methodologies: sequential, when the radiologists can assess the case before the DS is presented, and independent, when the case and DS are presented at the same time.
Overall, Barinov and colleagues found that there may be “a strong impact of confirmation bias in sequential study design” because “the deviation from the control assessment is significantly smaller in the sequential read verses independent read.” Independent reads, however, showed “dramatic shifts in reader performance and inter-operator variability," they added.
“The evidence provided in this study can be used to impact both study design when demonstrating efficacy of new diagnostic decision support tools, as well as their implementation in practical environments,” the authors concluded.
Michael has more than 18 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.