Upgrading to AI-based CAD software leads to fewer false-positive mammograms
Using computer-aided detection (CAD) software powered by artificial intelligence leads to fewer false-positive mammograms, according to new findings published by the Journal of Digital Imaging. Significant cost savings could also be realized by making such a switch.
“The largest disadvantage of using currently available CAD systems is the high rate of false-positive marks,” wrote author Ray Cody Mayo, MD, of the University of Texas MD Anderson Cancer Center in Houston, and colleagues. “One metric to assess the usefulness of CAD is the count of these marks on each image—false positives per image (FPPI). False-positive marks may distract the interpreting radiologist with too much “noise” and could lead to unnecessary workups and biopsies.”
The researchers used data from 245 mammograms performed using 2D full-field digital mammography from January 2013, to March 31, 2013. All examinations had previously been interpreted using CAD software and contained archived CAD markings. All images were anonymized before the study began. A head-to-head trial then occurred on the dataset as a recently developed AI-CAD algorithm was compared to traditional CAD software.
“To our knowledge, this is the first published study in the peer reviewed literature comparing the FPPI of an AI-CAD to a conventional CAD using the same test set,” the authors wrote.
The three cancer cases in the dataset were identified correctly by both the CAD system and the AI-CAD system. Using AI-CAD, however, led to a 69% reduction in overall FPPI. While there was an 83% reduction in FPPI for calcifications, there was a 56% reduction for mass. The overall FPPI was 0.29 for the AI-CAD system and 0.92 for the traditional CAD system.
Also, using traditional CAD, 21 cases were categorized as BI-RADS 0 and the patient was recalled. Eighteen of those cases were false-positive recalls, confirmed to be benign through either a biopsy or long-term follow-up. Looking closer at those 18 false-positive recalls, 15 of them had CAD marks while just eight had marks from the AI-CAD system.
“Ultimately, the radiologist will make the final decision regarding BIRADS assessment, but their decisions are influenced by CAD marks,” the authors wrote. “In order to reduce false-positive BIRADS assessments, the most desired improvement for imaging analysis software in mammography today is reduction in FPPI. In our study, 83% of the false-positive BIRADS 0 cases had CAD marks, whereas only 56% of these cases had AI-CAD marks.”
Mayo et al. noted this reduction in FPPI could lead to “fewer false recalls, improved workflow and decreased costs.”
“Fewer recalled patients means more availability to address backlogs for both screening and diagnostic appointments,” they wrote. “Departmental screening throughput may be increased without adding staff, equipment, or facility hours since typically two screening exams are often allocated the same time as one diagnostic exam on the schedule.”
Looking closer at the possible cost savings associated with moving to an AI-CAD system, the authors used an assumption of 100,000 breast cancer screening mammograms per year and the 2017 Medicare fee schedule to determine that “as much as $1,488,362 per year of increased global revenue” could be realized.
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.