Monday, December 1 | 3:00 p.m.-3:10 p.m. | M7-SSBR04-1 | Room E451A
Using AI as a standalone reader for screening mammograms when paired with one radiologist leads to more invasive cancers detected and fewer recalls, according to results to be shared in this session.
In her presentation, Karin Dembrower, MD, PhD, from the Karolinska Institute in Solna, Sweden, will discuss how implementing a commercial AI system adds diagnostic value to breast imaging for single reading.
“Compared to pairing AI with one radiologist and adding a second radiologist, the only difference was recalling more healthy women,” the research team highlighted.
The team analyzed three time periods with different reader settings. This included period one (two radiologists without AI), period two (two radiologists with AI), and period three (one radiologist with AI).
Period one had a recall rate of 2.6%, a cancer detection rate of 0.4%, a positive predictive value (PPV) of 16.6%, and a cancer detection rate of invasive cancers of 0.3%.
Period two showed a recall rate of 3% (+13.1%), cancer detection rate of 0.5% (+20.5%), PPV of 22.6% (+36.1%), and a cancer detection rate of invasive cancers of 0.4% (+28.9%).
And period three demonstrated a recall rate of 2.5% (-3.1%), cancer detection rate of 0.5% (+16.1%), PPV of 25.6% (+54.2%), and a cancer detection rate of invasive cancers of 0.4% (+34.5%).
The team also reported statistical significance for differences between the following measures: Lower recall rate for period two vs. period one (p = 0.013) and period three vs. period one (p < 0.001); lower PPV for period two vs. period one (p = 0.001) and period two vs. period three (p = 0.036); and higher invasive cancer detection rates for period three vs. period one (p = 0.022).
How much do these results point toward AI being favorably used as one reader of screening mammograms? Attend this session to find out more.
