Lyndon Rakusen¹ (), Brandon Eich¹, Stephen Adamo¹; ¹University of Arizona

For over 60 years, radiologists have known they are more likely to miss a second abnormality after a first is detected in the same image. This Subsequent Search Miss (SSM) effect (also known as Satisfaction of Search in radiology) persists across many imaging modalities and has more recently been shown to cause misses in breast cancer detection with radiologists and novice observers. Because breast cancer accounts for 31% of all new cancer diagnoses and 10 to 30% of breast cancers are not reported, efforts to improve detection are critical. 3D breast imaging (tomosynthesis) has yielded better detection of cancer and fewer false alarms than a standard mammogram (a 2D representation of breast tissue), as radiologists can search in-depth without normal tissue obstructing their view. However, it is unclear how search performance for multiple cancers is affected when searching with tomosynthesis, particularly the extent to which SSM errors persist compared to mammography. In the current study, novice observers were asked to detect up to two masses within synthesized virtual tomosynthesis and mammography images. These images were created using VICTRE, a fully in-silico tool for generating realistic-looking breast images. By using VICTRE to generate mammography and tomosynthesis images, we could control the mass locations between image types, allowing us to compare SSM errors between modalities. Overall, observers: 1) were more accurate in cancer detection in tomosynthesis, 2) were more accurate in finding a second target in tomosynthesis, and 3) had a significant SSM effect in both, with no differences between the size of their SSM effect. Together, these results suggest that tomosynthesis should be used more frequently when attempting to detect breast cancer as it can improve detection when there are standalone and multiple masses. However, tomosynthesis may not alleviate the SSM effect compared to mammography.

Acknowledgements: Grant from the National Institutes of Health, National Cancer Institute (4R00CA267163-03)