Room: Track 1
Purpose: To use eye tracking to compare different causes for missing a cancer (FN) when reading digital mammography (DM) compared to digital breast tomosynthesis (DBT).
Methods: Ten MQSA radiologists, or current women’s imaging fellows were recruited. Each observer read over 4 sessions, with at least a 2 week wash out period, the same 20 cases. 10 cases were biopsy proven positive cancers, subtle on imaging, and 10 were difficult normal cases. Each case was read as: a digital mammogram (DM), a DBT, and a DBT with a 2D preview. The preview was either a DM or a synthetic mammogram (SM) generated from the DBT acquisition. We used a SMI RED250mobile Eye Tracker set at a sampling frequency of 250Hz. We computed the number of search, recognition, and decision errors, as previously described in the literature.
Results: As expected, the FN rate was higher for DM and DBT. While there were search (n=6) and recognition (n=7) errors for DM (13 total), these were lower for DBT (5 total), and uncommon for DBT with a preview (2 total). For all 4 different reading conditions, a decision error was the dominant reason for a FN. Fewer search errors is consistent with the 2D area covered by radiologists’ search patterns, which was higher for DBT. However, the use of a 2D preview image, reduced the area covered, presumably because the preview keys the radiologist as to the location of a cancer. Further, there were differences between DM and DBT in terms the dwell time on a location in the image for true positive, false positive, false negative, and true negative decisions. Compared to DM, FN had longer dwell times than FP for DBT.
Conclusion: In spite of the larger volume of images to read, search and recognition errors were lower in DBT than DM.
Funding Support, Disclosures, and Conflict of Interest: Robert Nishikawa receives royalties from Hologic, Inc. and has research contracts with Hologic, Inc. and GE Healthcare.
Not Applicable / None Entered.