Retrospective analysis of the effect on interval cancer rate of adding an artificial intelligence algorithm to the reading process for two-dimensional full-field digital mammography.

Abstract

Interval cancers are a commonly seen problem in organized breast cancer screening programs and their rate is measured for quality assurance. Artificial intelligence algorithms have been proposed to improve mammography sensitivity, in which case it is likely that the interval cancer rate would decrease and the quality of the screening system could be improved. Interval cancers from negative screening in 2011 and 2012 of one regional unit of the national German breast cancer screening program were classified by a group of radiologists, categorizing the screening digital mammography with diagnostic images as true interval, minimal signs, false negative and occult cancer. Screening mammograms were processed using a detection algorithm based on deep learning. Of the 29 cancer cases available, artificial intelligence identified eight out of nine of those classified as minimal signs, all six false negatives and none of the true interval and occult cancers. Sensitivity for lesions judged to be already present in screening mammogram was 93% (95% confidence interval 68-100) and sensitivity for any interval cancer was 48% (95% confidence interval 29-67). Using an artificial intelligence algorithm as an additional reading tool has the potential to reduce interval cancers. How and if this theoretical advantage can be reached without a negative effect on recall rate is a challenge for future research.