Computer-Aided Breast Cancer Detection

Recent study findings raise questions about how well computers interpret mammograms.

We live in an ever increasing world of mechanized and automated processes, which are supposed to be better or faster than the work that humans can do. On occasion, researchers find that humans are superior to machines. In the August 3, 2011 issue of the Journal of the National Cancer Institute, researchers from the University of California in Davis reported that computer-aided mammography interpretation did not detect early or invasive breast cancers any better than mammograms interpreted by radiologists. In addition, humans were more accurate than computers in detecting abnormalities.

The researchers reviewed 1.6 million screening mammograms that were interpreted at Breast Cancer Surveillance Consortium facilities in 7 states from 1998 to 2006. Of the 90 facilities in the study, 27.8% had adopted computer-aided mammography detection during the study period. After adjusting for factors such as hormone therapy use, year of mammography interpretation, and other factors, use of computer-aided detection was associated with lower specificity and lower positive predictive value. There was no difference between the computer and humans in identifying conditions that warranted a biopsy, although the computer-read mammograms were associated with a higher rate of recall for further testing. Facilities using computer-aided detection to help interpret mammography films weren't more likely to find invasive or earlier stage tumors and actually had lower accuracy in detection.

The researchers concluded that computer assistance in analyzing routine screening mammograms is of little assistance in detecting breast cancer in clinical practice, and may be potentially harmful when it triggers unnecessary additional testing.

Reference

Fenton JJ, Abraham L, Taplin SH, et al. Effectiveness of computer-aided detection in community mammography practice. J Natl Cancer Inst 2011; e103 (15): 1152-1161.