Breast screening trial favors computer-aided detection

April 1, 2008

Method shows higher sensitivity, recall rate than single interpretation but lower recall rate than double reading


An evaluation of nearly a quarter million mammograms suggests screening mammography with computer-aided detection is more sensitive than double reads. The findings contradict a 2007 study that questioned breast CAD's effectiveness.

Mammographic CAD remains controversial. Radiologists argue its misuse drives up recall rates and, with them, the number of unwarranted biopsies and overall mammography costs. CAD came under close scrutiny after a study published last year in The New England Journal of Medicine concluded the technology could worsen the interpretation of mammograms rather than improve it.

The latest study, by Dr. Matthew Gromet, section chief of breast imaging at Charlotte Radiology in North Carolina, sets the pendulum swinging in the opposite direction.

Charlotte Radiology performed double reads of all screening mammograms until 2003, then shifted to single reading with CAD. The policy revision presented a perfect opportunity for comparing the two approaches. The findings were published in the April issue of the American Journal of Roentgenology.

Gromet retrospectively reviewed 231,221 screening mammograms interpreted by nine experienced mammographers from January 2001 to December 2005. Split nearly evenly between double and single reads, 112,413 (48.6%) studies were double-read, and 118,808 (51.4%) were single-read with CAD.

The first interpretation of the double-reading process yielded a recall rate of 10.2%. Sensitivity and positive predictive values were 81.4% and 4.1%, respectively, while the biopsy-confirmed cancer detection rate was 4.12 per 1000. After the second read, the recall rate and sensitivity rose&emdash;to 11.9% and 88%, respectively&emdash;as did the cancer detection rate (to 4.46 per 1000). The benefit came at the expense of 140 additional biopsies, however, resulting in a drop in positive predictive value to 3.7%.

Single reading with CAD yielded a recall rate of 10.6%, with sensitivity of 90.4%, positive predictive value of 3.9%, and cancer detection rate of 4.20 per 1000. CAD increased sensitivity and the recall rate compared with single interpretation without CAD and reduced the recall rate compared with double reading. All the findings were statistically significant (p

"As a large single-institution program with relative consistency of methods, patients, radiologists, and data collection, we believe that results from our experience before and after CAD implementation could provide additional evidence about the usefulness of this technology," Gromet said.