Report from ARRS: CAD scores high marks in digital mammography

May 9, 2007

The iCAD software package achieved high sensitivity with both computed radiography and full-field digital mammography, according to two new studies presented at the American Roentgen Ray Society meeting on Tuesday. In the same session, however, Dr. Edward Sickles warned that such positive studies may reflect optimal use and that radiologists should ensure they apply CAD software properly.

The iCAD software package achieved high sensitivity with both computed radiography and full-field digital mammography, according to two new studies presented at the American Roentgen Ray Society meeting on Tuesday. In the same session, however, Dr. Edward Sickles warned that such positive studies may reflect optimal use and that radiologists should ensure they apply CAD software properly.

Studies about CAD use with screen-film mammography abound, but research using digital imaging is limited, said iCAD vice president and medical director Dr. Jeffrey Hoffmeister, who presented results. The iCAD software is currently approved for FFDM and is being used on an investigational basis with Fuji's CR unit.

In the first study, researchers evaluated the use of the iCAD SecondLook v7.2 package in assessing 53 screen-detected cancers and 36 normal cases at two CR sites. CAD detected 47 out of 53 cancers, for a sensitivity of 89%, with 2.2 false positives per normal case. Sensitivity reached 92% for calcifications and 88% in masses. In the 18 small cancers, sized 1 mm to 10 mm, sensitivity was good at 83%.

Furthermore, CAD correctly marked 91% of cancers in nondense breasts and 85% in dense tissue. An analysis of CAD sensitivity by histologic type revealed that CAD marked 92% of ductal carcinoma in situ (12 of 13 cases), 89% of invasive ductal carcinoma (31 of 35), and 80% of invasive lobular and other cancers (31 of 35).

"CAD with CR has high sensitivity for detection of breast cancer and may result in early detection of breast cancer. Sensitivity was maintained with density," Hoffmeister said.

In the second study, researchers evaluated iCAD with the GE Senographe FFDM unit at five user sites and compared results with CAD in the screen-film environment. The study included 45 biopsy-proven cancers and 38 normal cases.

CAD correctly identified 40 out of 45 cancers, for a sensitivity of 89%, with 1.6 false positives per normal case. In comparison, CAD reached 90% sensitivity with 1.9 false positives per normal case in a large screen-film mammography study.

"CAD sensitivity is statistically the same in digital and film mammography, with a similar false-positive rate. The results indicate CAD with digital enhances radiologists' effectiveness in detecting breast cancer," Hoffmeister said.

During a lecture about outcomes research in the same session, however, Sickles warned that despite strong evidence about CAD's value, performance in general radiology practice may be lacking. Sickles was coauthor of a controversial study published in April in The New England Journal of Medicine that found that the use of CAD actually diminished performance (NEJM 2007:14:1399-1409).

"That doesn't mean CAD is not any good or you should not be using it. What it means is that you should read this [NEJM] paper carefully and understand how in general practice some radiologists are not using it correctly," said Sickles, breast imaging section chief at the University of California, San Francisco.

In particular, radiologists should read a mammogram and make a decision without CAD about whether they want to recall the patient, he said. If the findings are suspicious, they should not even use CAD, they should just recall the patient. If the image appears normal, CAD should be applied.

"If you read between the lines of the paper, the take-home message is clear: Thou shalt not use CAD to change your mind," Sickles said. "CAD is not designed for and should not be used in order to change an abnormal call into a normal call. CAD is not that good. It misses breast cancers."