Report from ARRS: CAD gives radiologists a slight boost in mammography

Computer-aided detection software results in just a slight improvement in breast cancer detection, and it misses a substantial number of malignancies, radiologists reported at the American Roentgen Ray Society meeting on Thursday.

In one prospective study, researchers at the University of California, San Francisco analyzed screening mammography performance before and after CAD at a high-screening-volume community hospital.

Screen-films were taken, then subsequently digitized and reviewed with R2's ImageChecker (both earlier and later versions of the product).

The researchers looked at 105,336 consecutive screening examinations performed from August 2004 to May 2005 and compared findings with 100,689 exams done prior to CAD.

Radiologists interpreted the exams, then applied the ImageChecker system. The researchers later assessed whether malignancies were found by the radiologist, CAD, or both.

Out of 105,336 screening exams, 341 cancers were identified, of which nine were detected by CAD only. Thus, CAD resulted in a 3% increase in cancer detection

"Had it not been for CAD, radiologists would have missed these nine cancers," said Dr. Jessica Leung, an assistant professor of radiology at UCSF.

Of the cancers seen by CAD alone, six (67%) were ductal carcinoma in situ, two were invasive ductal carcinoma, and one was a lobular carcinoma.

However, CAD missed 34 out of 341, or roughly 10%, of cancers that were found by radiologists.

"A substantial number are not marked by CAD. Radiologists need to be vigilant," Leung said.

She also noted that CAD led to a 12.1% increase in the recall rate, but that the higher rate is acceptable. There was no change in the number of cancers found relative to the number of screening exams: 3.2 per 1000 mammograms before and after CAD. There were also no statistically significant differences in the stage of cancers found.

In another prospective study, researchers at Tufts University assessed the value of an early version of the iCAD Mammo-Reader system in 5016 screening mammograms performed over the course of one year.

The radiologists involved in the study had more than 10 years' experience in mammography. They detected 42 cancers, and CAD alone detected an additional three malignancies. This 4.7% increase was not statistically significant, according to study authors.

The radiologists alone had a sensitivity of 90%, which increased to 94% with CAD. Specificity remained unchanged at 99% before and after CAD. There was no statistically significant increase in positive predictive value with CAD.

CAD missed eight cancers that were detected by the radiologists. The system performed well in microcalcifications but was not very good for low density masses and areas with architectural distortion.