Radiology: Mammo CAD, double reading each have advantages; neither infallible

Mammography computer-aided detection (CAD). Image Source: Definiens

Single mammography reading with computer-aided detection (CAD) and double reading yielded similar cancer detection rates; however, each reading regimen helped radiologists detect cancers missed by the other, according to the CADET II study published in the August issue of Radiology.

The U.K. relies on double reading of screening mammograms, which increases cancer detection rates by 4 percent to 13 percent, as standard practice. Because the model is more expensive than single reading and faces manpower limitations interest in CAD and in comparing CAD with double reading is on the rise in the U.K.

Designed to evaluate the mammographic features of breast cancer that favor lesion detection with single reading and CAD or double reading, the CADET II (Computer-Aided Detection Evaluation Trial) recruited 31,057 women from three screening centers between September 2006 and August 2007 and randomly assigned them to three groups: double reading, single reading with CAD or double reading and single reading with CAD at a ratio of 1:1:28, respectively.

Seventeen radiologists, two breast cancer clinicians and eight technologists trained to read film images served as readers in the study, which focused on primary outcome measures of cancer detection rates and recall rates in the 28,204 women whose images were single read with CAD and double read.

Researchers identified 227 cancers in women whose mammograms were single read with CAD and double read, and two breast radiologists retrospectively reviewed the screening mammograms. They recorded size, predominant radiographic feature and reading methods used to detect the cancer and reviewed CAD prompt data, said the study's lead author Jonathan J. James, MD, of Nottingham Breast Institute in England.

The ability of readers to correctly recall cancers in the basis of mammographic appearance varied significantly between reading regimens, wrote James. Double reading was more likely to recall cancers appearing as parenchymal deformities, while single reading with CAD showed a greater propensity to recall cancers appearing as asymmetric densities, which accounted for just three percent of lesions.

CAD correctly marked only 50 percent of the cancers that manifested as parenchymal deformities, wrote the authors. “Parenchymal deformity is the third most common manifestation of malignancy on mammograms, and it's one of the most challenging radiographic features for the reader to detect,” continued the authors.

"We know that [parenchymal deformity and asymmetric density] can be difficult features of cancers to appreciate on mammograms as they are also commonly the mammographic features of missed or interval cancers in a screening program. Cancers detected by double reading, but missed by a single reader using CAD were more likely to occur in women with a denser mammographic background pattern," noted James in an interview. Although double reading also can overlook cancers that manifest as parenchymal deformity, the authors posed that the double reading model offers the best prospect for detecting these cancers.

Both the single reading with CAD and double reading paradigms performed equally well at recalling patients with cancer with the predominant radiologic feature of either a mass or microcalcification, offered James, with CAD correctly marking 100 percent of microcalcifications.

“Readers using either single reading with CAD or double reading need to be aware of the strengths and weaknesses of reading regimens to avoid missing the more challenging cancer cases,” concluded James.