CAD boosts breast lesion detection by inexperienced readers

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon
 - breast cancer

Computer-aided diagnosis (CAD) systems improved inexperienced readers’ analyses of breast lesions in automated 3D breast ultrasounds but did little to better the interpretation of experienced readers, according to a study published in the November issue of Academic Radiology.

Tao Tan, MSc, of Radboud University Nijmegen Medical Centre, Nijmegen, the Netherlands, and colleagues examined the effectiveness of recently developed CAD systems in looking at malignant and benign breast lesions due to the demand for new screening technology. Breast cancer in dense breasts is incredibly difficult to determine, as reduced sensitivity often masks tumors from radiologists. CAD systems are designed to differentiate malignant breast lesions from benign ones, and include automated lesion segmentation in 3D.

Researchers organized an observer study in which the breast lesions of 88 patients were interpreted with and without CAD by 11 readers, seven of whom were radiologists and four of whom were residents. Forty-seven of the breast lesions were malignant and 41 were benign. Prior to the experiment, readers were trained by reading the lesions of 26 patients; 23 were benign and 14 were malignant. This time was used to acquaint the observers with the mechanics of the CAD system.

During the study, the cases were divided into two sets. The first part of the study asked readers to read the first set of data with CAD scores shown and the second part without. Readers were instructed to rate each lesion on a scale of 1 (benign) - 100 (malignant). The second session of the study maintained two sets of data but asked readers to read the first without CAD and the second with the system.  .

Receiver operating characteristic (ROC) was then utilized to analyze the performance of individual readers with and without CAD, as well as to compare reader performance with that of CAD on its own. Results indicated that the area under the ROC curve (AUC) for reader performances without CAD ranged from 0.81-0.95. Experienced readers who used CAD had an average AUC value of 0.89. An increase in AUC value from 0.85 to 0.90 was found for inexperienced readers. The difference in performance of readers with and without CAD was not significant for experienced readers. The difference was, however, significant for inexperienced readers.  The CAD’s AUC value on its own was 0.92.

Tan and colleagues remarked, “Our study confirms that there is a potential benefit of using CAD, though in our study, a significant improvement was obtained only for inexperienced readers. With CAD, they increased their performance to the level of the experienced readers. This might indicate that the benefits of a CAD system are greater for inexperienced readers than for experienced readers.”