Automated registration and classification techniques for interval change analysis in mammography

Our goal is to develop a computer-aided diagnosis (CAD) system for analysis of interval change of lesions (masses and microcalcification clusters) on serial mammographic exams. The CAD system includes two major parts: automated registration of corresponding lesions on temporal pairs of mammograms, and classification of the lesion as malignant or benign using interval change information. This study also evaluated the effect of the CAD system on the radiologists' performance for characterization of mammographic masses on serial mammograms. Using our region registration method (RRM) the average distance between the estimated and the true lesion centroids on the prior mammogram was 6.4 /spl plusmn/ 8.9 mm. The classification accuracy was measured using the area A/sub z/ under the receiver operating characteristic (ROC) curve. The temporal information classifier achieved a test A/sub z/ of 0.87. The radiologists reading with CAD statistically significantly improved (p = 0.005) their performance (A/sub z/ = 0.84) compared to the reading without CAD (A/sub z/ = 0.79).