Detection of Breast Mass in Mammogram Using Computer-Aided Diagnosis System

Purpose: Computer-aided diagnosis system was developed to improve the accuracy and the efficacy of the image interpretation. This article is to provide a possibility of computeraided diagnosis for detection of masses in mammograms. Methods: The craniocaudal and mediolateral images of 120 mammograms from 30 patients that were histologically proven to be malignant and 30 patients that were histologically proven to be benign were analysed using the mammography softwere. The contralateral mammograms were used as control images. Correct marks of the lesions were scored as a true positive and marks not at the location of the lesions were scored as a false negative. Any marks of the normal images were scored as a false positive and no mark of normal images were scored as a true negative. Results: It took approximately 2 min to scan and 1 min to process 24 by 18-cm mammograms. There was an average of 1.4, 2.0 and 2.1 marks per image in normal, benign and malignant mammograms respectively. Mass detection rate of malignant lesion was 90.0% (27 of 30) and that of benign lesion was 63.6% (21 of 33). Mass detection rate of dense breasts was 68.8% (22 of 32) and that of fatty breasts was 83.9% (26 of 31). Mass detection rate of BI-RADS category 4, 5 and 0 was 85.7% (42 of 49) and that of category 1, 2 and 3 was 42.9% (6 of 14). The overall sensitivity was 76.2% and specificity was 28.1%. Conclusion: In this study, mass detection rate for malignant lesions was higher than that of benign lesions and dense breast has lower detection rate than fatty breast. According to the BI-RADS category, mass detection rate was higher in the more malignant category. Computer-aided diagnosis system for this study had limited specificity but acceptable sensitivity. (Journal of Korean Breast Cancer Society 2003;6:20-23) ꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏꠏ