Original Jpn J Radiol Technol 2000 ;
Computerized Discrimination of Malignant and Benign
Microcalcification Clusters on Mammograms

RYOHEI NAKAYAMA, YOSHIKAZU UCHIYAMA, ISAMU HATSUKADE, KOJI YAMAMOTO,1) RYOJI WATANABE,2) KIYOSHI NAMBA,2) KAKUYA KITAGAWA,3) and KAN TAKEDA4)
Department of Computer Science and Systems Engineering, Miyazaki University
1)Medical Informatics Section, Mie University School of Medicine
2)Breastpia Namba Hospital
3)Department of Radiology, Matsuzaka Central Hospital
4)Department of Radiology, Mie University School of Medicine
Received Nov. 11, 1999; Revision accepted Jan. 6, 2000; Code No. 532
Summary
This paper introduces a classification of clustered microcalcifications that is based on the weighted-wavelet transform technique in digitized mammograms. The method uses three indicators of malignancy:(1)the standard deviation of the densities of individual microcalcifications within a cluster,(2)the coefficient of variation of their sizes within a cluster, and(3)the circularity of a cluster. The method was applied to the evaluation of malignancy in 62 microcalcification clusters selected as somewhat difficult cases from Breastpia Namba Hospital's patient files by an experienced mammographer. The results of the discriminant analysis using these indicators showed 85.3% sensitivity and 85.7% specificity.
Key words: Mammogram, Breast Cancer, Microcalcification, Diagnosis