Dense-breast classification using image similarity

Abstract

This paper describes the auto-analysis of the mammary gland visualized on mammography images to eliminate the subjective evaluation error between physicians using pixel values and image similarity, including pattern recognition. The mammography images including the heterogeneously dense and extremely dense images were divided into two groups based on the result of the subjective breast classification as the dense breast, and non-dense breast. One hundred and thirty images obtained during screening were set as search images, and 101 evaluation images were classified using zero-mean normalized cross-correlation. Concerning the conventional method, we employed the variance histogram analysis method of Yamazaki et al. The concordance rate for the subjective breast classification result obtained using the conventional and proposed methods was 79.2% and 89.1%. The image similarity evaluation method, which analyzes the pattern of the pixel values, enabled the breast classification while eliminating ambiguity in the subjective breast classifications among physicians.

Keywords: Auto analysis; Dense-breast classification; Mammogram; NCC; Similarity; Template matching.