Automatic identification and extraction of pectoral muscle in digital mammography

Abstract

Mammography is a worldwide image modality used to diagnose breast cancer and can be used to measure breast density (BD). In clinical routine, radiologist perform image evaluations through BIRADS assessment. However, this method has inter and intraindividual variability. An automatic method to measure BD could relieve radiologist’s workload by providing a first aid opinion. However, pectoral muscle (PM) is a high density tissue, with the same imaging characteristics as fibroglandular tissues, which makes its automatic detection a challenging task. The aim of this work is to develop an automatic algorithm to segment and extract PM in digital mammograms. A hybrid methodology has been developed using Hough transform, to find the edge of the PM, and active contour, to segment PM muscle. Seed of active contour is applied automatically in the edge of PM found by Hough transform. An experienced radiologist manually performed the PM segmentation. Manual and automatic methods were compared using the Jaccard index and Bland-Altman statistics. The comparison between methods presented a Jaccard similarity coefficient greater than 90% for all analyzed images. The Bland-Altman statistics compared the segmented PM area and showed agreement between both methods within 95% confidence interval. The method proved to be accurate and robust, segmenting rapid and free of intra and inter-observer variability.