Detection and Segmentation of Pectoral Muscle on MLO-View Mammogram Using Enhancement Filter

Abstract

The presence of predominant density region of the pectoral muscle in Medio-Lateral Oblique (MLO) view of the mammograms can affect or bias the results of mammograms processing for breast cancer detection using intensity based methods. Therefore, to improve the diagnostic performance of breast cancer detection using computer-aided system, identification and segmentation of pectoral muscle is an important task. This paper presents, an intensity based approach to identify the pectoral region in mammograms. In the presented approach enhancement mask and threshold technique is used to enhance and select the pectoral region and boundary points respectively, to find the boundary of pectoral muscle. Then curve fitting by Least Square Error (LSE) method is used to refine the rough initial boundaries. The proposed approach was applied on 320 mammograms from mini-Mammographic Image Analysis Society (mini-MIAS) database of 322 mammograms, with acceptable rate of 96.56% from radiologist experts. The performance evaluation for pectoral muscle segmentation, based on Hausdorff distance (H ), False Positive (FP) and False Negative (FN) rate, shows the usefulness and effectiveness of the proposed approach.

Breast pectoral muscle segmentation in mammograms using a modified holistically-nested edge detection network

“…Although there are many methods developed in the literature, the majority of them were evaluated qualitatively by expert radiologists (Kwok et al, 2004;Chen and Zwiggelaar, 2010) or have been evaluated using their own private datasets. It can be observed in Table 4 that our proposed method outperformed our previous method (Rampun et al, In terms of the false negative and false positive rates obtained with the MIAS dataset, our method produced F N R=3.2% and F P R=0.6%, respectively, which indicated that we quantitatively outperform recent studies (Vikhe and Thool, 2017;Chen et al, 2015;Yoon et al, 2016;Liu et al, 2014). Although the studies of Camilus et al (2010) and Ferrari et al (2004) reported small false positives, their proposed methods produced large false negatives of 5.58% and 5.77%, respectively.…”

“…Curve fitting-based techniques (Mustra and Grgic, 2013;Bora et al, 2016;Vikhe and Thool, 2017;Chen et al, 2015) have also been used as a part of the segmentation or post-processing step to estimate the pectoral muscle curve. Mustra and Grgic (2013) manually selected initial points for polynomial fitting to estimate the actual muscle boundary, which they assumed to be concave or convex.…”

“…Subsequently, a smooth pectoral boundary was obtained using Euclidean distance regression in conjunction with polynomial modelling. A similar approach was developed by Vikhe and Thool (2017) whose method used curve fitting by the Least Square Error (LSE) to refine the rough initial boundary points estimated via thresholding. Chen et al (2015) refined the initial boundary determined via shape-based region growing using a cubic polynomial function, whereas Yoon et al (2016) used quadratic curve fitting using the random sample consensus algorithm.…”

“…1. The majority of the studies (Kwok et al, 2004;Ferrari et al, 2004;Karssemeijer, 1998;Aylward et al, 1998;Chen and Zwiggelaar, 2010;Chakraborty et al, 2012;Rampun et al, 2017b;Taghanaki et al, 2017;Bora et al, 2016;Vikhe and Thool, 2017;Chen et al, 2015;Yoon et al, 2016) tried to manually model the curve structure of the pectoral muscle either using straight-line techniques, curve-based techniques or a combination of these two. In other words, the majority of the existing methods are hand-crafted models that require specific knowledge and were insufficient in dealing with the large variation of pectoral muscle boundaries.…”

“…One of the major limitations of the current methods in the literature (Kwok et al, 2004;Ferrari et al, 2004;Taghanaki et al, 2017;Bora et al, 2016;Vikhe and Thool, 2017;Chen et al, 2015) was their inability to deal with complex curves due to the geometrical architecture of the model. This can be seen in image InBreast1 where both of the methods of Taghanaki et al (2017) and Rampun et al (2017b) failed to estimate the pectoral boundary close to the ground truth, whereas our proposed method achievesJ = 89.8% andD = 94.6%.…”