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

Abstract

This paper presents a method for automatic breast pectoral muscle segmentation in mediolateral oblique mammograms using a Convolutional Neural Network (CNN) inspired by the Holistically-nested Edge Detection (HED) network. Most of the existing methods in the literature are based on handcrafted models such as straight-line, curve-based techniques or a combination of both. Unfortunately, such models are insufficient when dealing with complex shape variations of the pectoral muscle boundary and when the boundary is unclear due to overlapping breast tissue. To compensate for these issues, we propose a neural network framework that incorporates multi-scale and multi-level learning, capable of learning complex hierarchical features to resolve spatial ambiguity in estimating the pectoral muscle boundary. For this purpose, we modified the HED network architecture to specifically find 'contour-like' objects in mammograms. The proposed framework produced

Enhancing breast pectoral muscle segmentation performance by using skip connections in fully convolutional network

“…DCNN can automatically extract important features from the input images, which are best for pectoral muscle segmentation. Recently, Rampun et al presented a holistically nested edge detection network for the purpose of pectoral muscle segmentation from MLO view mammograms 25 …”

“…In the second stage, a Generative Adversarial Network is used for prediction of the shape of pectoral muscle region, which is used to learn the distribution of the dataset. Recently, DCNN architectures have been proposed 25–28 for pectoral muscle segmentation.…”

On Segmentation of Pectoral Muscle in Digital Mammograms by Means of Deep Learning

“…Subsequently, several studies aimed at designing the CNN architectures which were specifically tailored to the problem at hand. Among such works, the most promising results have been demonstrated in [17]- [19]. More specifically, in [17], [18], the CNNs were trained using image blocks (i.e., mini patches) extracted from the pectoral region, thus taking into account the structural appearance and photometric properties of the pectoral muscle.…”

“…Content may change prior to final publication. in [19] to subject the output of CNN-based classification to a post-processing stage involving morphological operations. It is worthwhile noting that, in its first stage, the method relied on a modified version of the hierarchical edge detection (HED) network of [20] which is capable of integrating the information on the location of pectoral edges across multiple resolution scales.…”

“…Motivated by the results of [19], the present study proposes a new approach to the problem of breast segmentation which, similarly to earlier works in the field, relies on a two-stage processing scheme. The first stage uses a modified version of the VGG16 network architecture [24].…”