Mammographic Breast Positioning Assessment via Deep Learning

TL;DR

This work tackles automated assessment of mammographic breast positioning quality in MLO views using a landmark-based deep learning pipeline. It introduces a CoordConv-enabled U‑Net with attention (CoordAtt UNet) to detect nipple and pectoralis landmarks and to draw a perpendicular posterior nipple line (PNL) for quantitative quality evaluation. On the VinDr Mammography dataset, CoordAtt UNet achieves top landmark localization (mean errors of a few millimeters) and strong classification performance ($88.63\% \pm 2.84$, $90.25\% \pm 4.04$, $86.04\% \pm 3.41$) with an angular error of $2.42^{\circ}$, outperforming regression and pure classification baselines. The study provides an open-source, automated, explainable tool to improve mammography screening reliability and reduce recall rates.

Abstract

Breast cancer remains a leading cause of cancer-related deaths among women worldwide, with mammography screening as the most effective method for the early detection. Ensuring proper positioning in mammography is critical, as poor positioning can lead to diagnostic errors, increased patient stress, and higher costs due to recalls. Despite advancements in deep learning (DL) for breast cancer diagnostics, limited focus has been given to evaluating mammography positioning. This paper introduces a novel DL methodology to quantitatively assess mammogram positioning quality, specifically in mediolateral oblique (MLO) views using attention and coordinate convolution modules. Our method identifies key anatomical landmarks, such as the nipple and pectoralis muscle, and automatically draws a posterior nipple line (PNL), offering robust and inherently explainable alternative to well-known classification and regression-based approaches. We compare the performance of proposed methodology with various regression and classification-based models. The CoordAtt UNet model achieved the highest accuracy of 88.63% $\pm$ 2.84 and specificity of 90.25% $\pm$ 4.04, along with a noteworthy sensitivity of 86.04% $\pm$ 3.41. In landmark detection, the same model also recorded the lowest mean errors in key anatomical points and the smallest angular error of 2.42 degrees. Our results indicate that models incorporating attention mechanisms and CoordConv module increase the accuracy in classifying breast positioning quality and detecting anatomical landmarks. Furthermore, we make the labels and source codes available to the community to initiate an open research area for mammography, accessible at https://github.com/tanyelai/deep-breast-positioning.

Figures (2)

• Figure 1: Illustration of concepts utilized in this study as part of an ablation study. At the input layer, a single-channel grayscale mammogram is augmented to a three-channel image by introducing two additional channels that encode the X and Y spatial coordinates of each pixel. The attention mechanism refines features, and skip connections preserve spatial information. The final layer outputs landmark coordinates, optimized using landmark-aware wing loss.
• Figure 2: Comparison of predicted (red) versus original (blue) landmarks in mammograms using different models: UNet, Attention UNet, CoordAtt UNet, R-ResNeXt50, and ResNeXt50. The rightmost column shows heatmaps for ResNeXt50.