A Short Review: Semantic Segmentation for Breast Cancer Detection in MRI Images

Automated breast cancer detection is of paramount importance as it could potentially reduce false-negative examinations and allow for earlier treatments, which would have a significant impact on patients and healthcare system costs. However, insufficient data, imbalanced classes, the detection of cancer in dense breasts, and the difficulty in distinguishing breast lymph nodes hinder the application of automated methods. To address these problems, in this paper, we presented an overview of deep learning models for the segmentation of breast cancer at an early stage by examining the contrast-enhanced MRI images. We examined the effectiveness of supervised machine learning models (UNet, VNet, SegResNet, and HighResNet) in the detection of breast cancer. We used 3D and 2.5D semantic segmentation models for pixel-level segmentation, which overcame the drawbacks of region-based segmentation. All the models are trained, validated, and tested on a private dataset collected from the Champalimaud Foundation, Portugal, and evaluated quantitatively and qualitatively.