Measuring domain shift effect for deep learning in mammography

Abstract

[en] Breast cancer remains a global challenge, affecting over 2.3 million women in 2020 (refs WHO). The most common screening technology is mammography. The use of deep learning approaches such as Convolutional Neural Networks has recently shown promising results. However, these models are constrained by the limited size of publicly available mammography datasets. Moreover, these models are highly dependent on the quality of the provided training data. In this work, we will study the breast cancer classification problem by using Convolutional Neural Networks. We will show the effectiveness of Convolutional neural networks in breast cancer problems, and we will explore the domain shift problem by using different mammography datasets. Extensive validation will be presented to show the strengths and limitations of breast cancer classification.