Deep Learning and Machine Learning in Mammography-Based Breast Cancer Detection: A Systematic Review of Advances, Challenges, and Future Directions

Abstract

Breast cancer is one of the most common causes of deaths caused by cancer in women, which requires early and precise detection of the disease in order to increase the chances of patient survival. One of the primary tools used to detect the disease is mammography, which, however, is prone to a number of limitations related to breast density, false positives, and inconsistencies in reader interpretations. The current systematic review provides an overview of the most recent achievements in the application of machine learning (ML) and deep learning (DL) algorithms to the problem of breast cancer detection via mammography. The paper reviews classical ML approaches, convolutional neural networks (CNNs), transformer-based approaches, self-supervised learning, generative models, multimodal learning, federated learning, and explainable artificial intelligence (XAI). In addition, the paper reviews publicly available mammography datasets, evaluation metrics, and major challenges in the field of mammogram-based detection of breast cancer, including class imbalance, dataset heterogeneity, domain shift, algorithmic biases, and lack of clinical validation.