Deep Learning driven Multi-Modal Breast Cancer Detection for Early and Accurate Daigonosis

Abstract

Breast cancer remains one of the leading causes of mortality among women worldwide, making early and reliable detection a clinical priority. Mammography continues to serve as the primary screening modality however, the subtle appearance of microcalcifications, dense tissue patterns, and variability in lesion morphology pose significant challenges for traditional computer-aided diagnosis systems. This review systematically examines recent advancements in segmentation and classification techniques for microcalcification detection, with a particular emphasis on the transition from classical image-processing and machine-learning approaches to modern deep-learning architectures, including convolutional neural networks (CNNs), vision transformers (ViTs), and hybrid feature-fusion models. The analysis highlights methodological strengths, dataset characteristics, evaluation strategies, and performance limitations reported in the literature. Furthermore, the review identifies persistent barriers such as limited annotated datasets, inter-patient heterogeneity, and reduced robustness in dense breast categories. By synthesizing current progress and gaps, this study underscores the growing potential of multimodal and hybrid deep-learning frameworks to enhance diagnostic accuracy, improve clinical decision support, and pave the way for more interpretable and generalizable breast cancer detection systems.