Recent Advances in an Efficient Cascaded Deep Capsule Cell Attention Network Model for Breast Cancer Molecular Subtypes Prediction Using Mammogram Images: A Systematic Review

Abstract

Breast cancer remains one of the leading causes of mortality among women worldwide, making early detection and accurate molecular subtype classification essential for effective treatment planning and improved patient outcomes. Molecular subtypes such as Luminal A, Luminal B, HER2-enriched, and triple-negative breast cancer play a significant role in determining prognosis and therapeutic strategies. Traditional subtype classification methods primarily depend on invasive biopsy procedures and immunohistochemical analysis, which are costly, time-consuming, and resource-intensive. Recent advancements in deep learning techniques, including convolutional neural networks (CNNs), capsule networks, and attention mechanisms, have shown significant potential in automating breast cancer diagnosis using mammogram images. This systematic review explores recent developments in cascaded deep capsule cell attention network models for predicting breast cancer molecular subtypes from mammographic data. Capsule networks preserve spatial and hierarchical relationships among features, while attention mechanisms allow models to focus on critical tumor regions for enhanced classification accuracy. Cascaded architectures further improve feature extraction and diagnostic performance. The findings indicate that hybrid capsule-attention models outperform conventional CNN-based approaches in accuracy, sensitivity, robustness, and reliability. These intelligent systems offer a promising non-invasive alternative to biopsy-based diagnosis and support personalized treatment planning. However, challenges such as limited annotated datasets, class imbalance, and lack of interpretability continue to hinder large-scale clinical implementation and real-world applicability.