Conceptual Model for Breast Cancer Diagnosis Using Machine Learning on Mammogram Data

Abstract

Breast cancer diagnosis from mammogram data has increasingly leveraged advanced machine learning (ML) techniques to enhance accuracy, reduce false positives/negatives, and support radiologists in clinical decision-making. This study focuses on developing a conceptual model for breast cancer diagnosis by integrating multi-view mammogram analysis with state-of-the-art ML algorithms. Modern approaches often emphasize deep learning (DL) architectures such as convolutional neural networks (CNNs), Vision Transformers (ViTs), and hybrid models, which combine local and global feature extraction for robust classification. In particular, multi-view methods, which analyze complementary information from craniocaudal (CC) and mediolateral oblique (MLO) views, are foundational to improving diagnostic accuracy. Transformers and attention-based mechanisms facilitate inter-view correlation learning, enhancing integration and interpretability. Meanwhile, weakly supervised techniques, such as Multiple Instance Learning (MIL), enable tumor localization and classification with limited annotated data. Addressing challenges related to imbalanced datasets and data scarcity, preprocessing methods (e.g., augmentation, GAN-based synthesis) and transfer learning have emerged as critical tools. Additionally, explainable AI (XAI) methods, such as Gradient-weighted Class Activation Mapping (Grad-CAM) and Shapley Additive Explanations (SHAP), improve clinical trust by aligning model outputs with radiological expertise. Despite advancements, obstacles such as dataset diversity, model generalizability, and architectural standardization remain. This study synthesizes key innovations in multi-view ML frameworks, weak supervision, and interpretability to propose a robust, conceptually integrated diagnostic model. The findings aim to bridge the gap between AI advancements and clinical applicability, offering a foundation for improving breast cancer screening outcomes. Further work is needed to standardize methodologies and validate models across diverse populations.

Topics & Concepts