Bridging the Gap between AI Performance and Clinical Practice: A Comprehensive Scoping Review of Breast Cancer Diagnosis

Abstract

Breast cancer remains a leading cause of mortality among women worldwide, necessitating early and accurate diagnosis. Recent advancements in artificial intelligence (AI), particularly machine learning and deep learning, have shown significant potential in improving diagnostic performance using medical imaging. This study presents a comprehensive scoping review conducted using a structured search and screening framework, through which 22 studies published since 2022 were systematically identified and included. The review focuses on AI-based breast cancer diagnosis employing machine learning, deep learning, and computer vision techniques. The key outcomes of this review indicate a clear transition from traditional machine learning methods to deep learning architectures, with Convolutional Neural Networks dominating current approaches, while emerging models such as Vision Transformers and multimodal frameworks are gaining traction. Additionally, although many studies report high diagnostic performance (accuracy and AUC >90%), these results are largely confined to controlled experimental settings, with limited generalizability to real-world clinical environments. The review further identifies critical challenges, including poor cross-dataset performance, limited model interpretability, and barriers to clinical integration. Emerging directions such as multimodal learning and explainable AI demonstrate strong potential to address these limitations but require further validation. Overall, this review emphasizes the need to move beyond performance-focused research toward the development of robust, interpretable, and clinically deployable AI systems to achieve meaningful real-world impact in breast cancer diagnosis.