Breast cancer is the most common forms of cancer and a leading cause of mortality in women. Early and correct diagnosis is, therefore, essential to save lives. The development of diagnostic imaging applied to the breast has been impressive in recent years and the most used diagnostic test in the world is mammography, a low-dose X-ray technique used for imaging the breast. In the first half of the 20th century, the diagnosis was in practice only clinical, with consequent diagnostic delay and an unfavorable prognosis in the short term. The rise of organized mammography screening has led to a remarkable reduction in mortality through the early detection of breast malignancies. This historical review aims to offer a complete panorama of the development of mammography and breast imaging during the last century. Through this study, we want to understand the foundations of the pillar of radiology applied to the breast through to the most modern applications such as contrast-enhanced mammography (CEM), artificial intelligence, and radiomics. Understanding the history of the development of diagnostic imaging applied to the breast can help us understand how to better direct our efforts toward an increasingly personalized and effective diagnostic approach. The ultimate goal of imaging applied to the detection of breast malignancies should be to reduce mortality from this type of disease as much as possible. With this paper, we want to provide detailed documentation of the main steps in the evolution of breast imaging for the diagnosis of breast neoplasms; we also want to open up new scenarios where the possible current and future applications of imaging are aimed at being more precise and personalized.

Artificial intelligence (AI) is increasingly integrated into breast imaging workflows, offering the potential to enhance diagnostic accuracy, efficiency, and early cancer detection. Image interpretation plays a pivotal role in the breast cancer diagnostic pathway, directly influencing therapeutic decisions and patient outcomes. However, the effective implementation of AI-assisted systems relies not only on technical performance but also on radiologists’ trust, acceptance, and readiness to incorporate these tools into clinical practice. In addition, system-related, perceptual, and cognitive factors may contribute to diagnostic errors, ultimately affecting overall accuracy and reliability. This paper provides a comprehensive overview of the cognitive and systemic sources of diagnostic inaccuracies in breast imaging, emphasizing the growing role of AI as both a supportive and potentially bias-modulating tool. Recent prospective studies have demonstrated the clinical safety and effectiveness of AI-assisted mammography screening, reporting improved cancer detection rates and reduced workload. Nonetheless, the integration of AI into diagnostic workflows without an appropriate knowledge of the consequences may introduce new cognitive biases, such as anchoring, automation, and confirmation bias, that influence radiologists’ decision-making and counteract the intended benefits. To address these challenges, the paper outlines strategies to mitigate diagnostic errors and foster appropriate integration of AI into clinical practice. These include targeted training programs, enhanced interdisciplinary communication, and standardized interpretation workflows that promote consistent evidence-based practice. Furthermore, the adoption of explainable AI approaches is identified as a key factor in improving model transparency and interpretability, allowing radiologists to understand algorithmic reasoning and engage in a more informed, confidence-based human-AI collaboration. Ultimately, a balanced and context-sensitive integration of AI, grounded in continuous professional education and cognitive awareness, is essential for improving diagnostic accuracy while preserving radiologists’ critical analytical skills.

Teaching point: Extrapelvic endometriosis involving pudendal and sciatic nerve may be a cause of lower limb pain.

Objectives: This study evaluates the impact of neoadjuvant chemotherapy (NAT) on breast tissue density and examines the correlation between density changes and pathological response, while also assessing inter-reader agreement among radiologists and artificial intelligence (AI) tools. Methods: The study included 135 women with triple-negative and/or HER2+ invasive ductal breast cancer who underwent NAT. Digital mammography was performed pre- and post-NAT, with evaluations by radiologists of varying experience levels and comparison using an in-house developed AI tool based on AlexNet CNN architecture. Statistical analyses included Wilcoxon signed-rank test, linear regression, McNemar test, Spearman’s rank correlation, and intraclass correlation coefficients (ICC). Results: Post-NAT assessments showed a significant decrease in breast tissue density, with initial heterogeneously dense and extremely dense breasts decreasing from 32% and 15% to 18% and 7%, respectively (P < .003). Significant relationships were found between density reduction and age, menopausal status, and baseline density, and the reduction correlated moderately with pathological response to NAT. Intraclass correlation coefficient values reflected good agreement among experienced radiologists (0.68), fair agreement between expert and resident (0.55), poor agreement between expert and non-radiologist (0.39), and substantial agreement between the expert and AI tool (0.78). Conclusions: Neoadjuvant chemotherapy significantly reduces breast tissue density, correlating with pathological response, and AI tools provide consistent density evaluations. Artificial intelligence tools show promise in improving assessment consistency, with substantial agreement with expert radiologists. The findings highlight the need for standardized protocols and training to optimize breast density evaluations and suggest further refinement and validation of AI tools for clinical use. Advances in knowledge: This study evaluates the impact of NAT on breast tissue density while investigating its correlation with pathological response and inter-reader agreement among radiologists and AI tools. It contributes to the field by demonstrating the potential of AI to improve consistency in density evaluations and emphasizing the importance of standardized protocols and training for optimized clinical assessments.