Interpretability of Machine Learning Models for Breast Cancer Identification: A Review

Over recent years, machine learning models have enhanced breast cancer detection, especially in its early stages. Nevertheless, their integration into clinical practices remains limited despite their proven efficacy in early-stage detection among women. However, the results obtained by these approaches are poorly interpretable. This study seeks to demystify early-stage breast cancer detection and boost clinicians' trust in these methods by leveraging eXplainable Artificial Intelligence (XAI). This research underscores the potential of XAI as a foundational step to initiate conversations about adopting supportive AI tools in the clinical sphere. By incorporating these XAI methods, clinicians can better understand why a specific prediction has been made, promoting trust and facilitating more informed decision-making in breast cancer detection and treatment. This study uniquely investigates the potential of advanced XAI techniques to enhance the trustworthiness and reliability of machine learning models, specifically in the early detection and diagnosis of breast cancer. The different XAI approaches are critically reviewed, underlying the current limitations and proposing future work directions.