Generative Artificial Intelligence Models in Breast Cancer Diagnosis: A Systematic Review

Background: This systematic review aims to evaluate and synthesize the current research on generative artificial intelligence (AI) models, focusing on their roles, performance, datasets, and evaluation metrics in improving breast cancer detection. Method: A systematic search using the PRISMA guideline was conducted. Four electronic databases (Science Direct, IEEE Explore, SpringerLink, and Scopus) and Google Scholar were searched using specific search strings for online articles published between 2020 and 2024. After applying specified inclusion and exclusion criteria, 91 articles were selected for data extraction, synthesis, and quality assessment. Results: This study reveals that generative AI models are used in breast cancer diagnosis for data augmentation, classification, and segmentation. For data augmentation, GAN has proven to be of highest effectiveness, while transformers have the highest effectiveness in terms of classification and segmentation. This review highlights the growing adoption of various publicly and privately available datasets, which have played a crucial role in training and validating these models. The findings showed the importance of selecting appropriate datasets and metrics to enhance the effectiveness of generative AI models in breast cancer diagnosis. A notable gap identified is the underrepresentation of African populations in the datasets used. Furthermore, the limited applicability of these models in real-world clinical settings is attributed to the lack of explainability. Conclusions: This review provides valuable insights into current practices, identifies existing gaps, and highlights areas for future research, aiming to improve the reliability and applicability of AI-based diagnosis in clinical settings.