AI in breast imaging: A systematic review of reader studies evaluating design, performance, and clinical impact

Abstract

Artificial intelligence (AI) is rapidly transforming breast imaging by supporting radiologists in cancer detection and diagnostic decision-making. Although many AI algorithms show strong technical performance in research, their real-world clinical utility remains under active investigation. Reader studies-where radiologists interpret imaging cases with and without AI assistance-offer a robust framework for evaluating the impact of AI systems in realistic diagnostic workflows. This systematic review, conducted in accordance with PRISMA 2020 guidelines, involved a literature search across PubMed, Web of Science, IEEE Xplore, and Google Scholar for studies published from January 2010 to February 2025. A total of 267 records were identified prior to screening, and 47 eligible reader performance studies were included in the final analysis. We focus on reader studies that directly compare radiologists' performance with versus without AI support across multiple breast imaging modalities, including mammography, ultrasound, and magnetic resonance imaging (MRI). The AI tools assessed span both computer-aided detection (CADe) and computer-aided diagnosis (CADx) systems. Compared to unaided interpretation, AI-assisted reading generally improves radiologists' performance and efficiency across a range of tasks, including detection and diagnosis. However, the degree of benefit varies across studies and is influenced by task type, reader experience, and the mode of AI integration. Despite encouraging findings, challenges remain, including the design of reader studies, validation through prospective trials, and the integration of human-AI collaboration strategies. We highlight existing evidence gaps and ou tline directions for future reader studies to support the clinical use of AI in breast imaging.

Ähnliche Arbeiten