Artificial intelligence in breast imaging: Areas of application from a clinical perspective

Abstract

Clinical/methodological issue: Central to breast imaging is the coordination of clinical and multimodal imaging information with percutaneous image-guided biopsies and surgical procedures. A wide range of problems arise due to this complexity: missed cancers, overdiagnosis, false-positive findings, unnecessary further imaging, biopsies and surgeries. Standard radiological methods: Breast imaging comprises the following diagnostic tests: mammography, tomosynthesis, contrast-enhanced mammography, (multiparametric) ultrasound, magnetic resonance imaging, computed tomography, nuclear medicine derived imaging and hybrid methods. Methodological innovations: Artificial intelligence (AI) promises to alleviate practically all these problems of breast imaging. AI has the potential to avoid missed cancers and false-positive findings. Furthermore, it could guide an efficient use of imaging methods and it may potentially be used to define biological phenotypes of breast cancer. Performance: AI-based software is being developed for various applications. Most developed are systems that support mammography screening. Problems are monocentric approaches and the focus on short-term financial success. Achievements: AI promises to improve breast imaging by simplifying and speeding up the workflow, by reducing monotonous tasks and by pointing out problems. This is likely to set free physician capacities that could be invested in improved communication with patients and interdisciplinary colleagues. Practical recommendations: The present article mainly addresses clinical needs in breast imaging, pointing out potential areas of use for artificial intelligence. Depending on the definition, a wide array of helpful software tools for breast imaging are already available. Global solutions, however, are still missing.