Thesis abstract:

«p»Artificial Intelligence (AI) has played a crucial role in advancing the medical imaging field, particularly in breast imaging. Previous studies have demonstrated that AI's diagnostic performance is not inferior to that of average human readers. It has also been confirmed in previous studies that with AI assistance, the diagnostic performance of radiologists may significantly improve. However, while various studies have assessed the performance of AI across different populations, breast densities, and vendors, there remains a gap in knowledge regarding the performance of AI in perceiving cases with various levels of difficulty and in different groups (symptomatic vs asymptomatic women). This study aims to bridge these gaps by exploring how different AI systems cope with various test-set difficulties and whether there is a difference in performance given different dataset natures.«/p»