?
Variability in Decision Making of Breast Cancer Radiologists and Commercial AI Models: Experimental Study
Breast cancer represents a major challenge in women's healthcare globally, underscoring the need for precise and dependable diagnostic techniques. Although substantial progress has been achieved in imaging technologies and physician training, significant variation exists in how radiologists interpret mammographic images. This study investigates discrepancies between diagnoses made by experienced radiologists and those generated by commercial artificial intelligence platforms. Through an experimental framework, we contrast outputs from AI systems against diagnoses rendered by skilled radiologists, paying particular attention to cases affected by cognitive biases. Our results indicate strong correlations between radiologists' decisions and AI outputs, raising few concerns about AI reliability. However, they highlight the necessity of careful implementation and monitoring when incorporating AI into clinical workflows, ensuring compliance with established diagnostic norms, and principles of informed risk aversion.