Personalized explanations for clinician-AI interaction in breast imaging diagnosis by adapting communication to expertise levels
Francisco Maria Calisto, João Abrantes, Carlos Santiago, Nuno Nunes, Jacinto C. Nascimento
Abstract
This paper investigates the impact of personalized AI communication on clinical outcomes in breast cancer diagnosis. Our study examines how different AI communication styles influence diagnostic performance, workload, and trust among clinicians, focusing on imaging diagnosis. We engaged 52 clinicians, categorized as interns, juniors, middles, and seniors, who diagnosed patient cases using conventional and assertiveness-based AI communication. Results show that personalized AI communication reduced diagnostic time by a factor of 1.38 for interns and juniors, and by a factor of 1.37 for middle and senior clinicians, without compromising accuracy. Interns and juniors reduced their diagnostic errors by 39.2% with a more authoritative agent, while middle and senior clinicians saw a 5.5% reduction with a more suggestive agent. Clinicians preferred assertiveness-based AI agents for their clarity and competence, valuing detailed, contextual explanations over numerical outputs. These findings underscore the need for adaptable AI communication to build trust, reduce cognitive load, and streamline clinical workflows. This work offers valuable insights for designing effective AI systems in high-stakes domains, contributing significantly to the Human–Computer Interaction community by enhancing our understanding of AI-mediated clinical support.