의사-환자 의사소통 향상 및 환자의 합리적인 의사결정을 위한 Explainable AI(XAI) 기반의 진료보조 서비스 시스템 제안: 치과 진료 사례를 중심으로

초록

Background Dental care often relies on image-based diagnosis, specialized terminology, and limited consultation time, which often hinders patient understanding and engagement. This environment reinforces one-way, clinician-centered explanations and information asymmetry. Existing artificial intelligence (AI) applications have focused primarily on improving diagnostic accuracy, with limited impact on personalized explanations and patient–clinician communication. This study proposes an explanatory AI collaboration model that integrates human-centered AI (HCAI), explainable AI (XAI), and human-in-the-loop (HITL) approaches, tailored for dental care. Methods To define the human–AI collaboration structure, we conducted theoretical research, in- depth interviews, and user observations to derive personas and user journeys. Collaboration opportunity factors were then translated into scenarios. Design directions were refined through Likert scale surveys and in-depth interviews with clinicians and patients, followed by prototype development and user testing to propose an integrated service process. Results The proposed system comprises four core services—X-ray Explanation, MyChart Summary, Personalized Treatment Simulation, and Intelligent Post-Care Chatbot—designed as a cyclical collaboration model linking clinician judgment, AI interpretation, and patient engagement. Usability evaluations showed high satisfaction among both patients and clinicians. Qualitative analysis revealed that clinicians emphasized explanation reliability, personalization, and enhanced visualization/summary functions, while patients requested clear indication of expert involvement and improved chatbot interactivity. These insights informed design improvements balancing trust and patient autonomy. Conclusions This study redefines explainable AI as an “explanation partner” through an HCAI– XAI–HITL framework, presenting a collaboration structure that safeguards both patient autonomy and clinician decision authority. AI intervention levels were structured into four stages—Automate, Enhance, Amplify, and Augment—linking explanation depth with collaboration modes to clarify the relationship between autonomy and explanation levels. The model expands the theoretical applicability of explainable AI in healthcare and provides a practical blueprint covering pre-, intra-, and post-consultation phases for AI system design and evaluation. However, as the study is limited to dental care, future work should address scalability to other medical fields, cultural adaptability, and legal/ethical considerations in cases of AI explanation failure.

키워드