From Insights to Interface: Exploring Human-AI Interaction in Clinical Decision-Making for Ophthalmology

Abstract

Despite the considerable potential inherent in the integration of AI into healthcare, its practical application remains limited. In a preceding study (Theilmann et al., 2025), semi-structured expert interviews were conducted to identify key factors for successfully integrating AI into healthcare. Factors identified include ease of use, alignment with clinical workflows, the incorporation of domain-specific knowledge and the involvement of stakeholders through co-design methods. This paper explores these factors in practice by implementing a low-fidelity prototype to support ophthalmologists in clinical decision-making based on optical coherence tomography (OCT) and fundus scans was implemented. It supports multimodal interaction modalities, editable AI-generated suggestions, and interactive visual overlays. To evaluate the user interface and interaction design, structured usability testing was carried out with practising ophthalmologists at a German ophthalmology clinic. The study employed a combination of quantitative and qualitative methodologies, encompassing think-aloud protocols, the System Usability Scale (SUS), and an A/B testing setup. The findings suggest that interaction design tailored to the specific needs of ophthalmology, such as visual overlays and multimodal interaction types, improves the efficiency of Human–AI collaboration. A strong preference for interpretable and editable AI outputs was identified, as these outputs allow for greater control over final decisions and increased transparency. The study outlines a human-centred design process and demonstrates how structured feedback loops, domain-specific adaptations and user-centred design can facilitate a more effective adoption of AI in healthcare. These insights could inform the development of future interactive AI systems that support, rather than replace, medical expertise.

Ähnliche Arbeiten