Towards a Visual Pattern Language for Generative AI in Healthcare

Generative AI (GenAI) applications consist of several interconnected components that collaborate to generate meaningful outputs. To effectively communicate the architecture of an LLM application to software developers, it is crucial to adopt a standardized set of symbols that represent different components and their interactions. We propose establishing a unified set of visual annotations and guidelines to ensure consistency, transparency, and a shared understanding among all stakeholders, drawing from our experience in the healthcare domain. We will first identify recurring and reusable components from the systematic literature review of GenAI applications in healthcare. Next, we will categorize the components based on the problem space where they are applied. Further, we will define the potential connectors and interactions between these components. Finally, we will assign appropriate visual annotations to the set of components, connectors, and interactions. The VPL will be refined through collaborative discussion and consensus-building, considering the diverse perspectives and needs of all stakeholders using the Delphi method. We anticipate two outcomes from our research: a vocabulary of VPL that facilitates stakeholder communication and component reusability, and a methodology to establish such a consensus-based vocabulary.