Large Multimodal Model-Driven Clinical Decision Support Systems: Mapping Architectures to Decision Requirements—A Scoping Review Protocol

Project and scope: The rapid evolution of Large Multimodal Models (LMMs) has introduced a "technical velocity" that frequently outpaces the development of clinical governance frameworks. Current research often treats clinical tasks as static snapshots or one-off judgments, failing to account for the complex, longitudinal nature of medical reasoning. The primary purpose of this research is to bridge the interdisciplinary void between advanced AI architectures and real-world clinical deployment. By employing a decision-centric framework, this study moves beyond treating medical specialties as simple evaluation categories. Instead, it analyzes them as active governors of system design, where specific clinical constraints, such as inferential depth, regulatory requirements (e.g., EU AI Act), and longitudinal reasoning trajectories, force specific architectural evolutions. The project is strictly delimited to the role of the Digital Practitioner, focusing on licensed healthcare professionals within professional workflows to ensure high precision and clinical relevance. Methodology: This study utilizes a JBI-compliant scoping review design, enhanced by a hybrid screening pipeline: Human Expertise: Two independent reviewers conduct contextual screening of titles and abstracts to ensure clinical relevance. AI-Driven Deep Audit: An AI system performs full-text analysis of the documents, completing a technical checklist to identify specific architectural responses such as Multi-Agent Systems (MAS), Retrieval-Augmented Generation (RAG), and Explainable AI (XAI). Quantitative Synthesis: Data will be ingested into a Python-based pipeline to calculate inter-rater reliability and generate multidimensional evidence maps. Expected Outcomes: The anticipated outcomes of this research project include: A Formal Taxonomy: The construction of a taxonomy of clinical decision space archetypes that reflects the structural specifications of diverse medical tasks. Architectural Mapping: A clear mapping of how high-complexity clinical requirements necessitate specific technical paradigms, such as MAS for integrative breadth or temporal modules for longitudinal reasoning. Deployment Guidelines: An evaluation of the determinants that influence the transition from experimental prototypes to clinical-grade implementations capable of maintaining logical fidelity and trustworthiness. Evidence Base for the TUAI Network: A zero-redundancy baseline that informs the development of future AI systems designed to align with the professional needs of the Digital Practitioner.