An Adaptive Immersive AI Framework for Clinical Training and Performance Evaluation Using Graph-Augmented Large Language Models

Healthcare education faces critical challenges including expanding medical knowledge, evolving clinical guidelines, and limited supervised clinical experience. While artificial intelligence (AI) and immersive technologies offer promising solutions, current systems operate in isolation rather than synergistically. This paper presents a systematic review of 165 AI-enabled clinical training systems (2022-2025), examining five technological dimensions: LLM-based virtual patients, VR/AR platforms, graph-augmented AI, adaptive learning, and performance evaluation. Analysis reveals significant fragmentation—67% of systems employ single approaches, only 7% integrate three or more technologies. Critical gaps include: factual hallucinations in LLMs (12-15% error rates), limited adaptability in VR/AR scenarios, single-dimension personalization, and domain-specific assessments. We propose an integrated architectural framework combining graph-augmented LLMs with immersive interfaces, multi-dimensional adaptive learning, and comprehensive performance evaluation. Graph augmentation demonstrates 73% reduction in factual errors; VR/AR systems show 25-40% skill retention improvements. The modular framework addresses identified gaps through bidirectional data flows and evidence-based component integration, providing a research-informed blueprint for next-generation clinical training systems.