Artificial intelligence in biomedical visualization and education: a narrative review

Biomedical visualization increasingly spans traditional medical illustration, synthetic medical imaging, vision-language systems, and patient-specific 3D reconstruction. This narrative review examines applications of artificial intelligence (AI) across this field, with particular attention to medical education and scientific and clinical communication. We searched PubMed/MEDLINE and IEEE Xplore for English-language literature published from January 2017 through September 2026, supplemented by targeted searches of publisher databases and reference-list screening. Included publications addressed AI-enabled medical or anatomical illustration, text-to-image generation, synthetic medical imaging, vision-language models, 3D reconstruction, or immersive biomedical visualization. Current evidence indicates that diffusion models can rapidly generate visually persuasive and customizable drafts, whereas segmentation-guided synthesis and anatomy-aware 3D methods offer stronger structural control in defined tasks. However, direct evaluations of general-purpose text-to-image systems consistently identify fabricated anatomy, incorrect labels, prompt- and model-dependent variability, demographic bias, and limited reliability for complex structures or procedures. Vision-language models can support prompt refinement, annotation, and consistency checking, but they cannot ensure scientific accuracy and may evaluate flawed AI outputs more favorably than human experts. Evidence supporting AI-generated educational illustrations remains less mature than that supporting conventional 3D and virtual-reality learning environments. We therefore distinguish established applications from emerging or prospective uses and propose an expert-supervised workflow incorporating documented prompts and model versions, structure-specific validation, manual correction, and transparent disclosure. AI should currently be treated as a drafting and visualization aid, with domain-expert verification required before educational, clinical, or scientific use.