生成式AI在医学可视化教学中的应用与局限性：基于双案例的探索性比较分析

Application and Limitations of Generative AI in Medical Visualization Education: An Exploratory Comparative Analysis Based on Two Cases

摘要: 生成式人工智能(GAI)正加速渗透医学可视化教学，但其图像的高似真性与低准确性构成显著悖论。本研究选取医学机制图(纳米药物递送与肿瘤免疫调控)与人体解剖结构图(臀区梨状肌与坐骨神经关系)两类典型图像，采用视觉内容分析与比较分析方法，系统评估GAI在医学可视化教学中的应用边界。结果表明，GAI在整体构图、视觉风格统一与初稿快速生成方面具有明显效率优势，但在因果时序逻辑、空间层次关系及动态过程呈现等核心医学知识编码能力上存在系统性缺陷，且该缺陷跨场景一致，提示其源于模型对医学逻辑理解的根本不足。基于此，本文提出医学可视化教学应实现从技能训练向认知培养的目标转向，构建“AI生成–专家校正–课堂反馈”的闭环机制，将AI图像作为批判性审图素材，而非标准化教学成果。研究为AI时代医学可视化课程改革提供了实证依据与教学路径参考。

Abstract: Generative artificial intelligence (GAI) is rapidly penetrating medical visualization education, yet its generated images exhibit a notable paradox of high visual plausibility and low factual accuracy. This study selected two representative image types—a mechanistic diagram (nanomedicine delivery and tumor immune modulation) and an anatomical diagram (relationship between piriformis muscle and sciatic nerve)—and employed visual content analysis and comparative analysis to systematically evaluate GAI’s application boundaries in medical visualization teaching. Results show that GAI excels in overall composition, visual style consistency, and rapid draft generation, but exhibits systematic deficiencies in encoding causal-temporal logic, spatial hierarchical relationships, and dynamic processes. These defects persist across both cases, suggesting they stem from the model’s fundamental inability to comprehend medical logic. Accordingly, this paper argues that medical visualization education should shift its focus from skill training to cognitive development, and establish a closed-loop mechanism of “AI generation-expert correction-classroom feedback,” using AI-generated images as materials for critical review rather than as standardized teaching outputs. This study provides empirical evidence and pedagogical guidance for curriculum reform in the AI era.