Eldesoqui Mamdouh, Albadawi Emad A, AlQumaizi Khalid I, Radwan Maryam Nizar Mohammad, Ebrahim Hasnaa Ali, Elsaid Manar Abd Elaziz
Department of Basic Medical Sciences, College of Medicine, AlMaarefa University, Riyadh, Saudi Arabia.
Department of Human Anatomy and Embryology, Faculty of Medicine, Mansoura University, Mansoura, Egypt.
Clin Anat. 2025 Jul 9. doi: 10.1002/ca.70002.
Historically, human anatomy education has been an essential part of medical training, depending on cadaveric dissection and anatomical representations. However, financial and ethical limitations have resulted in a decline in conventional teaching techniques, necessitating the investigation of alternative resources such as digital drawings and artificial intelligence (AI). The aim of this research was to assess and compare the anatomical precision of graphics produced by four AI text-to-image generators: Microsoft Bing, DeepAI, Freepik, and Gemini, emphasizing their value in medical education. On February 6, 2025, four AI text-to-image generators were used. Prompts for creating intricate anatomical images included the human heart, brain, skeletal thorax, and hand bones. Two anatomists and a radiologist evaluated the pictures produced according to anatomical standards. Bing and Gemini generated anatomically correct representations of the human heart, but DeepAI and Freepik were less accurate. All generators offered accurate reconstructions of the human brain; however, there were disparities in sulci and gyri, with Gemini performing best. Only Gemini delivered a correct sternum; the other generators misrepresented the rib count. The Gemini platform provided a satisfactory depiction of the human hand skeleton, but the outputs from other text-to-image generators were not anatomically accurate. This work examines the potential of generative AI in medical illustration, noting significant limitations in accuracy and detail, especially with bony structures. Although AI accelerates the drawing process, it cannot replace the proficiency of skilled medical illustrators. Continuous assessment and improvement of AI-generated material are essential to ensure that the criteria mandated for medical education are met.
从历史上看，人体解剖学教育一直是医学培训的重要组成部分，依赖于尸体解剖和解剖学图示。然而，经济和伦理限制导致传统教学技术有所下降，因此有必要研究数字绘图和人工智能（AI）等替代资源。本研究的目的是评估和比较由四个AI文本到图像生成器（Microsoft Bing、DeepAI、Freepik和Gemini）生成的图形的解剖学精度，强调它们在医学教育中的价值。2025年2月6日，使用了四个AI文本到图像生成器。创建复杂解剖图像的提示包括人类心脏、大脑、胸廓和手部骨骼。两名解剖学家和一名放射科医生根据解剖学标准对生成的图片进行了评估。Bing和Gemini生成了人体心脏的解剖学正确表示，但DeepAI和Freepik的准确性较低。所有生成器都提供了人类大脑的准确重建；然而，脑沟和脑回存在差异，Gemini表现最佳。只有Gemini生成了正确的胸骨；其他生成器对肋骨数量的表示有误。Gemini平台对人类手部骨骼的描绘令人满意，但其他文本到图像生成器的输出在解剖学上并不准确。这项工作研究了生成式AI在医学插图中的潜力，指出了在准确性和细节方面的重大局限性，尤其是在骨骼结构方面。尽管AI加速了绘图过程，但它无法取代熟练医学插画家的专业技能。对AI生成的材料进行持续评估和改进对于确保符合医学教育要求的标准至关重要。
