BACKGROUND The increasing accessibility of artificial intelligence (AI) text-to-image generators offers a novel avenue for exploring societal perceptions. The present study assessed AI-generated images to examine the representation of gender and racial diversity among orthopaedic surgeons. METHODS Five prominent text-to-image generators (DALL·E 2, Runway, Midjourney, ImagineAI, and JasperArt) were utilized to create images for the search queries "Orthopedic Surgeon," "Orthopedic Surgeon's Face," and "Portrait of an Orthopedic Surgeon." Each query produced 80 images, resulting in a total of 240 images per generator. Two independent reviewers categorized race, sex, and age in each image, with a third reviewer resolving discrepancies. Images with incomplete or multiple faces were excluded. The demographic proportions (sex, race, and age) of the AI-generated images were then compared with those of the 2018 American Academy of Orthopaedic Surgeons (AAOS) census. RESULTS In our examination across all AI platforms, 82.8% of the images depicted surgeons as White, 12.3% as Asian, 4.1% as Black, and 0.75% as other; 94.5% of images were men; and a majority (64.4%) appeared ≥50 years old. DALL·E 2 exhibited significantly increased diversity in representation of both women and non-White surgeons compared with the AAOS census, whereas Midjourney, Runway, and ImagineAI exhibited significantly decreased representation. CONCLUSIONS The present study highlighted distortions in AI portrayal of orthopaedic surgeon diversity, influencing public perceptions and potentially reinforcing disparities. DALL·E 2 and JasperArt show encouraging diversity, but limitations persist in other generators. Future research should explore strategies for improving AI to promote a more inclusive and accurate representation of the evolving demographics of orthopaedic surgery, mitigating biases related to race and gender. CLINICAL RELEVANCE This study is clinically relevant as it investigates the accuracy of AI-generated images in depicting diversity among orthopaedic surgeons. The findings reveal significant discrepancies in representation by race and gender, which could impact societal perceptions and exacerbate existing disparities in health care.

中文翻译：

---

骨科手术中的人工智能描绘：使用文本到图像生成器分析性别和种族多样性。


背景人工智能 （AI） 文本到图像生成器的日益普及为探索社会观念提供了一条新的途径。本研究评估了 AI 生成的图像，以检查骨科医生中性别和种族多样性的代表性。方法 五个著名的文本到图像生成器 （DALL·E 2、Runway、Midjourney、ImagineAI 和 JasperArt）用于为搜索词“Orthopedic Surgeon”、“Orthopedic Surgeon's Face”和“Portrait of an Orthopedic Surgeon”创建图像。每个查询生成 80 张图像，因此每个生成器总共有 240 张图像。两名独立审阅者对每张图像中的种族、性别和年龄进行分类，第三名审阅者解决差异。排除了不完整或多个人脸的图像。然后将 AI 生成图像的人口比例（性别、种族和年龄）与 2018 年美国骨科医师学会 （AAOS） 人口普查的人口比例进行比较。结果在我们对所有 AI 平台的检查中，82.8% 的图像将外科医生描绘为白人，12.3% 为亚洲人，4.1% 为黑人，0.75% 为其他;94.5% 的图像是男性;大多数 （64.4%） 看起来年龄在 ≥50 岁。DALL·与 AAOS 人口普查相比，E 2 在女性和非白人外科医生的代表性方面表现出显著增加的多样性，而 Midjourney、Runway 和 ImagineAI 的代表性显着降低。结论 本研究强调了 AI 对骨科医生多样性描述的扭曲，影响了公众的看法并可能加剧了差异。DALL·E 2 和 JasperArt 显示出令人鼓舞的多样性，但其他生成器仍然存在限制。 未来的研究应探索改进 AI 的策略，以促进更具包容性和更准确的骨科手术不断发展的人口统计数据，减轻与种族和性别相关的偏见。临床相关性 这项研究具有临床相关性，因为它调查了 AI 生成的图像在描绘骨科医生多样性方面的准确性。研究结果揭示了种族和性别的代表性存在重大差异，这可能会影响社会观念并加剧医疗保健方面的现有差异。