Scholarlink

Background <p>Clinical images are essential in plastic and reconstructive surgery education, particularly for understanding pathology, planning&#xa0;reconstructions, and preparing for board examinations. However, the use of real patient photographs raises ethical concerns related to privacy, confidentiality, and informed consent, especially in head-and-neck conditions,&#xa0;where anonymization is challenging. Recent advances in artificial intelligence (AI) have enabled text-to-image generation, offering a potential alternative to real patient images for educational&#xa0;use.</p> Methods <p>This cross-sectional evaluation study assessed the ability of modern AI&#xa0;image generation&#xa0;applications to produce clinically accurate and realistic images of head-and-neck conditions based solely on standardized written case scenarios. Expert consultants developed examination-style clinical scenarios representing common benign and malignant head-and-neck conditions, intentionally omitting&#xa0;the&#xa0;diagnoses&#xa0;section. Scenarios were entered verbatim into three AI platforms: ChatGPT (DALL·E 3), Midjourney, and Google Gemini 3 (Nano Banana Pro). One image per scenario&#xa0;and&#xa0;platform was generated without prompt refinement.&#xa0;The images&#xa0;were anonymized, randomized, and evaluated by plastic surgeons using a structured 5-point Likert questionnaire assessing fidelity, realism, completeness, errors or hallucinations, and educational usefulness. Non-parametric statistical analyses were performed.</p> Results <p>All AI models demonstrated the ability to generate clinically plausible images. No significant differences were observed between&#xa0;the&#xa0;models for basal cell carcinoma. For melanoma, squamous cell carcinoma, melanocytic nevus, and neurofibromatosis, Gemini achieved higher overall scores and outperformed other models&#xa0;in&#xa0;several clinically relevant domains, particularly surface texture accuracy, representation of key features, and overall clinical accuracy.</p> Conclusion <p>AI-generated images derived from written clinical scenarios can produce realistic and educationally valuable visual representations of head-and-neck conditions. Such images may serve as a feasible&#xa0;and&#xa0;ethical alternative to real patient photographs in surgical education, with variability in performance observed across AI platforms.</p> Level of Evidence V <p>This journal requires that authors assign a level of evidence to each article. For a full description of these Evidence-Based Medicine ratings, please refer to the Table of Contents or the online Instructions to Authors&#xa0;<a href="http://www.springer.com/00266">www.springer.com/00266</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AI-Generated Clinical Images of Head-and-Neck Lesions in Plastic Surgery Education: An Expert Evaluation

  • 期刊论文

  • 发表日期：

    2026年4月15日

查看全文
  • Fahad K. Aljindan,
  • Zainalabden E. Jefri,
  • Abdulaziz M. Alghamdi,
  • Lamees A. Altamimi,
  • Mohammed Alharbi,
  • Hind M. Alotaibi,
  • Ibrahim H. Alhaig

摘要

Background

Clinical images are essential in plastic and reconstructive surgery education, particularly for understanding pathology, planning reconstructions, and preparing for board examinations. However, the use of real patient photographs raises ethical concerns related to privacy, confidentiality, and informed consent, especially in head-and-neck conditions, where anonymization is challenging. Recent advances in artificial intelligence (AI) have enabled text-to-image generation, offering a potential alternative to real patient images for educational use.

Methods

This cross-sectional evaluation study assessed the ability of modern AI image generation applications to produce clinically accurate and realistic images of head-and-neck conditions based solely on standardized written case scenarios. Expert consultants developed examination-style clinical scenarios representing common benign and malignant head-and-neck conditions, intentionally omitting the diagnoses section. Scenarios were entered verbatim into three AI platforms: ChatGPT (DALL·E 3), Midjourney, and Google Gemini 3 (Nano Banana Pro). One image per scenario and platform was generated without prompt refinement. The images were anonymized, randomized, and evaluated by plastic surgeons using a structured 5-point Likert questionnaire assessing fidelity, realism, completeness, errors or hallucinations, and educational usefulness. Non-parametric statistical analyses were performed.

Results

All AI models demonstrated the ability to generate clinically plausible images. No significant differences were observed between the models for basal cell carcinoma. For melanoma, squamous cell carcinoma, melanocytic nevus, and neurofibromatosis, Gemini achieved higher overall scores and outperformed other models in several clinically relevant domains, particularly surface texture accuracy, representation of key features, and overall clinical accuracy.

Conclusion

AI-generated images derived from written clinical scenarios can produce realistic and educationally valuable visual representations of head-and-neck conditions. Such images may serve as a feasible and ethical alternative to real patient photographs in surgical education, with variability in performance observed across AI platforms.

Level of Evidence V

This journal requires that authors assign a level of evidence to each article. For a full description of these Evidence-Based Medicine ratings, please refer to the Table of Contents or the online Instructions to Authors www.springer.com/00266.