文献速递美容整形

一种轮廓统领所有?神经网络分析初次鼻整形术的同质化效应

One Profile to Rule Them All? A Neural Network Analysis of the Homogenizing Effect of Primary Rhinoplasty.
PubMed 文献 · Aesthet Surg J · 2025-01-01

本研究利用Amazon Rekognition神经网络比较3位独立经验丰富的鼻整形外科医生共145例白人女性初次鼻整形患者的标准术前与术后照片相似度。结果显示术后患者相似度显著增加17.58%(P < .001),且医生组内相似度增加(30.89%)显著大于组间(14.15%),提示初次鼻整形术具有同质化效应并体现特定外科医生的可量化「签名」效应。

要点

  • 初次鼻整形术后患者面部相似度显著增加17.58%(P < .001)。
  • 组内相似度增加30.89%显著大于组间14.15%,体现医生特异性签名效应。
  • 该神经网络的同质化量化结果可能影响医生对患者期望的设定。

核心结论

  • 初次鼻整形患者术后同质性增加,尤其在医生组内。
  • 首次用神经网络量化鼻整形的同质化效应及外科医生签名效应。

研究设计与样本

  • 回顾性分析145例白人女性初次鼻整形患者的标准术前术后照片(正面、侧面、斜位)。
  • 样本来自3位独立经验丰富的鼻整形外科医生的公开网站:医生A n=50,医生B n=50,医生C n=45。
  • 使用Amazon Rekognition计算照片间相似度,配对t检验评估绝对相对变化,简单t检验评估组间与组内差异。

主要发现

  • 术后患者整体显著更相似(17.58%,P < .001),侧面19.17%、斜位19.02%、正面14.38%。
  • 亚组分析:医生A相似度增加37.50%,医生B增加24.60%,医生C增加17.68%。
  • 组间相似度增加(14.15%)显著小于组内增加(30.89%)。

临床怎么用

  • 医生可参考同质化效应与签名效应来设定患者对术后外观的期望。
  • 神经网络相似度分析或有助于评估不同医生手术风格的群体影响。

局限性

  • 仅纳入白人女性初次鼻整形患者,结果外推需谨慎。
  • 样本来自公开网站照片,可能存在选择偏倚。
  • 证据等级为3(治疗性)。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

BACKGROUND: A goal of aesthetic rhinoplasty is to improve nasal aesthetics while preserving individuality. OBJECTIVES: We hypothesized that primary rhinoplasty has a homogenizing effect on a patient's appearance and in a distinct manner for a given rhinoplasty surgeon. METHODS: Standardized, unedited preoperative and postoperative photographs, including frontal, lateral, and oblique views of 145 Caucasian female primary rhinoplasty patients from the public websites of 3 independent, experienced rhinoplasty surgeons were included (Surgeon A: n = 50, Surgeon B: n = 50, Surgeon C: n = 45). Utilizing Amazon Rekognition, similarities between photographs were calculated by comparing all preoperative photographs to each other and all postoperative photographs to each other. Differences between preoperative and postoperative views were calculated intracohort and intercohort and standardized by the relative change in similarity. Paired t tests evaluated the absolute relative changes, whereas a simple t test assessed the intercohort and intracohort differences. RESULTS: Patients were significantly more similar postsurgery (17.58%, P < .001) in the lateral (19.17%), oblique (19.02%), and frontal (14.38%) views. In the subgroups, Surgeon A similarity increased 37.50%, Surgeon B had a 24.60% increase, and Surgeon C a 17.68% increase. The intercohort similarity increase (14.15%) was significantly less than the intracohort increase (30.89%). CONCLUSIONS: This study illustrates that primary rhinoplasty patients have increased homogeneity postoperatively, particularly within surgeon cohorts. This is the first study in which the homogenizing effect of rhinoplasties are quantified with neural networks and show the quantifiable "signature" effect of a specific surgeon on their patients' appearances as a population. This study may impact how surgeons set patients' expectations. Level of Evidence: 3 (Therapeutic).

来源:Aesthet Surg J,查看原文。版权归原作者所有。

引用

  1. Khaw KL, Lu SM. One Profile to Rule Them All? A Neural Network Analysis of the Homogenizing Effect of Primary Rhinoplasty. Aesthet Surg J. 2025. PMID: 40501168. doi:10.1093/asj/sjaf109

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。