应用面部识别软件量化眼睑成形术后面部感知年龄的降低
本研究从美国整形外科医师学会网站提取至2023年12月的眼睑成形术患者术前术后图像,用两种卷积神经网络平台估算感知年龄,并由两名志愿者用整体美学改善量表评分。共纳入94例患者(平均年龄52.4±10.5岁,84.0%女性),术后CNN感知年龄平均降低3.2岁(术前估算50.0岁 vs 术后估算46.8岁,P<0.01),GAIS评分与感知年龄降低正相关(r=0.33,P<0.05)。
要点
- CNN估算术后感知年龄较术前平均降低3.2岁(50.0岁 vs 46.8岁,P<0.01)。
- GAIS评分与CNN感知年龄降低正相关(r=0.33,P<0.05)。
- 术前被估算比实际年龄老的患者,术后感知年龄降低更大(5.0岁 vs 2.3岁,P<0.05)。
核心结论
- 卷积神经网络可量化眼睑成形术后的感知年龄降低,结果与人工评估一致;对术前看起来比实际年龄老的患者,感知年龄降低最大。
- 作者认为这些发现支持CNN作为评估美学效果的客观工具的潜在用途,并可能有助于术前引导患者对术后年龄降低的期望。
研究设计与样本
- 回顾性研究,从美国整形外科医师学会网站提取至2023年12月的眼睑成形术患者术前术后图像,共纳入94例患者(平均年龄52.4±10.5岁,84.0%女性)。
- 使用FacePlusPlus(中国北京)和Amazon Rekognition(美国西雅图)两个基于CNN的平台估算术前术后年龄,并由两名受训志愿者用整体美学改善量表(GAIS)评分。
- 统计分析采用配对t检验、线性回归和方差分析。
主要发现
- 术前CNN估算年龄比实际年龄年轻2.4岁(估算年龄50.0岁 vs 实际年龄52.4岁,P<0.05)。
- 术后CNN感知年龄平均降低3.2岁(术前估算年龄50.0岁 vs 术后估算年龄46.8岁,P<0.01)。
- 感知年龄降低与性别、真实术前年龄或手术类型无关(P>0.05)。
- GAIS评分与感知年龄降低正相关(r=0.33,P<0.05)。
- 术前被估算比实际年龄老的患者较被估算比实际年龄年轻的患者,CNN感知年龄降低更大(5.0岁降低 vs 2.3岁降低,P<0.05);术前估算年龄与实际年龄的差值与术后年龄降低相关(r=0.31,P<0.05)。
临床怎么用
- 作者认为CNN可作为评估美学效果的客观工具的潜在用途,并可能有助于术前引导患者对术后年龄降低的期望。
局限性
- 原文未报告局限性。
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
INTRODUCTION: Changes to periorbital morphology, including decreased skin elasticity and ptosis, contribute to the appearance of an aging face. Consequently, many patients seek blepharoplasty surgery to address these changes. However, objective measures of surgical success remain sparse. Therefore, we investigated the ability of convolutional neural networks (CNNs) to assess differences in perceived age before and after blepharoplasty and examined correlations between CNN-generated results and human evaluations. METHODS: Pre- and postoperative patient blepharoplasty images from inception through December 2023 were extracted from the American Society of Plastic Surgeons website. Patient age, follow-up time, gender, and type of procedure were recorded. Two CNN-based platforms, FacePlusPlus (Beijing, China) and Amazon Rekognition (Seattle, WA), were used to estimate patients' pre- and postoperative ages. Two trained volunteers rated patients' aesthetic changes using the Global Aesthetic Improvement Scale (GAIS). Statistical analyses to compare patients' pre- and postoperative CNN-estimated ages and factors associated with perceived age reduction included paired t tests, linear regressions, and ANOVA tests. RESULTS: Ninety-four patients were included in the analysis (mean age, 52.4 ± 10.5 years; 84.0% female). Preoperatively, the CNNs estimated patients to be 2.4 years younger than their true ages (estimated age, 50.0 years; true age, 52.4 years; P < 0.05). Postoperatively, the CNNs perceived an average of 3.2 years of age reduction (estimated preoperative age, 50.0 years; estimated postoperative age, 46.8 years; P < 0.01). Perceived age reduction was not associated with gender, true preoperative age, or procedure type (P > 0.05). GAIS scores positively correlated with perceived age reduction (r = 0.33, P < 0.05). Patients estimated as older than their true preoperative age had greater CNN-perceived age reductions compared to those estimated as younger (5.0-year reduction vs. 2.3-year reduction, P < 0.05). The discrepancy between preoperative estimated age and true age correlated with postoperative age reduction (r = 0.31, P < 0.05). CONCLUSIONS: Convolutional neural networks quantified reductions in perceived age following blepharoplasty, with results aligning with human evaluations. CNN-perceived age reduction was greatest in patients who appeared older than their true age, particularly for those with larger discrepancies. These findings support the potential utility of CNNs as objective tools for assessing aesthetic outcomes and may help preoperatively guide patient expectations for postoperative age reduction.
引用
- Khong J, Davis AJ, Wei O, et al. Using Facial Recognition Software to Quantify Perceived Age Reduction in Patients Undergoing Blepharoplasty. Ann Plast Surg. 2025. PMID: 40167099. doi:10.1097/SAP.0000000000004319
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。