Lian Xiang, Hu Xin, Li Guannan, Wu Siqi, Liu Yihao, Qin Ke, Liu Kai
Department of Plastic and Reconstructive Surgery, Shanghai Ninth People's Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China.
Shanghai Key Laboratory of Tissue Engineering, Shanghai Ninth People's Hospital, National Tissue Engineering Center of China, Shanghai Jiao Tong University School of Medicine, Shanghai, China.
Front Med (Lausanne). 2025 Mar 6;12:1532195. doi: 10.3389/fmed.2025.1532195. eCollection 2025.
Early detection of periocular aging is a common concern in cosmetic surgery. Traditional diagnostic and treatment methods often require hospital visits and consultations with plastic surgeons, which are costly and time-consuming. This study aims to develop and evaluate an AI-based decision-making system for periocular cosmetic surgery, utilizing a Hierarchical Attention Transformer (HATrans) model designed for multi-label classification in periocular conditions, allowing for home-based early aging identification.
This cross-sectional study was conducted at the Department of Plastic and Reconstructive Surgery at Shanghai Jiao Tong University School of Medicine's Ninth People's Hospital from September 1, 2010, to April 30, 2024. The study enhanced the Vision Transformer (ViT) by adding two specialized branches: the Region Recognition Branch for foreground area identification, and the Patch Recognition Branch for refined feature representation via contrastive learning. These enhancements allowed for better handling of complex periocular images.
The HATrans model significantly outperformed baseline architectures such as ResNet and Swin Transformer, achieving superior accuracy, sensitivity, and specificity in identifying periocular aging. Ablation studies demonstrated the critical role of the hierarchical attention mechanism in distinguishing subtle foreground-background differences, improving the model's performance in smartphone-based image analysis.
The HATrans model represents a significant advancement in multi-label classification for facial aesthetics, offering a practical solution for early periocular aging detection at home. The model's robust performance supports its potential for assisting clinical decision-making in cosmetic surgery, facilitating accessible and timely treatment recommendations.
眼部周围衰老的早期检测是整形手术中常见的关注点。传统的诊断和治疗方法通常需要前往医院并咨询整形外科医生，这既昂贵又耗时。本研究旨在开发并评估一种基于人工智能的眼部整形手术决策系统，利用一种为眼部状况多标签分类设计的分层注意力Transformer（HATrans）模型，实现居家早期衰老识别。
本横断面研究于2010年9月1日至2024年4月30日在上海交通大学医学院附属第九人民医院整复外科进行。该研究通过添加两个专门分支增强了视觉Transformer（ViT）：用于前景区域识别的区域识别分支，以及用于通过对比学习进行精细特征表示的补丁识别分支。这些增强功能使得能够更好地处理复杂的眼部图像。
HATrans模型显著优于诸如ResNet和Swin Transformer等基线架构，在识别眼部周围衰老方面实现了更高的准确性、敏感性和特异性。消融研究证明了分层注意力机制在区分细微的前景-背景差异方面的关键作用，提高了模型在基于智能手机的图像分析中的性能。
HATrans模型代表了面部美学多标签分类方面的重大进展，为居家早期眼部周围衰老检测提供了切实可行的解决方案。该模型的稳健性能支持其在整形手术中辅助临床决策的潜力，便于提供可及且及时的治疗建议。
