人工智能驱动的三维分析用于面部白癜风客观定量评估
本研究开发并验证了一种基于人工智能(AI)的三维重建系统,用于客观定量评估面部白癜风皮损。回顾性分析329例面部白癜风患者的标准三视角临床照片,深度学习分割模型在边界勾画上F1分数>0.98,中位Dice相似系数约0.85。AI辅助皮肤科医生评估时,皮损定量的平均绝对误差从6.71降至3.13,与参考皮损面积的Pearson相关系数从0.84提升至0.92,表明AI辅助提高了准确性和一致性。
要点
- AI辅助三维重建系统可客观定量面部白癜风皮损,减少观察者间变异。
- AI辅助使皮损定量平均绝对误差从6.71降至3.13,相关系数从0.84升至0.92。
- 该系统整合面部曲率信息,提供标准化测量,有望优化患者监测和治疗规划。
核心结论
- AI辅助三维重建系统能精确、标准化地定量面部白癜风皮损,提高诊断准确性和一致性。
研究设计与样本
- 诊断性研究,回顾性分析2022至2024年三级转诊中心329例面部白癜风患者的标准三视角临床面部照片。
主要发现
- AI分割模型边界勾画F1分数>0.98,中位Dice相似系数约0.85。
- AI辅助评估时皮损定量平均绝对误差为3.13,手动评估为6.71。
- AI辅助评估与参考皮损面积的Pearson相关系数r=0.92,手动评估为0.84。
- Bland-Altman分析显示AI辅助量化系统偏差减小、一致性界限更窄。
临床怎么用
- 可将AI驱动的三维工具整合到常规皮肤科实践,用于面部白癜风患者监测和治疗规划。
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: Traditional vitiligo assessment methods rely on subjective evaluation and two-dimensional photographic analysis, limiting consistency and accuracy in clinical practice. OBJECTIVES: To develop and validate an artificial intelligence (AI)-assisted, three-dimensional (3D) reconstruction-based system to accurately and objectively quantify facial vitiligo lesions. METHODS: In this diagnostic study, standardized three-view clinical facial photographs from patients with facial vitiligo seen at a tertiary referral centre between 2022 and 2024 were retrospectively analysed. After excluding incomplete or poor-quality images, 329 patients were included. A deep learning segmentation model was trained to detect depigmented lesions and applied to 3D reconstructions with surface mapping to assess facial curvature. Performance was compared with manual assessments made by board-certified dermatologists, with and without AI guidance, across two sessions. RESULTS: The AI segmentation model achieved consistently high boundary delineation (F1 score > 0.98) and robust lesion identification accuracy (median Dice similarity coefficient approximately 0.85). When integrated into dermatologists' evaluations, AI assistance markedly reduced mean absolute error in lesion quantification (3.13 vs. 6.71) compared with manual assessment alone. Correlation with reference lesion areas also showed a higher Pearson correlation coefficient (r = 0.92 vs. 0.84), indicating enhanced reliability. Bland-Altman analysis further revealed reduced systematic bias and narrower limits of agreement, supporting the reproducibility of AI-assisted quantification. Collectively, these findings indicate that AI guidance not only improves segmentation performance but also augments the accuracy and consistency of clinical interpretation. CONCLUSIONS: This study demonstrates the potential clinical impact of AI-assisted 3D reconstruction. The system provides precise and standardized quantification of facial vitiligo lesions by incorporating surface curvature into lesion mapping. Compared with manual assessments, AI-based support improved diagnostic accuracy, reduced interobserver variability and offered consistent measurements across sessions. These findings highlight the promise of integrating AI-driven 3D tools into routine dermatological practice to enhance patient monitoring and optimize treatment planning. Vitiligo is a long-term skin disease. It affects about 30 million people worldwide. Vitiligo occurs when the cells that give skin its colour stop working well. This causes white patches of skin to appear on the body, often on visible areas such as the face, hands and joints. It can affect someone’s emotional wellbeing as well as their physical appearance. Doctors usually check vitiligo by looking at photographs of a person’s skin. They also use scoring systems, such as the Vitiligo Area Scoring Index. However, these methods can take time. Results may also differ depending on which doctor evaluates the photos. Doctors have tried some new methods, including deep learning approaches and special cameras. These often require expensive equipment and still provide only two-dimensional measurements. We have developed a new way to measure vitiligo. It uses artificial intelligence (AI) and three-dimensional computer technology. We used this new AI system to look at routine medical photographs. Doctors who used the AI-based system were better at detecting and measuring white patches on the face than those who did not. In conclusion, this new approach could make it easier to track changes in vitiligo. This will help doctors to see how well treatments are working. In the future, this new approach may also support more personalized treatment plans for people with vitiligo.
引用
- Lee S, Kim YH, Park Y, et al. Objective and quantitative assessment of facial vitiligo via artificial intelligence-powered three-dimensional analysis for diagnostic support. Br J Dermatol. 2026. PMID: 42473859. doi:10.1093/bjd/ljag290
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。