深度学习如何解析你的鼻唇沟？

目标: To establish an AI‑powered automated grading system for nasolabial fold (NLF) severity based on the Wrinkle Severity Rating Scale (WSRS). Learners will understand how deep learning can standardize aesthetic assessment, reduce inter‑observer variability, and support objective treatment planning and outcome evaluation in facial rejuvenation. 介绍: Nasolabial folds are key markers of facial aging and primary targets for non‑surgical aesthetic treatments. Traditional WSRS grading is subjective and prone to rater inconsistency. Artificial intelligence enables precise, reproducible facial image analysis, offering a solution to standardize NLF severity evaluation for clinical practice and research. 材料 / 方法: This study constructed a dataset of 6718 standardized facial images, including 1718 clinical outpatient photographs and 5000 public CelebA images. Each image was divided into left and right halves and independently annotated by 3 senior plastic surgeons using the 5-grade WSRS. We built an ensemble deep learning model with ResNet-50 as the backbone, using focal loss to address class imbalance and majority voting to enhance stability and robustness. 结果: The DeepFold ensemble model achieved 0.917 accuracy and 0.917 F1 score, outperforming single baseline models. It reduced grading variance, improved robustness under class imbalance, and focused activation on clinically relevant NLF regions, providing reliable WSRS‑aligned grading 总结: Deep learning‑based automated NLF grading delivers high accuracy and objectivity. It standardizes aesthetic evaluation, supports consistent treatment decisions and outcome monitoring, and can be integrated into clinical workflows to enhance reproducibility in non‑surgical facial rejuvenation.