Feature Feedback-Based Pseudo-Label Learning for Multi-Standards in Clinical Acne Grading

Accurate acne grading is critical in optimizing therapeutic decisions yet remains challenging due to lesion ambiguity and subjective clinical assessments. This study proposes the Feature Feedback-Based Pseudo-Label Learning (FF-PLL) framework to address these limitations through three innovations: (1) an acne feature feedback (AFF) architecture with iterative pseudo-label refinement to improve the training robustness, enhance the pseudo-label quality, and increase the feature diversity; (2) all-facial skin segmentation (AFSS) to reduce background noise, enabling precise lesion feature extraction; and (3) the AcneAugment (AA) strategy to foster model generalization by introducing diverse acne lesion representations. Experiments on the ACNE04 and ACNE-ECKH benchmark datasets demonstrate the superiority of the proposed framework, achieving accuracy of 87.33% on ACNE04 and 67.50% on ACNE-ECKH. Additionally, the model attains sensitivity of 87.31%, specificity of 90.14%, and a Youden index (YI) of 77.45% on ACNE04. These advancements establish FF-PLL as a clinically viable solution for standardized acne assessment, bridging critical gaps between computational dermatology and practical healthcare needs.