Key points are not available for this paper at this time.
Scar treatment planning in dermatology and plastic surgery requires the joint interpretation of clinical assessment findings, visual scar morphology, and prior treatment context. However, existing approaches often rely on subjective scoring, single-modality analysis, or shallow multimodal fusion, which limits their ability to support consistent and clinically meaningful treatment decisions. To address this issue, we propose DAF-HPNet, a multimodal decision-support framework that integrates structured clinical attributes, scar images, and treatment-history information through a dual-attention fusion mechanism. The model is designed to jointly perform scar severity grading, treatment recommendation, and treatment outcome prediction. Extensive experiments show that DAF-HPNet consistently outperforms representative image-only, shallow-fusion, and attention-based multimodal baselines, achieving the best overall performance across all three tasks, including an outcome-prediction AUC of 0.911, a severity grading Macro-F1 of 0.813, and a treatment recommendation Macro-F1 of 0.806. Additional analyses further demonstrate favorable early-planning capability, cross-cohort generalization, robustness to incomplete or noisy inputs, and clinically aligned explanatory evidence. These results suggest that DAF-HPNet is a promising auxiliary framework for intelligent scar treatment planning and multimodal clinical decision support.
Wang et al. (Mon,) studied this question.