Skin diseases significantly affect the quality of life of approximately 190 million individuals worldwide.The complexity and diversity of their clinical manifestations are the major challenges for traditional diagnostic approaches,and exploring novel diagnostic strategies has become an urgent priority.In recent years,deep learning(DL)technology has been increasingly applied in the intelligent recognition of skin diseases,demonstrating substantial potential.This study provides a systematic review of the research progress of DL in dermatological diagnosis from three major dimensions.First,at the data input level,it focuses on the characterization and preprocessing of multimodal data,including dermoscopic images,ultrasound images,and histopathological slides.Second,at the algorithmic model level,it explores ensemble learning frameworks,multimodal data fusion strategies,multicenter collaborative training approaches,and interpretable model construction.Finally,at the task recognition level,it evaluates the performance of DL models in benign skin disease screening,malignant skin lesion differentiation,and binary as well as multiclass classification tasks.By comprehensively reviewing advancements in DL-based skin disease diagnostic models from multiple perspectives,this paper aims to provide valuable insights for the further optimization and clinical translation of intelligent diagnostic systems.