Skin-R1: Clinical Knowledge-Guided Dermatological Diagnosis Using Vision-Language Models

A research team has proposed Skin-R1, a vision-language model designed to improve dermatological diagnosis by integrating clinical knowledge with reinforcement learning, according to a preprint posted on arXiv [1]. The model, detailed in a paper submitted on 18 November 2025 and revised on 26 June 2026, aims to address persistent limitations in current medical vision-language models, including inconsistent diagnostic labels across datasets and the difficulty of scaling knowledge from small, annotated collections to larger, sparsely labeled ones [1][2]. The authors, including Zehao Liu, constructed a textbook-based reasoning generator that produces hierarchy-aware and differential-diagnosis trajectories derived from authoritative dermatology sources [2]. These trajectories are used first for supervised fine-tuning, establishing a clinically grounded reasoning foundation, and then within a reinforcement-learning framework that incorporates the hierarchical structure of skin diseases into its reward design [2]. In experiments across multiple dermatology benchmarks, Skin-R1 consistently improved diagnostic accuracy and robustness compared to state-of-the-art Med-VLM baselines [2]. Ablation studies highlighted the importance of the grounded reasoning supervision introduced during the supervised fine-tuning stage [2]. The work appears on arXiv, an open-access repository for electronic preprints that, as of late 2024, receives approximately 24,000 submissions per month and is not peer-reviewed [6]. The paper is available through arXivLabs, a framework launched in 2020 that allows community collaborators to develop and share experimental tools directly on the platform, such as bibliographic explorers and code finders [5][4].