Longitudinal Treatment-Aware Multimodal AI for Dermatology: A Scoping Review

The field of dermatological AI has rapidly advanced from simple single-image classifiers to large, multimodal foundation models. Despite achieving high diagnostic accuracy, current systems mostly analyze a single point in time and do not incorporate patient history or previous treatments. This review explores the evidence for longitudinal, treatment-aware, and multimodal large language model (LLM) approaches in dermatology, with an emphasis on chronic inflammatory conditions such as eczema, acne, psoriasis, and rosacea. Researchers examined five databases, PubMed, ACM, IEEE Xplore, Web of Science, and Scopus, covering studies from 2020 to 2026. Using PRISMA-ScR guidelines, 30 studies were selected from 1,644 citations. The review identifies five key themes: (1) shift from static to longitudinal AI; (2) multimodal vision-language integration; (3) treatment-aware decision support systems; (4) temporal reasoning capabilities; and (5) benchmarks and evaluation methods. Currently, no models combine longitudinal imaging, medication logs, and LLM reasoning specifically for chronic inflammatory skin diseases. This review highlights five major research gaps and calls for longitudinal, treatment-annotated benchmarks to advance progress in the field.