Frontiers in medicine · 2026
Evaluation of multimodal large language models for psoriasis diagnosis, severity grading, and treatment recommendations from clinical photographs: ChatGPT shows superior performance compared to other large language models.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Boostani Mehdi, Gisondi Paolo, Bellinato Francesco, Kiss Tara, Zouboulis Christos C, Kuraitis Drew, Goldfarb Noah, Nádudvari Nóra, Farabi Banu, Hodosi Balazs, Holló Péter, Wikonkal Norbert M, Lőrincz Kende, Banvölgyi András, Paragh Gyorgy, Kiss Norbert
Аннотация
INTRODUCTION: Psoriasis is common, but diagnosis and early severity assessment can be delayed because of variable presentation and overlap with mimicking dermatoses. Multimodal large language models (LLMs) may assist image-based triage.
OBJECTIVE: To evaluate web-based multimodal LLMs for psoriasis identification, Physician Global Assessment (PGA) scoring, and treatment recommendation quality from clinical photographs.
METHODS: We retrospectively analyzed 303 standardized photographs from 160 patients (Semmelweis University, May 2022-January 2025), including 163 psoriasis lesions and 140 mimickers. Reference diagnosis and PGA were assigned by two dermatologists with third-expert adjudication, treatment outputs were rated for appropriateness. ChatGPT-5, ChatGPT-4o, Gemini 2.5 Flash, and Claude Sonnet 4.5 received identical prompts for diagnosis and, when psoriasis was predicted, PGA and treatment; sessions were periodically reset.
RESULTS: Diagnostic accuracy was highest for ChatGPT-5 (93.1%) and ChatGPT-4o (90.1%), followed by Claude (83.6%) and Gemini (61.5%). Among correctly identified psoriasis cases, PGA accuracy was 93.3% (ChatGPT-5), 92.9% (ChatGPT-4o), 82.7% (Gemini), and 78.1% (Claude). Appropriate treatment recommendations were most frequent for ChatGPT-5 (83.7%) and ChatGPT-4o (82.1%), then Claude (75.0%) and Gemini (53.2%); test-retest agreement favored the OpenAI models.
CONCLUSION: Under standardized conditions, general-purpose multimodal LLMs, especially ChatGPT-5 and ChatGPT-4o, showed strong performance for psoriasis recognition and reasonable support for PGA scoring and treatment suggestions, supporting potential use by primary care physicians when dermatology specialist access is limited.
OBJECTIVE: To evaluate web-based multimodal LLMs for psoriasis identification, Physician Global Assessment (PGA) scoring, and treatment recommendation quality from clinical photographs.
METHODS: We retrospectively analyzed 303 standardized photographs from 160 patients (Semmelweis University, May 2022-January 2025), including 163 psoriasis lesions and 140 mimickers. Reference diagnosis and PGA were assigned by two dermatologists with third-expert adjudication, treatment outputs were rated for appropriateness. ChatGPT-5, ChatGPT-4o, Gemini 2.5 Flash, and Claude Sonnet 4.5 received identical prompts for diagnosis and, when psoriasis was predicted, PGA and treatment; sessions were periodically reset.
RESULTS: Diagnostic accuracy was highest for ChatGPT-5 (93.1%) and ChatGPT-4o (90.1%), followed by Claude (83.6%) and Gemini (61.5%). Among correctly identified psoriasis cases, PGA accuracy was 93.3% (ChatGPT-5), 92.9% (ChatGPT-4o), 82.7% (Gemini), and 78.1% (Claude). Appropriate treatment recommendations were most frequent for ChatGPT-5 (83.7%) and ChatGPT-4o (82.1%), then Claude (75.0%) and Gemini (53.2%); test-retest agreement favored the OpenAI models.
CONCLUSION: Under standardized conditions, general-purpose multimodal LLMs, especially ChatGPT-5 and ChatGPT-4o, showed strong performance for psoriasis recognition and reasonable support for PGA scoring and treatment suggestions, supporting potential use by primary care physicians when dermatology specialist access is limited.