A new era of care: Artificial intelligence in psoriasis

The management of psoriasis is undergoing a significant transformation with the integration of artificial intelligence (AI). AI will alleviate diagnosis, in particular for non-dermatologists, and is a great time-saving tool, for example by reducing the need of documentation.

As Prof. Alexander Navarini (University Hospital of Basel, Switzerland) pointed out, an area where AI will crucially change the field of dermatology is the diagnosis of psoriasis [1]. AI-driven diagnostic tools are particularly valuable in identifying early manifestations of psoriasis.

AI-driven diagnostic tools

Machine-learning models, trained on large datasets of dermatological images, have shown to accurately differentiate psoriasis from other skin conditions with high precision. Multiple convolutional neural networks (CNNs) have shown to be competitive with human dermatologists. In a Chinese study, for example, the percentage of misdiagnoses by the AI model was 4%, versus 10% by dermatologists [2]. At present, 2 dermoscopy-driven models can successfully distinguish psoriasis from other erythrosquamous conditions, e.g. scalp psoriasis from seborrheic dermatitis, and, thus, provide real-time diagnostic support [3]. Such systems can reduce diagnostic errors and provide support for non-dermatologists, for example in teledermatology cases. A study showed that psoriasis diagnosis with AI support could be improved from 10% (with general practitioners) up to 12% (with nurses) [4].

Every year, 10 billion text searches for skin, hair, and nails are performed in Google. So far, no flood of patients is looking for confirmation of “Google diagnoses” but this might change in the future. There was an initial plan to make Google Lens a full “Derm-Assistant” with history taking including skin type and duration of symptoms. This plan has been officially abandoned due to lacking FDA approval. Currently, Google Lens has an integrated image search that indicates visual matches only. ChatGPT4 by OpenAI also allows image classification. A comparison between ChatGPT and Google Lens evaluating 200 images of psoriasis showed that 190 cases were diagnosed correctly by GPT4 and 168 by Google Lens [5].

AI assessment of PASI scores

Score evaluation is another interesting AI application. Psoriasis Area Severity Index (PASI) assessments can be time-consuming and scores can vary between dermatologists [6]. Thus, a validated AI tool that assesses psoriasis severity would be a great time-saving application. A small study showed that assistance from a smartphone-based PASI evaluation improved the PASI scoring performed by medical students and dermatologists. A drawback of PASI evaluation by AI is the fact that most current systems are trained on white skin only, where erythema is easily visible [1]. Thus, re-training with pigment-rich images is warranted. Prof. Navarini argued that clinical scores will be among the most useful services offered by AI.

Other applications

Another valuable application might be to embed a treatment suggestion system within the diagnostic algorithm to suggest the best treatment options for psoriasis variants using a Multi-Criteria Decision Making (MCDM) method [7]. Finally, large language models (LLM) such as Llama 3.1 will lead to less or no documentation in practices and clinics. Prof. Navarini stated that audio-LLM combinations have impacted the daily life of physicians like nothing since the introduction of the electronic patient file. According to his prediction, they will be rapidly adopted in private practice and later in hospitals, and can save time and money.

1. 
   
   1. Navarini A. AI in psoriasis Care – A vision for the future. Abstract 7829, EADV 2024 Congress, 25–28 September, Amsterdam, the Netherlands.
   2. Zhao S, et al. J Eur Acad Dermatol Venereol 2020;34:518-24.
   3. Goessinger EV, et al. AM J Clin Dermatol 2024; Sep 11. DOI: 10.1007/s40257-024-00883-y.
   4. Jain A, et al. JAMA Netw open 2021;4:e217249
   5. Praveenraj T, et al. Int J Dermatol 2024; Jul 22. DOI: 10.1111/ijd.17392.
   6. Okamoto T, et al. J Eur Acad Dermatol Venereol 2022;36:2512-5.
   7. Yaseliani M, et al. Comput Ind Eng 2024;187:109754.

Copyright ©2024 Medicom Medical Publishers



Posted on 6 November 20246 November 2024