Can AI Revolutionize Psoriasis Care: From Diagnosis to Education?

🔍 Key Finding Machine learning demonstrates potential for enhancing psoriasis care across diagnosis, severity quantification, biomarker discovery, personalized treatment, and patient education. However, challenges regarding dataset biases and the need for dermatologist oversight require attention to ensure equitable and effective AI integration.

🔬 Methodology Overview

• Design: Comprehensive review of the impact of machine learning (ML) on psoriasis care.
• Data Sources: Published literature on ML applications in dermatology, with a focus on psoriasis.
• Selection Criteria: Studies highlighting the use of ML in psoriasis diagnosis, severity quantification, biomarker identification, precision medicine, and AI-driven education.
• Analysis Approach: Qualitative synthesis of findings from selected studies, summarizing the current state of ML in psoriasis care and identifying key advancements and challenges.
• Scope: Current applications of ML in five key areas of psoriasis care, along with discussion of limitations and future research directions.

📊 Results

• Diagnosis: A two-stage deep learning model outperformed 25 Chinese dermatologists in diagnosing psoriasis from images. Another CNN achieved similar diagnostic accuracy to board-certified dermatologists using dermoscopic images. A separate CNN differentiated scalp psoriasis from seborrheic dermatitis more accurately than dermatologists.
• Severity Quantification: An image-AI model outperformed the average performance of 43 dermatologists in calculating PASI scores from over 14,000 images, showing a 33.2% performance gain. AI also demonstrated 90%+ accuracy in BSA calculation, differing only 5.9% from dermatologist markings. Deep learning achieved 88% AUROC in calculating modified NAPSI for fingernail psoriasis.
• Biomarker Identification: Multiple studies using machine learning on RNA sequencing data identified potential psoriasis biomarkers, including ADAM23, GJB2, IRS1, RAI14, ARH-GEF10, and PI3, with varying expression levels in psoriatic lesions compared to controls. Other studies explored imaging and metabolomic biomarkers using AI.
• Personalized Treatment: Machine learning models predicted patient responses to NB-UVB phototherapy with 84% sensitivity and predicted short remission and good outcome with 85% and 75% accuracy, respectively. Other models predicted week 12 response to systemic therapies with 97% accuracy after 1–2 weeks of therapy. AI also predicted 5-year biologic discontinuation with 65.3–77.5% accuracy.
• Education: ChatGPT-4 accurately translated dermatopathology reports into patient-friendly language. While dermatologists’ responses to patient portal messages were preferred by reviewers, ChatGPT responses showed no factual inaccuracies (hallucinations). ChatGPT also generated understandable patient education materials at appropriate reading levels for common dermatological conditions.

💡 Clinical Impact AI-powered tools show promise in enhancing psoriasis care by improving diagnostic accuracy, personalizing treatment plans, and facilitating patient education, especially in areas with limited access to dermatologists. Successful integration hinges on dermatologist oversight to ensure responsible implementation and maintain the essential human element in patient care.

🤔 Limitations

• Underrepresentation of diverse skin tones in training datasets.
• Suboptimal performance of AI across different demographics.
• Datasets primarily comprising images from limited geographic locations (e.g., China).
• Lack of detailed demographic information in some studies, affecting generalizability.
• Reliance on PASI and BSA automated calculations that often exclude palmoplantar lesions.
• Limited exploration of AI for assessing induration, scalp psoriasis severity, and applicability in clinical trials.
• Lack of clinically validated biomarkers for PsA despite promising AI models.
• Potential for ChatGPT/chatbot hallucinations, outdated information, and plagiarism.
• Risk of misinformation from AI-generated materials without clinician review.
• Limited research on psoriasis/PsA-specific AI-driven education and engagement.

✨ What It Means For You AI tools can enhance diagnostic accuracy and personalize psoriasis treatment, particularly in underserved areas, by aiding in diagnosis, severity quantification, biomarker identification, and treatment response prediction. However, dermatologists must oversee AI integration to ensure responsible use, address dataset biases, and maintain the crucial human element in patient care. Further research with diverse datasets is needed to refine AI tools and ensure equitable healthcare delivery.

Reference Smith P, Johnson CE, Haran K, Orcales F, Kranyak A, Bhutani T, Riera-Monroig J, Liao W. Advancing Psoriasis Care through Artificial Intelligence: A Comprehensive Review. Current Dermatology Reports. 2024;13:141–147. https://doi.org/10.1007/s13671-024-00434-y