Artificial Intelligence-Based Psoriasis Severity Assessment: Real-world Study and Application

Area of Science:

• Dermatology and Artificial Intelligence
• Medical Image Analysis
• Computational Pathology

Background:

• Psoriasis is a common inflammatory skin condition.
• Objective assessment of psoriasis severity is challenging due to the subjectivity of current metrics like the Psoriasis Area and Severity Index (PASI).
• Tele-dermatology requires reliable tools for objective severity assessment.

Purpose of the Study:

• To develop and validate an image-based artificial intelligence (AI) system for objective psoriasis severity assessment.
• To facilitate the long-term management of psoriasis patients through accurate and efficient severity scoring.
• To overcome the limitations of subjective PASI scoring in clinical practice.

Main Methods:

• A deep learning system was trained on 14,096 psoriasis images from 2,367 patients to estimate PASI scores.
• A multiview feature enhancement block and a combined classification-regression header with a cross-teacher header were utilized for improved accuracy.
• The model's performance was evaluated using Mean Absolute Error (MAE) and compared against 43 experienced dermatologists.

Main Results:

• The AI model achieved a 33.2% performance gain over the average of 43 dermatologists in estimating overall PASI scores.
• The model demonstrated a minimum MAE of 2.05 with three input images, indicating high accuracy in severity prediction.
• The SkinTeller app, incorporating the AI model, was utilized 3,369 times across 1,497 patients and received positive feedback from users.

Conclusions:

• An image-AI-based system can automatically calculate PASI scores efficiently, objectively, and accurately.
• The SkinTeller app shows promise as an alternative tool for dermatologists' assessments and for psoriasis patient self-management.
• AI-driven tools can enhance the objective evaluation and management of chronic skin conditions like psoriasis.