Image-based automated Psoriasis Area Severity Index scoring by Convolutional Neural Networks

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• The Psoriasis Area and Severity Index (PASI) is crucial for monitoring psoriasis severity and treatment effectiveness.
• Automating PASI scoring with deep learning, specifically Convolutional Neural Networks (CNNs), offers potential for objective and efficient assessments.

Purpose of the Study:

• To evaluate the performance of CNNs in automated, image-based PASI scoring across different anatomical regions.
• To compare the accuracy of CNN-based PASI scoring with that of human physicians.

Main Methods:

• CNNs were trained on standardized imaging datasets of trunk, arm, and leg regions for individual PASI subscores (erythema, desquamation, induration, area).
• Physicians also retrospectively scored trunk images to enable direct comparison.
• Agreement was quantified using intraclass correlation coefficients (ICCs).

Main Results:

• CNNs demonstrated good agreement with real-life PASI scores, with ICCs for the trunk region ranging from 0.580 to 0.793.
• CNN performance was comparable or slightly superior to physicians in image-based scoring for erythema, induration, and area.
• Physicians showed slightly better performance than CNNs in desquamation scoring.

Conclusions:

• CNNs show significant potential for objective, automated, image-based PASI scoring at the anatomical region level.
• The findings suggest CNNs can serve as a valuable tool in clinical practice and research for psoriasis assessment.
• CNNs achieved comparable or superior performance to physicians in specific PASI subscore evaluations.