Artificial Intelligence-Based Quantification to Assess the Automatic Psoriasis Area and Severity Index

Authors

Taig Mac Carthy, Daniel Dagnino, Alfonso Medela, Gerardo Fernández, Andy Aguilar, Antonio Martorell, Pedro Gómez-Tejerina, Gastón Roustán-Gullón.

Abstract

The Psoriasis Area and Severity Index is widely used in clinical practice and trials, but scoring is time-consuming and can vary between observers. This study evaluated automated methods for rating erythema, induration, and desquamation and for segmenting psoriatic lesions from clinical photographs.

The analysis used 2,857 images annotated independently by three dermatology experts. Among the tested classification models, MiT_b2 achieved accuracies of 60.6% for erythema, 54.3% for induration, and 61.8% for desquamation. For lesion segmentation, Xception achieved an intersection over union of 0.752.

The findings support the feasibility of automated, reproducible assessment of PASI components while identifying room for further validation and model improvement.

Publication

Published in JEADV Clinical Practice, volume 5, pages 82-90. First published online on 8 October 2025.

Available at https://doi.org/10.1002/jvc2.70143.