Smart Imaging and Deep Learning for Objective Psoriasis Lesion Scoring: A Scarletred Proof-of-Concept Study

The assessment of psoriasis severity requires detailed, and robust lesion segmentation and tissue classification. This proof-of-concept study aimed to automate the classification of psoriasis disease severity on skin lesions using sophisticated deep learning algorithms. In this study, which was part of a randomized, controlled trial in psoriasis patients (NCT04394936), we used a total of 152 images of representative target plaques and 206 images of local body regions from 26 patients affected by psoriasis, both acquired with Scarletred®Vision, a clinically validated and approved Software as a Medical Device (SaMD) platform operating through an iOS-App on smartphones. As for the modeling aspect, we have trained a deep learning algorithm that employs target lesion scores derived from experts and manually delineated lesion areas. Our algorithm achieved high precision and accuracy in lesion segmentation, yielding a dice coefficient of 0.85. whereas for tissue classification, our model achieved a test accuracy and F-score of 97%. Additionally, our model estimated lesion severity for erythema, scaling and induration attaining correlations of 60%, 65% and 83% respectively between experts and AI scores, using a four-class prediction framework for grades 0 to 3. Similarly, our approach involved automatic body segmentation, enabling lesion localisation and estimation of affected body surface area. Notably, we demonstrated that successful AI classification was attained using a significantly reduced input dataset, showcasing a noteworthy advancement in rapid AI prototyping in clinical and hybrid trial environments, aimed at enhancing expert decision making.