Advancement and independent validation of a deep learning-based tool for automated scoring of nail psoriasis severity using the modified nail psoriasis severity index

• DOI: 10.3389/fmed.2025.1574413
• Publication Date: 2025-04-02T06:58:34Z
• Authors: Stephan Kemenes; Liu Chang; Maja Schlereth; Rita Noversa de S...; Ioanna Minopoulou; Pauline Fenzl; Giulia Corte; Melek Yalcin Mutlu; Michael Wolfgang ...; Ioannis Sagonas; Birte Coppers; Anna-Maria Liphart; David Simon; Arnd Kleyer; Lukas Folle; Michael Sticherling; Georg Schett; Andreas Maier; Filippo Fagni
• Keywords: Nail disease

To improve and validate a convolutional neural network (CNN)-based model for the automated scoring of nail psoriasis severity using modified Nail Psoriasis Severity Index (mNAPSI) with adequate accuracy across all classes without dependency on standardized conditions. Patients (PsO), psoriatic arthritis (PsA), non-psoriatic controls including healthy individuals patients rheumatoid were included training, while validation utilized an independent cohort patients. photographs pre-processed segmented mNAPSI scores annotated by five expert readers. A CNN based Bidirectional Encoder representation from Image Transformers (BEiT) architecture pre-trained ImageNet-22k was fine-tuned classification. Model performance compared human annotations area under receiver operating characteristic curve (AUROC) other metrics. reader study performed to assess inter-rater variability. In total, 460 providing 4,400 in training dataset. The dataset 118 further who provided 929 photographs. demonstrated high classification dataset, achieving mean (SD) AUROC 86% ± 7% classes. Performance remained robust 80% 9%, despite variability imaging Compared annotation, achieved Pearson correlation 0.94 patient-level, which consistent We developed validated that enables automated, objective reliability need image standardization. This approach has potential clinical utility enabling time-efficient assessment involvement disease possibly as self-reporting tool.