Deepnapsi: Deep Learning for Nail Psoriasis Prediction

Nail psoriasis is a frequent condition that is associated with a severe course of rheumatic diseases. Thus, it may be used to adapt the therapy of patients according to the change of nail psoriasis. The nail psoriasis severity index (NAPSI) was developed to measure this condition. Nevertheless, there is a lack of application for the NAPSI, as its use in the clinic is too time-consuming. In this work, we propose the use of advancements in deep learning in a new domain by recording, annotating, and predicting the NAPSI based on photographs of the hand and term our approach DeepNAPSI. This allows not just the automated recording of the NAPSI in the clinic, but also patient self-assessment from home. Our method achieved an area-under-receiver-operator-characteristic curve (AUROC) of 0.83 and 0.86 for macro and micro averaging, respectively, and a mean absolute error of 0.55.