In the practice of psoriasis, nail lesions must be accurately evaluated in addition to skin and joint lesions. Scoring with Nail Psoriasis Severity Index (NAPSI) is often used as a tool to assess the severity of nail psoriasis, but there are several vexing issues. First, calculating NAPSI scores on multiple digits is rather time-consuming. Second, NAPSI scores often vary widely among scoring dermatologists. In this study, we developed a deep learning-based NAPSI prediction system to solve the current problems of NAPSI scoring and make it easier and more effective to implement in daily clinical practice. The "nail region detector" that detects nails in images was constructed utilizing an object detection model. The "NAPSI predictor" that predicts NAPSI of the detected nails was constructed using a classification model. Those were combined as an integrated system. A total of 1876 hand and foot images containing nails used for the "nail region detector" were collected from Google searches as well as from our patients with nail psoriasis. We prepared 2939 nail images for the "NAPSI predictor" by manually cropping out the nail regions from the images of our patients. Annotation of NAPSI scores was performed by dermatologists with abundant clinical experience in nail psoriasis. The performance of the integrated system was then evaluated by comparison of the NAPSI scores between the system and the dermatologists. The results showed that the "nail region detector" recognized nails with 93.8% accuracy, and the "NAPSI predictor" proposed NAPSI scores with 82.8% accuracy. The integrated system was also able to predict NAPSI scores with higher accuracy and lower variability than the dermatologists. Accordingly, it was shown that deep learning can be used to quickly and accurately isolate nail regions from images and estimate NAPSI scores at a level comparable to that of a dermatologist. The method used in this study is expected to be applicable not only to nail psoriasis but also to image analysis of other nail diseases in future investigations.

In the management of psoriasis, nail lesions must be accurately evaluated in addition to skin and joint lesions. Scoring with Nail Psoriasis Severity Index (NAPSI) is often used as a tool to assess the severity of nail psoriasis, but there are several limitations. First, calculating NAPSI scores on multiple digits is rather time-consuming. Second, NAPSI scores often vary widely among scoring dermatologists. In this study, we developed a deep learning-based NAPSI estimating system to solve the current problems of NAPSI scoring and make it easier and more effective to implement in daily clinical practice. The "nail region detector" that detects nails in images was constructed utilizing an object detection model. The "NAPSI estimator" that estimates NAPSI of the detected nails was constructed using a classification model. Those were combined as an integrated system. A total of 1876 hand and foot images containing nails used for the "nail region detector" were collected from Google searches as well as from our patients with nail psoriasis. We prepared 2939 nail images for the "NAPSI estimator" by manually cropping out the nail regions from the images of our patients. Annotation of NAPSI scores was performed by dermatologists with abundant clinical experience in nail psoriasis. The performance of the integrated system was then evaluated by comparison of the NAPSI scores between the system and six dermatologists who are different from the annotators. The results showed that the "nail region detector" recognized nails with 93.8% accuracy, and the "NAPSI estimator" proposed NAPSI scores with 82.8% accuracy. The integrated system was also able to predict NAPSI scores with higher accuracy than the dermatologists. Accordingly, it was shown that deep learning can be used to quickly and accurately isolate nail regions from images and estimate NAPSI scores at a level comparable to that of a dermatologist. The method used in this study is expected to be applicable not only to nail psoriasis but also to image analysis of other nail diseases in future investigations.

ESDR102 - A deep learning system enables a prompt and objective scoring of Nail Psoriasis Severity Index

The scientific meeting of the European Society for Dermatological Research (ESDR) is held yearly, usually in September. With an international attendance of more than 1000 skin scientists from around the globe, the four-day event includes short oral presentations from peer-reviewed abstracts, keynote lectures from international experts, sponsored symposia, satellite programs, educational events as well as an important focus on posters, networking and building collaborations amongst scientists. For more information about this year’s meeting visit our website.