Automatic evaluation of Nail Psoriasis Severity Index using deep learning algorithm.

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Nail psoriasis causes nail dystrophy, impacting nail matrix and bed.
• Nail Psoriasis Severity Index (NAPSI) quantifies nail psoriasis severity but is time-consuming.
• Current NAPSI assessment poses challenges for real-world clinical application.

Purpose of the Study:

• To develop a deep learning algorithm (DLA) for rapid and reliable NAPSI evaluation.
• To overcome the time-consuming nature of manual NAPSI assessment.
• To provide clinical and research advantages through automated NAPSI scoring.

Main Methods:

• A dataset of 7054 fingernail images from 634 psoriasis patients was created.
• Eight NAPSI features were annotated using bounding boxes on single nail images.
• A YOLOv7-based deep learning algorithm was trained on the annotated dataset.

Main Results:

• The DLA achieved high accuracy, with 98.6% of images differing by only 2 points from ground truth.
• Model accuracy was 67.6% with a mean absolute error of 0.449.
• An intraclass correlation coefficient of 0.876 indicated good agreement with expert assessment.

Conclusions:

• The developed DLA can rapidly and accurately evaluate NAPSI scores.
• This automated approach offers significant benefits for clinical practice and research.
• The DLA enables efficient analysis of large nail image datasets for psoriasis research.