Detection of Nailfold Capillaries Using Smartphone Microscopy and Deep Learning Analysis

Nailfold capillaroscopy is an efficient, noninvasive diagnostic method for predicting various diseases. However, its conventional application is constrained by the need for specialized equipment, hospital visits, and the reliance on the subjective judgment of the examiner, which undermines its objectivity. In this study, we propose a mobile-based capillary analysis system integrating a 3D-printed smartphone microscope and a deep learning model. We evaluated six segmentation models using data collected from the developed device, and U-Net demonstrated the best performance with a Dice Similarity Coefficient of 96.10% and an Accuracy of 93.23%. Furthermore, we implemented a client-server architecture application that enables real-time visualization of segmented images on mobile systems. This study demonstrates the potential to enhance healthcare accessibility by reducing economic and temporal burdens and offering a user-friendly diagnostic environment, particularly in underserved areas.