Automated capillary flow segmentation and mapping for nailfold video capillaroscopy.

ABSTRACT

OBJECTIVE: This study aimed to develop an automated image analysis method for segmentation and mapping of capillary flow dynamics captured using nailfold video capillaroscopy (NVC). Methods were applied to compare capillary flow structures and dynamics between young and middle-aged healthy controls. METHODS: NVC images were obtained in a resting state, and a region of the vessel in the image was extracted using a conventional U-Net neural network. The approximate length, diameter, and radius of the curvature were calculated automatically. Flow speed and its fluctuation over time were mapped using the Radon transform and frequency spectrum analysis from the kymograph image created along the vessel's centerline. RESULTS: The diameter of the curve segment (14.4 µm and 13.0 µm) and the interval of two straight segments (13.7 µm and 32.1 µm) of young and middle-aged subjects, respectively, were significantly different. Faster flow was observed in older subjects (0.48 mm/s) than in younger subjects (0.26 mm/s). The power spectral analysis revealed a significant correlation between the high-frequency power spectrum and the flow speed. CONCLUSIONS: The present method allows a spatiotemporal characterization of capillary morphology and flow dynamics with NVC, allowing a wide application such as large-scale health assessment.