Vessel Extraction of Microcirculatory Video Recordings Using Multi-thresholding Based Verification Algorithm

This study aims to develop a fully-automated algorithm utilizing image processing methods to segment capillaries in microcirculatory video recordings and to quantitatively analyze the changes occur in microcirculation. The video recordings are captured from lingual surfaces of animal subjects at healthy condition to create a baseline and during hemorrhagic shock. The percentage of active blood vessels in the frames will lead to the measurement of microcirculation. Video stabilization, pre-processing, vessel segmentation, identification of active blood vessels and calculating the Functional Capillary Density of the video recordings are the five main steps of the algorithm. Five subregions for each video recording are selected to stabilize the frames through maximizing cross-correlation. Adaptive histogram equalization and median filtering are main steps of pre-processing to enhance the contrast. Multiple level thresholding is applied to pre-processed images and each pixel is verified based on geometric and contrast parameters at each threshold level. The difference of segmented consecutive frames is calculated to detect active capillaries. The algorithm is tested on publicly available retinal image database (DRIVE)and the vessel segmentation results show 93% accuracy when it is compared to manual segmentation results. The Functional Capillary Density calculated for microcirculation video recordings is used to identify healthy and hemorrhagic subjects successfully.