Ground Truth from Multiple Manually Marked Images to Evaluate Blood Vessel Segmentation

Blood vessel segmentation from digital images is one of the valuable processes for medical diagnosis. Many researchers have proposed blood vessel segmentation algorithms, which can segment the blood vessels automatically or with minimum human interventions. One of the popular blood vessel segmentation branches is edge-based segmentation. In this approach, only the edges are detected by the algorithm. While developing edge segmentation algorithms, researchers must evaluate their proposed methods’ performance. If full-reference-based quality measures are utilized, the ground truth, which shows the targetted segmentation output, is needed. This ground truth is commonly generated manually, where human experts identify and draw the edges. However, the manually segmented edges may differ depending on the experts due to several factors, including individual preference. The work in this paper aims to give some insight into how to combine these images. This paper suggests that the edges be classified as useful edges, weak edges, and unintentional edges.