Similarity evaluation of retinal vascular network based on tree edit distance

The similarity of retinal vascular networks has important applications in the early diagnosis, prediction, and treatment of retinal diseases, such as diabetic retinopathy and age-related macular degeneration (AMD). Unlike direct comparison of vascular parameters or vessel segmentation pixels, a method for calculating the distance between vascular networks based on the tree edit distance is proposed in this paper to represent the similarity between vascular networks. We model the retinal vessel as a tree structure and calculate the tree edit distance between vessel trees which is sensitive to geometrical and topological changes of vessels. Since the retinal vascular network is composed of multiple vessel trees, we weight the distance between vessels and sum them up to represent the distance between networks. The smaller the distance, the higher the network similarity. Finally, experimental results verify that our proposed network distance algorithm can evaluate the similarity between networks objectively.