Automatic Extraction of the Structure of the Retinal Blood Vessel Network of Premature Infants
Lassada Sukkaew MSc*, Bunyarit Uyyanonvara PhD*, Sarah Barman PhD**, Alistair Fielder MD***, Ken Cocker MD***
Affiliation : * Sirindhorn International Institute of Technology, Thammasat University, Thailand ** Kingston University, United Kingdom *** City University, United Kingdom
Objective : Automatically detect the structure of blood vessels in ROP infants to allow ophthalmologist to
analyze and detect the symptom early.
Materials and Methods : This study presents a set of methods for detection of the skeletonized structure of
premature infant’s low-contrast retinal blood vessel network. Steps has been optimized for this study, namely
statistically optimized LOG edge detection filter, Otsu thresholding, Medial Axis transform skeletonization,
pruning, and edge thinning.
Results : A set of 100 test images are grouped together into five testing groups based on their similar charac-
teristics and clinicians suggestions. The authors applied the series of methods proposed on all the 100 images.
The result from the algorithm was compared with ophthalmologists’ hand-drawn ground truth and it can
detect the blood vessel with a high specificity of 0.9879 and sensitivity of 0.8935.
Conclusion : The authors’ algorithm can detect blood vessels effectively even though the image quality may
not be good, have high noise, and low contrast. The algorithm can also detect the blood vessel at important
locations such as the edge of the retina.
Keywords : Retinal vessel extraction, ROP, Infant retinal image
