Automated Detection and Segmentation of Laser Marks in Retinal Fundus Images of Preterm Infants

Retinopathy of Prematurity (ROP) is a sight-threatening disorder that affects the retina of preterm infants. Laser photo-coagulation is an established treatment for severe ROP to suppress the growth of abnormal blood vessels. This leaves scars or laser marks on the retina's surface, and these laser marks can be detected falsely as blood vessels during followup visits. Therefore, this paper proposes efficient methods for detection, segmentation and removal of laser marks, as a preprocessing step to vessel segmentation. Removal of laser marks from the image will improve the visualization and segmentation of blood vessels and hence will be useful in disease prognosis. Relevant features that perfectly characterize the structure, intensity and texture of the laser marks are identified by training a K Nearest Neighbor classifier. The optimal feature set gave a detection accuracy of 98% and a sensitivity of 100% when tested on a real dataset of infant fundus images.