A CURVATURE BASED APPROACH TO TORTUOSITY MEASUREMENT OF RETINAL BLOOD VESSELS

摘要

Retinopathy of prematurity (ROP) is an eye disease thataffects prematurely-born babies. It may even lead to blindnessin serious cases. Among others, ROP can be recognizedby evaluating the tortuosity of retinal vessels.An automatic detection and evaluation of tortuous vesselscan support ophthalmologists when judging retinal images.There already exist various methods for computing vesseltortuosity. However, they do not always cover the medicalproperties and requirements. Thus, we propose an efficientmeasure, which is based to the total curvature and the numberof curvature sign changes (twists) of the vessel. In orderto achieve a stable detection of twists we represent theextracted vessels by curves composed of circular arcs andline segments: The vessels center pixels are approximatedby smoothly joined segments of constant curvature. Forany maximum tolerance we obtain the minimum number ofsegments. This way vessels are represented by preferablylong curve segments of constant curvature. Together withthe proposed measure this representation enables an effectiveand stable method for evaluating vessel tortuosity. Itsperformance is examined by means of both synthetic dataand vessels of real retinal images.