People with diabetes should be periodically vetted for unusual retinal vessels, to help avoid visual damage and blindness. But how should the progression of new vessels in retinal images be effectively identified for early treatment? Welikala and colleagues put forward a twofold classification system that assists in the identification of new deficient retinal vessels. At the outset, the solid retinal image is regularized to assure healthy image resolution. The image is then preprocessed to boost the illumination of the retinal blood vessels.

Two alternative partitioning algorithms are used to construct two binary vessel maps. The intensity of each pixel of a grey-level image is between 0 and 255. The line strength of a target pixel is the largest average grey-level value of the pixels leaning on a line going by it, minus the average grey-level value of equally positioned pixels in the region. The standard line operator (SLO) segmentation algorithm is used to create a binary vessel map from the alignment strengths of the image pixels with the retinal vessels. A newly adapted line operator (ALO) segmentation procedure is applied to construct a binary vessel map from the alignment strengths of the tilting left and right sides, and the slanted away large nearby image pixels with the retinal vessels. The bulky sections of the regular vasculature are eliminated from the binary vessel maps, to allow accurate detection of new vessels. The binary vessel maps are transformed into segments before computing the density and the numbers of pixels, segments, and orientations of vessels; these features are then normalized. A support vector machine is used to discriminate (1) new vessels from normal vessels of the feature set derived by the SLO, and (2) new vessels from exudates of the feature set generated by the ALO. The incorrect new vessel results are eliminated by comparing the results from the two classifiers.

The dual classification approach was used to process the retinal images in a public database and at a hospital. The methodology was accurate in classifying new retinal vessels. Although no effort was made to pinpoint distinct new vessel pixels or segments in the research, the authors have created a valuable classification system for use by ophthalmologists in validating self-demarcated new vessel areas.