Automated retinal optic disc detection using pixel based multi fractal analysis

This paper speaks about a novel method for detection and tracing of optic disc (OD) in retinal fundus images. Detection of diabetic retinopathy (DR), Glaucoma and Edema are the ones in which OD detection finds the meaning and need. The method proposed here introduces a unique concept of fractal analysis in its own way, exploiting the geometric structure of eye called Pixel Based Multi Fractal Analysis (PBMFA). At the very outset the retinal fundus images are pre-processed employing a modified Contrast Limit Adaptive Histogram Equalization (CLAHE) method. Furthering Mean Based Localization is carried out on the pre-processed image to locate the Region of Interest (ROI) over which further processes are carried out helping in improvising time complexity and efficiency of the algorithm developed. Modified Canny Edge detection leveraged with Gabor operator is performed over the ROI located, resulting in extraction of blood vessels from the retinal image. Pixel Based Multi Fractal Analysis is done locating the origin of vascular tree (center of the OD), based on the fact that it carries highest fractal dimension. This helps us locate the center most accurately followed by which OD of the retinal image is traced. This algorithm was implemented on DSK6713T making it a real time system and also an end product ready. Hardware implementation results in improvisation of the time complexity of the system too. An evaluation of the proposed algorithm was run over a set of 50 images each from STARE, MESSIDOR and DRIVE projects, containing images from both normal and pathological subjects. Detection and tracing of Optic Disc was found to be accurate up to 98.66%. Abnormal cases like expansion OD in the final stage of diabetic retinopathy were also detected successfully.