Developing a Real Time Algorithm for Diagnosing Glaucoma

Abstract

A Glaucoma is a group of eye diseases causing optic nerve damage, as it has no symptoms and if not detected at an early stage it may cause permanent blindness. Glaucoma progression precedes some structural damage to the retina are the marked symptoms of Glaucoma. Mainly, it is diagnosed by examination of size, structure, shape, and color of the optic disc (OD) and optic cup (OC) and retinal nerve fiber layer (RNFL), which suffer from the subjectivity of human due to experience, fatigue factor etc. Fundus camera is among one of the biomedical imaging techniques to analyze the internal structure of retina, with the widespread adoption of higher quality medical imaging techniques and data, there are increasing demands for medical image-based computer-aided diagnosis (CAD) systems for glaucoma detection, because the human mistakes, other retinal diseases like Age-related Macular Degeneration (AMD) and the existing medical devices like Optical Coherence Tomography (OCT) and Heidelberg Retinal Tomography (HRT) affecting in early glaucoma detection. The proposed technique provides a novel algorithm to detect glaucoma from digital fundus image using combined features set and evaluation of proposed algorithm is performed using a RIM_ONE (version two) database containing fundus images from 158 patients (118 healthy and 40 glaucoma image), Drishti_GS which contain 101 fundus images (70 glaucoma image and 31 healthy images), and RIM_ONE (version one) which contain (200 glaucoma and 255 healthy images) via Matlab software. The proposed system used to detect glaucoma via 3 steps; firstly, OD and OC segmentation. In OD and OC segmentation several steps were done like prepressing, thresholding, boundary smoothing and disc reconstruction to be a full circle where, OD segmentation achieved best dice coefficient (DSC) 90% and Structural Similarity (SSIM) 83% and OC segmentation results are dice coefficient 73% and Structural Similarity (SSIM) 93%, secondly shape, color, and texture features were extracted from segmented parts and then select the most relevant features, thirdly and finally many classifiers were applied to find the best classification accuracy, which was the support vector machine (SVM). This research proposes a novel combination of color-based, shape-based and texture features by extract 13 shape features from disc and cup, extract 25 texture features from RNFL(retinal nerve fiber layer) using gray level co-occurrence method and Tamar algorithm and 3 color feature for each of disc, cup, and RNFL. Next, best features were selected by T-test method and Sequential feature selection (SFS) to introduce eight features with average accuracy 97%, maximize area under vii curve (AUC) 0.99, specificity 96.6% and sensitivity 98.4% to the first database and 91.5 and 94.5 to the second, and third database respectively with training time 1.5623 sec and prediction time 2600 obs/sec (One billionth of a second). The proposed algorithm achieved excellent performance compared with previous studies from 2011 until now in features types and overall performance. The key contribution in this work is the proposed real-time algorithm for glaucoma detection with high accuracy achieved, the proposed method can make a valuable contribution to medical science by supporting medical image analysis for glaucoma detection. Future works suggested to design a complete, integrated, automated system to classify all different types of glaucoma namely: Primary Open-Angle Glaucoma, Normal Tension Glaucoma, Angle Closure Glaucoma, Acute Glaucoma, Exfoliation Syndrome and Trauma-Related Glaucoma, and to upgrade the system to compute the progress of the disease by comparing different image of the same patient to be used for follow up.