Abstract:
In papilledema is a human eye disease, where Optic Nerve Head (ONH) of human eye is swollen. Detection of papilledema predicts increased Intracranial Pressure (ICP) and other underlying diseases. If not detected at early stage, it may show devastating effect on visionary power and in some cases becomes fatal. Currently, ophthalmologists are using various invasive methods like Lumber Puncture, Magnetic Resonance Imaging, Optical Coherence Tomography, Computerized Tomography Scans etc., for detection of papilledema. These methods are time consuming and require expertise, therefore, need to develop an automated system for detection of papilledema is inevitable. The goal of this research is to formulate an automated and non-invasive system for the detection of papilledema by using the features of fundus images. In this research, 80 retinal fundus images from STARE database and local sources are chosen for the detection of papilledema. An algorithm is proposed, which extracts 12 features from fundus images. Out of those, 7 features belong to the retinal blood vessels properties and remaining 5 features belong to the Gray Level Co-occurrence Matrix (GLCM) based textural properties of fundus images. Algorithm detects variations in the blood vessels and texture of retinal nerve fibre layer of ONH and its peripapillary region, due to papilledema. A supervised State Vector Machine (SVM) based Classifier is used with its Radial Basis Function (RBF) kernal to classify mixed images into normal and papilledema affected images with the help of features matrix of both type of images. Vascular and textural features have shown better results and algorithm giving accuracy of 97.50%. 40 Papilledema images with mixed severity levels are further tested by using the proposed method to ascertain their severity level that either they are mild papilledema images or severe papilledema images. Proposed system successfully distinguished the mild and severe papilledema images with an accuracy of 75.0%. Results confirm that the proposed system has high rate of accuracy to correctly pronounce the images as papilledema or normal and also determines the severity of papilledema images. This system will find a great utility in the area of automated disease classification.
