Abstract

Optic disc detection is important in the computer-aided analysis of retinal images. It is crucial for the precise identification of the macula to enable successful grading of macular pathology such as diabetic maculopathy. However, the extreme variation of intensity features within the optic disc and intensity variations close to the optic disc boundary presents a major obstacle in automated optic disc detection. The presence of blood vessels, crescents and peripapillary chorioretinal atrophy seen in myopic patients also increase the complexity of detection. Existing techniques have not addressed these difficult cases, and are neither adaptable nor sufficiently sensitive and specific for real-life application. This work presents a novel algorithm to detect the optic disc based on wavelet processing and ellipse fitting. We first employ Daubechies wavelet transform to approximate the optic disc region. Next, an abstract representation of the optic disc is obtained using an intensity-based template. This yields robust results in cases where the optic disc intensity is highly non-homogenous. Ellipse fitting algorithm is then utilized to detect the optic disc contour from this abstract representation. Additional wavelet processing is performed on the more complex cases to improve the contour detection rate. Experiments on 279 consecutive retinal images of diabetic patients indicate that this approach is able to achieve an accuracy of 94% for optic disc detection