Optic disc segmentation applied to retinal images with glaucoma, diabetic retinopathy, and optic disc edema

Precise detection of the optic disc (OD) is an important task in the diagnosis of several eye diseases. To manage the massive visual-impaired population, there is a huge demand for efficient and remote retinal imaging techniques that can visualize the health of the retina. In this regard, the use of handheld mobile cameras attached to a smartphone is a promising approach. With the advancement in retinal lens technology, a smartphone could offer a low-cost imaging system alternative to standard equipment. However, smartphone retinal images are often of low quality compared to those obtained on standard equipment. They also have a narrow field of view, an incomplete vessel structure, and uneven illustration. Existing image processing algorithms have a problem with handling retinal images acquired from a mobile phone. In this work, we present OD detection and segmentation methods for mobile-phone retinal images of eyes with glaucoma and diabetic retinopathy (DR). The first work provides an automatic glaucoma pre-screening method based on OD-initialization by the exclusion method (EM) and the alternative deflation/inflation gradient vector flow (ADI-GVF) segmentation method. To find an OD location, EM scans for bands across the image that have a low number of vessels in the horizontal direction. For each band, it scans for a place where the number of vessels in the vertical direction is the greatest. The ADI-GVF segments the OD and optic cup (OC) areas using an initial location resulting from EM. The cup-to-disc area ratio (CDAR) is measured using the segmented areas of OD and OC. A CDAR cutoff of 0.3 is used to classify the glaucomatous and healthy images. The classification performance is evaluated on a mobile phone retinal dataset and two standard datasets, such as Drishti-GS and High-Resolution Fundus (HRF) datasets. The proposed method outperforms state-of-the-art methods such as the traditional gradient vector flow, the region growing, and the watershed transformation. Our glaucoma detection method shows a high true positive rate (93.33%) and a low false omission rate (6.66%) for the mobile-phone dataset. The second work provides automatic OD detection and segmentation methods for retinal images with DR. Eyes with DR experience the leakage of blood and other fluids from retinal vessels. The EM fails for cases when the vessels distort, swell, and are incomplete in the image. Therefore, we propose a new, fully automatic hybrid method for OD localization (HLM). It is designed for and verified on mobile phone retinal images. The HLM analyzes the vessel structure and finds the OD locations by using the EM when an image has a complete vessel system and a newly proposed line detection method otherwise. For OD segmentation, an active contour model followed by the circle fitting approach is integrated into the HLM. The proposed method was tested on three mobile-phone datasets (Exudate, Hemorrhage, and Healthy) and four datasets (STARE, DIARETDB0, DIARETDB1, and Retinopathy of prematurity (ROP)) obtained by standard equipment. For mobile-phone datasets, the HLM achieves an average accuracy of 98% for OD localization. The segmentation routine obtains an average precision of 92.64% and an average recall of 82.38%. Testing against several state-of-the-art methods on the standard datasets shows comparable performance.Optic disc edema (ODE) is an important ocular manifestation in neuro-ophthalmic diseases. The presence of ODE is recommended to detect at an early stage to avoid blindness and visual impairment. The ODE area was suggested as a reliable index to evaluate the severity of ODE. Although retinal images from standard cameras have good quality, those with ODE may have different textures from a typical OD and abnormal characteristics such as dimmer color than usual, blurring and enlargement of OD, and obscuring veins. These unique characteristics make existing OD segmentation methods fail to work. Lastly, this work also presents an ODE segmentation method using a factorization-based active contour model based on HLM initialization. The proposed method was tested on a total of 35 images with ODE. It obtains a localization success rate of 97.14% and an average precision of 91.74%, and an average recall of 79.21% for ODE segmentation performance.