Fully-Automated Lesion Characterization in Ultrawide-Field Retinal Images

AbstractIn this grant application we propose to develop, EyeReadUWF, a fully automated tool for lesion characterization in ultra-widefield scanning laser ophthalmoscopy (UWF SLO) images. In recent times non mydriatic UWF SLO imaging has been shown to be a promising alternative to conventional digital color fundus imaging for grading of diabetic eye diseases, with advantages including 130°-200° field-of-view showing more than 80% of the retina in a single image, no need for multiple fields, multiple flashes, or refocusing between field acquisitions, ability to penetrate media opacities like cataract, and lower rate of ungradable images. UWF SLO images are particularly suitable for detecting predominantly peripheral lesions (PPLs), which have been associated with higher risk of diabetic retinopathy (DR) progression. Accurate quantification of presence and extent of PPLs can only be done by a robust automated tool that is specifically designed for the pseudo-colored images of UWF SLO modality. EyeReadUWF will automatically characterize lesions in pseudo colored UWF images while handling possible artifacts from eyelashes/eyelids and determine the lesion predominance in peripheral and central regions of UWF image. The ability to accurately quantify the presence and extent of predominantly peripheral lesions in UWF SLO images can enable clinicians to develop a more precise DR scoring scheme. This would help identify patients with higher risk of DR progression and onset of PDR, have a positive impact on diabetic patient management, and aid drug discovery research.