Integrated computational system for portable retinal imaging

Abstract

This thesis introduces a system to improve image quality obtained from a low-light CMOS camera-specifically designed to image the surface of the retina. The retinal tissue, as well as having various diseases of its own, is unique being the only internal tissue in the human body that can be imaged non-invasively. This allows for diagnosis of diseases that are not limited to eye conditions, such as diabetes and hypertension. Current portable solutions for retinal imaging such as the Panoptic and indirect ophthalmoscopes require expensive, complex optics, and must be operated by a trained professional due to the challenging task of aligning the pupillary axis with the optical axis of the device. Our team has developed a simple hardware/software solution for inexpensive and portable retinal imaging that consists of an LED light source, CMOS camera, and simple LCD display built into a pair of sunglasses. This thesis presents a multistage solution that registers the retinal tissue on the sensor; identifies its shape and size; performs local integration with phase correlation; and performs global integration with panoramic mosaicing. Through this process we can increase signal to noise ratio, increase image contrast, create super-resolution images, and obtain a large field of view image of the retina. We also lay the groundwork for possible 3D reconstruction to increase the amount of information present. The main contributions of this thesis are in computational methods for improving overall image quality, while other team members focused on the illumination source, optics, or other hardware components.