Multi-image Stitching for Smartphone-based Retinal Fundus Stitching

Abstract

Adequate visualization of the retina is an important factor in assisting ophthalmologists in diagnosing various eye conditions and diseases. Retinal images pertaining to the posterior surface of the eye are captured using specialized devices called fundus cameras. Shortages of such devices are prevalent in areas without sufficient funding due to their expensive costs and, therefore, introduce a barrier to proper eye care. The advent of portable, low-cost smartphone-based fundus cameras has presented a promising solution to address this issue. However, images captured from recently developed smartphone-based portable fundus cameras and ophthalmoscopes have a lower field-of-view and quality compared to expensive tabletop fundus cameras. This paper proposes a method to enhance the diagnostic capacity of such smartphone-based fundus cameras and ophthalmoscopes by stitching multiple retinal images into a mosaic with a greater field-of-view of the retina. The key steps include feature detection, feature correspondence, image warping, and image blending. The method is implemented as a smartphone app intended to be used with low-cost portable fundus cameras and ophthalmoscopes. It successfully generates high-quality mosaics that retain the landmark structures and other notable clinically relevant features present in the original images while maintaining low profile image boundaries. The proposed method is validated by comparison with a standard stitching method on a success rate metric corresponding to the success and quality of image warping and blending. The dataset used for the validation includes 16 retinal fundus mosaicking image sets captured by smartphone-based retinal imaging devices. The comparison results indicate the proposed method is much more effective in stitching retinal fundus images than standard stitching techniques.