Application of Image and Video Processing Techniques to Eye Fundus with Retinopathy of Prematurity Acquired from a Smartphone

Diagnosis of retinopathy of prematurity, a disorder affecting retinal vasculature, is limited by the subjective definition of disease stages and the lack of experts trained in its identification. Although indirect ophthalmology continues to be the gold standard for this diagnosis, there is a trend towards the use of digitized imaging systems to improve examination conditions. The continuous improvement in image quality of smartphones, together with their widespread availability, make these devices an attractive alternative for screening of this pathology. This article presents a pipeline for conditioning retinal recordings obtained by smartphones for retinopathy of prematurity diagnosis, through the stages of normalization, artifact removal, subsampling and registration in order to create a retinal image mosaic and a processed video with the most relevant information. The resulting images and videos were evaluated by means of a perceptual assessment with four ophthalmologists, using graded scales, and the level of inter-observer agreement was analyzed. Results indicate that ophthalmologists found at least 87% of video reviews and 92.1% of image mosaics acceptable, while inter-observer agreement turned out to be lower than expected by chance according to the statistics used.