Ensemble For Quality Assessment of Eye Fundus Images Involving Deep Learning Methods

Accurate diagnosis in Telemedicine relies on high-quality eye fundus images. However, the increasing adoption of imaging technology and the increasing volume of images taken per patient have the potential to result in a decrease in image quality. This paper aims to create a reliable method of automated image quality assessment by creating an ensemble of methods to evaluate deep features of the images as well as additional indicators such as blood vessel presence and clarity, optic disc appearance, and image blurriness. The ensemble was tested on two digital fundus image datasets and achieved high accuracy rates of 0.9146 and 0.9845, demonstrating the potential for machine learning algorithms to be used for reliable automated image quality assessment in Telemedicine, helping to ensure accurate diagnoses and better patient care.