An Efficient Artificial Intelligence-based approach for Diagnosis of Media Haze Disease

Abstract

The media opacity developed by media haze in the eye can be a mark to introduce cataracts, corneal edema, vitreous opacities, or small pupils. Hence, the timely and accurate diagnosis of media haze is required to be important and can even impart the prevention from loss of vision that might occur if the disease is not treated on time. In this paper, an optimized and low-computational-cost framework for the diagnosis of media haze disease in color fundus images is carried out using a convolutional neural network which is trained on the RFMiD dataset. This work proposed an approach using a lightweight convolutional neural network for extraction of features from multiple number of layers to diagnose the eyes and then after the extracted features are subjected to classification through three fully connected layers. The comparison is outlined for the proposed work with other deep learning models and based on the achieved experiment results; the proposed framework achieved the highest accuracy of 100% on the training set, 96.95% on the validation set, and 93.89% on the test set that supports the robustness of the proposed framework. This will help in the development of a diagnosis framework which can be easily portable to single board computers with low-time-complexity and faster processing the images frames containing different fundus images.