Retinal Fundus Image Retrieval and Classification using Optimal Deep Learning Model

An increasing number of initiatives for screening people for retinal disorders with widespread obtainability of digital fundus (retina) cameras results in huge collection of color fundus images. The capability to retrieve images relevant to pathological conditions was a powerful functionality that contains extensive applications in automated computer aided diagnosis (CAD), evidence-related medicine, and training ophthalmologists. The automatic extraction of significant attributes namely exudates will be helpful for physicians to effectively overcome several diseases affecting the patient. Though there exist numerous techniques for extracting such features, they lack in retrieval aspect of the data or the precision of the feature extraction. This study designs a Retinal Fundus Image Retrieval and Classification utilizing Optimal Deep Learning (RFIRC-ODL) model. The presented RFIRC-ODL model encompasses two major modules namely retrieval and classification. For image retrieval process, the presented RFIRCODL model employs Residual Network (ResNet) based feature extraction and Manhattan distance based similarity measure. Next, in the second stage, the salp swarm optimization (SSO) algorithm with long short term memory (LSTM) approach can be exploited for fundus image classification. The use of SSO based hyperparameter optimization process helps in attaining maximum fundus image classification outcomes. For investigating the improved efficacy of the projected RFIRC-ODL model, a wide ranging experimental analysis was made on benchmark database. The simulation outcomes demonstrate the betterment of the RFIRC-ODL algorithm over other techniques.