An Efficient Investigation on Age-Related Macular Degeneration Using Deep Learning with Cloud-Based Teleophthalmology Architecture

Abstract

AMD, or age-related macular degeneration, is the fourth most common visual ailment leading to blindness worldwide and mostly affects persons over the age of 60. Early-stage blindness may be reduced with timely and precise screening. High-resolution analysis and identification of the retinal layers damaged by illness is made possible by optical coherence tomography (OCT), a diagnostic technique. Setting up a comprehensive eye screening system to identify AMD is a difficult task. Manually sifting through OCT pictures for anomalies is a time-consuming and error-prone operation. Automatic feature extraction from OCT images may speed up the diagnostic process and reduce the potential for human mistake. Historically, several methods have been developed to identify characteristics in OCT pictures. This thesis documents the development and evaluation of many such algorithms for the identification of AMD. In order to minimize the severity of AMD, retinal fundus images must be employed for early detection and classification. In this work, we develop a useful deep learning cloud-based AMD categorization model for wearables. The suggested model is DLCTO-AMDC model, a patient outfitted with a head-mounted camera (OphthoAI IoMT headset) may send retinaldehyde fundus imageries to a secure virtual server for analysis. The suggested AMD classification model employs Inception v3 as the feature extractor and a noise reduction approach based on midway point filtering (MPF). The deep belief network (DBN) model is also used to detect and classify AMD. Then, an AOA-inspired hyperparameter optimisation method is used to fine-tune the DBN parameters. To ensure the DLCTO-AMDC model would provide superior classification results, extensive simulations were done using the benchmark dataset. The findings prove the DLCTO-AMDC model is superior to other approaches already in use.