<p>Ocular disease is the combination of various kinds of eye diseases that majorly affect the sharp and central vision, which requires early intervention and treatment. Fundus images captured through eye screening are used as the source in identifying eye-related diseases, as they are less expensive and accurate. While examining the fundus images, it is necessary to correlate the data from both eye images that were ignored in the past studies. Features representing the ocular diseases are slightly deviated from each other, and thus, it is necessary for careful observation. Using advanced deep learning and feature extraction mechanisms can aid in creating a multi-label ocular disease classification model. Here, an intelligent deep learning architecture is developed for analyzing fundus images “and categorizing several types of ocular diseases from the images.” The required fundus images for the research are obtained from public resources related to ocular diseases. The related features representing the various ocular diseases are extracted using the Variational Autoencoder (VAE) model, which represents the features with a broad range of representations. In the ocular disease classification, features representing different labels are computationally intensive, and direct processing may lead to high computational complexity. This is avoided by a novel weighted features selection approach through the proposed Revamped Fitness-based Teamwork Optimization Algorithm (RF-TOA), which selects appropriate weights for generating a weighted feature set. These weighted features are sent over to the proposed Hybrid Deep Multi-label Classification Network (HDMcCNet) that is developed by integrating the deep CapsNet with the Bayesian Learning network for “multi-ocular disease classification.” The evaluation is performed to check the performance of the developed ocular disease classification model by comparing it with conventional techniques. The outcomes reveal that the proposed HDMcCNet obtained an accuracy of 94.86% leading to quick and better diagnosis and possibly minimizing blindness, and is well suited to multi-label ocular disease classification. “The research study has made its implementation” available at <a href="https://github.com/merliny379/An-Effective-Ocular-Disease-Classification-Using-Hybrid-Deep-Multi-Label-Classification-Network">https://github.com/merliny379/An-Effective-Ocular-Disease-Classification-Using-Hybrid-Deep-Multi-Label-Classification-Network</a>.</p>

HDMcCNet: A Hybrid Deep Multi-label Classification Network for Diagnosing Multiple Ocular Disease via Fundus Images

期刊论文
发表日期：

2025年12月8日

查看全文

R. Merlin,
F. Vincy Lloyd

摘要

Ocular disease is the combination of various kinds of eye diseases that majorly affect the sharp and central vision, which requires early intervention and treatment. Fundus images captured through eye screening are used as the source in identifying eye-related diseases, as they are less expensive and accurate. While examining the fundus images, it is necessary to correlate the data from both eye images that were ignored in the past studies. Features representing the ocular diseases are slightly deviated from each other, and thus, it is necessary for careful observation. Using advanced deep learning and feature extraction mechanisms can aid in creating a multi-label ocular disease classification model. Here, an intelligent deep learning architecture is developed for analyzing fundus images “and categorizing several types of ocular diseases from the images.” The required fundus images for the research are obtained from public resources related to ocular diseases. The related features representing the various ocular diseases are extracted using the Variational Autoencoder (VAE) model, which represents the features with a broad range of representations. In the ocular disease classification, features representing different labels are computationally intensive, and direct processing may lead to high computational complexity. This is avoided by a novel weighted features selection approach through the proposed Revamped Fitness-based Teamwork Optimization Algorithm (RF-TOA), which selects appropriate weights for generating a weighted feature set. These weighted features are sent over to the proposed Hybrid Deep Multi-label Classification Network (HDMcCNet) that is developed by integrating the deep CapsNet with the Bayesian Learning network for “multi-ocular disease classification.” The evaluation is performed to check the performance of the developed ocular disease classification model by comparing it with conventional techniques. The outcomes reveal that the proposed HDMcCNet obtained an accuracy of 94.86% leading to quick and better diagnosis and possibly minimizing blindness, and is well suited to multi-label ocular disease classification. “The research study has made its implementation” available at https://github.com/merliny379/An-Effective-Ocular-Disease-Classification-Using-Hybrid-Deep-Multi-Label-Classification-Network.