Home // EMERGING 2025, The Seventeenth International Conference on Emerging Networks and Systems Intelligence // View article
Authors:
Ishaan Kunwar
Keywords: cataract, convolutional neural network, vision transformer, multilayer perceptron.
Abstract:
Cataract is one of the most prevalent eye diseases affecting the elderly population. In underserved regions, a low ophthalmologist-to-patient ratio and a scarcity of specialized medical devices pose challenges for early detection. This study aims to harness recent advancements in Deep Learning (DL) to automate cataract detection. Although numerous studies have been conducted in this area, improving model accuracy and minimizing overfitting, all while maintaining a simple architecture that requires fewer computational resources, remains challenging. This research proposes a hybrid method that merges featurization achieved by a Convolutional Neural Network (CNN) with classification techniques to improve prediction accuracy. The model's predictive performance is evaluated not only on the original test dataset but also on a newly acquired image set collected independently from a hospital. Experiments are conducted across different model architectures, such as CNNs and hierarchical Vision Transformers (ViTs) in combination with classifiers, such as multi-layer perceptron (MLP), K-nearest neighbors, and RandomForest. The highest accuracy is achieved using a combination of the ConvNeXtXLarge architecture for feature extraction coupled with a MLP classifier, reaching 92.3% on the original test dataset and improving to 94% on the new hospital-based dataset.
Pages: 20 to 29
Copyright: Copyright (c) IARIA, 2025
Publication date: September 28, 2025
Published in: conference
ISSN: 2326-9383
ISBN: 978-1-68558-292-0
Location: Lisbon, Portugal
Dates: from September 28, 2025 to October 2, 2025