Computer-aided Cataract Diagnosis with Fundus Retinal Images under Noisy Conditions

Cataract, a leading cause of vision impairment worldwide, demands accurate and efficient diagnosis for timely intervention. Traditional cataract detection methods often face challenges in noisy environments, which adversely affect diagnostic accuracy. In this paper, a novel deep learning approach is proposed to enhance cataract diagnosis using fundus retinal images, with a specific focus on tackling image noise. The proposed model is designed to dynamically adapt its processing based on the fundus retinal image quality, efficiently handling both good-quality and noisy fundus retinal images. For noisy images, a pre-processing filter is applied, enabling the use of DenseNet-201, which has shown resilience to noise-induced distortions. An extensive experiments has conducted on a dataset of fundus retinal images, validated the performance of the proposed method against existing methods. The results demonstrate superior performance in diagnostic accuracy, even in challenging noisy environments.