<![CDATA[Retinal diseases are serious conditions that can cause vision impairment and, in severe cases, blindness, affecting 6.3% to 17.9% of cases per 100,000 people annually worldwide. Early diagnosis is crucial but often time-consuming, prompting the use of Artificial Intelligence (AI) models like DenseNet, part of the Convolutional Neural Network (CNN) architecture, to streamline the process. This study utilizes the Retinal OCT Images dataset from Kaggle, comprising 83,600 images categorized into four classes. To address the low contrast in Optical Coherence Tomography (OCT) images, the Contrast Limited Adaptive Histogram Equalization (CLAHE) technique was applied during preprocessing. Results indicate that DenseNet without CLAHE achieved an accuracy, precision, recall, and F1-score of 95%, while incorporating CLAHE improved these metrics to 98%. The application of CLAHE also reduced classification bias and error, enhancing model reliability despite requiring more training epochs (43 compared to 39 without CLAHE). These findings demonstrate the potential of CLAHE to optimize DenseNet performance in retinal disease classification. Future research could explore other image enhancement techniques or apply the method to different retinal disease datasets, contributing to improved diagnostic accuracy in clinical settings.]]>
Copyrights © 2024
