Integrating Neural Networks for Enhanced Detection and Classification of Retinal Diseases in OCT Images

Abstract

Listen

Translate article

With the use of neural networks, and more especially a hybrid model combining Convolutional Neural Networks (CNN) and Autoencoders (AE), this study introduces a new strategy for improving the identification and categorization of ocular disorders in Optical Coherence Tomography (OCT) pictures. Early and precise diagnosis is crucial for treating retinal illnesses such diabetic retinopathy as well as age-related macular degeneration, which are among the most common causes of visual impairment. Our goal hybrid model draws on the strengths of both CNNs and AE to enhance the precision and stability of illness identification and categorization by extracting features and performing reducing dimensionality and denoising, respectively. In this approach, OCT images are first processed through the CNN component to extract relevant features capturing intricate retinal structures and pathological characteristics. Subsequently, the AE component reduces the dimensionality of the extracted features while enhancing their clarity by mitigating noise and irrelevant information. The integrated features from both CNN and AE are then utilized to classify retinal diseases, providing a more discriminative and informative representation. The suggested hybrid model is shown to be better to traditional techniques by experimental findings on a broad collection of OCT images. When it comes to detecting and classifying retinal illnesses, the hybrid neural network shows increased specificity as well as sensitivity, which helps physicians make more precise diagnosis faster. Experiment is carried in python and the proposed model outperforms all others with an outstanding accuracy of 98.41%. It excels in sensitivity (97.14%) and specificity (98.12%), affirming its remarkable precision in identifying retinal diseases. The precision and F1 score for the suggested model are 97.03% and 97.56%, respectively, signifying its exceptional capacity for accurate disease classification.