An Efficient Image Segmentation Approach Using Fully Convolutional Neural Networks with Attention Fusion for Ophthalmic Disease Classification

Résumé fourni par la source

The process of segmenting retinal images and classifying ophthalmic diseases establishes a crucial pathway for both early disease identification and clinical treatment choices. Deep Learning (DL) methods demonstrate effective results however, previous research suffers from three critical issues, which include their failure to extract sufficient features, their inability to generalize well and their weak connection between preprocessing and augmentation methods. This research develops an Attention-Based Fully Convolutional Neural Network (FCNN) framework to solve existing problems in ophthalmic disease classification through Ocular Image Analysis – Ophthalmic Disease Intelligent Recognition (OIA-ODIR) retinal fundus image dataset analysis. The proposed methodology applies image preprocessing methods, which include resizing, denoising and histogram equalization, together with data augmentation techniques, which include rotation, scaling, normalization and cropping to generate new dataset samples and enhance the quality of images. The attention-based FCNN architecture classifies processed retinal images through its ability to extract retinal features that distinguish different eye conditions. The experimental results show that the developed model reaches an accuracy of 99.20% with a precision of 98.17% and a recall of 99.35%, and an F1-score value of 98.74%, which surpasses multiple current algorithms. The study concludes that the integration of preprocessing, augmentation, and attention mechanisms significantly enhances segmentation and classification performance. The proposed framework delivers accurate results that demonstrate strong performance and dependable outcomes for classifying ophthalmic diseases through retinal fundus image analysis.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.