What question did this study set out to answer?

The aim is to enhance the detection and classification of glaucoma using a hybrid deep learning approach.

March 14, 2026Open Access

A hybrid deep learning approach for efficient detection and classification of internal eye diseases

Key Points

The aim is to enhance the detection and classification of glaucoma using a hybrid deep learning approach.
Proposed a hybrid architecture combining CNN for spatial feature extraction and handcrafted descriptors.
Utilized Gated Recurrent Unit to model temporal dynamics.
Implemented the maximum relevance minimum redundancy method for feature optimization.
Classified data using multiple machine learning classifiers including MLP, GRU, SVM, RF, and KNN.
Tested on publicly available glaucoma fundus image datasets.
Achieved 99.7% accuracy in glaucoma detection.
Reported 99.6% precision and 99.5% recall.
F1-score reached 99.55%, surpassing existing methods.

Abstract

Glaucoma is a progressive optic neuropathy and one of the most common causes of the irrecoverable blindness in the world. Traditional diagnostic tools like Optical Coherence Tomography (OCT) and visual field testing are also dependable but usually costly, laborious, and not accessible in low-resource healthcare settings. Also, the interpretation of fundus images manually causes inter-observer variations and delays in diagnosis. Despite some promising results in the field of deep learning applications in the detection of glaucoma, the current models are often limited due to overfitting and low ability to generalize as well as poor temporal progression. This research suggests an innovation to overcome all these challenges by proposing a hybrid deep learning architecture that combines spatial feature extraction of convolutional neural network (CNN) with handcrafted descriptors such as Local Binary Patterns (LBP), Histogram of Oriented Gradients (HOG), and Speeded-Up Robust Features (SURF). The temporal dynamics are shaped with the help of the Gated Recurrent Unit (GRU), and the fusion of the features is optimized with the help of the Maximum RelevanceMinimum Redundancy (mRMR) approach to decrease the redundancy and maximize the discriminative power. The classification is done with various machine learning classifiers which are Multilayer Perceptron (MLP), GRU, Support Vector Machine (SVM), Random Forest (RF), and k-Nearest Neighbors (KNN) to maintain robustness and interpretability. The framework is tested using publicly available datasets of glaucoma fundus images, such as ORIGA, REFUGE and G1020, after being preprocessed (resizing, normalization, region of interest (ROI) localization, and segmentation using U-Net). Experimental findings show better performance with 99.7% accuracy, 99.6% precision, 99.5% recall and 99.55% F1-score, which is better than the state of art methods. The proposed system is a scalable, cost-effective, and clinically relevant system implemented in Python with the help of TensorFlow, OpenCV, and Scikit-learn, to detect glaucoma in its early stages and monitor progression.

Read Full Paperexternally

Mark Helpful

Bookmark

Relay

View Full Paper