An Ensemble Stacked Bi-LSTM with ResNet50 Method for Glaucoma Classification in IoT Framework

AN ENSEMBLE METHOD FOR GLAUCOMA CLASSIFICATION IOT FRAMEWORK

Abstract

Rural areas in India face significant healthcare challenges, particularly in managing diabetic complications such as glaucoma due to the lack of timely medical facilities. This study proposes an IoT-based healthcare framework designed to connect rural populations with distant healthcare units, enabling medical professionals to provide necessary interventions promptly. The framework employs an ensemble learning-based Bidirectional Long Short-Term Memory (Bi-LSTM) architecture integrated with ResNet50 for glaucoma classification and detection. The methodology involves pre-processing input images, extracting features, and balancing the dataset using the Synthetic Minority Oversampling Techniques (SMOTE). The balanced dataset is then fed into the model, and the results are classified using a sigmoid function. The framework was validated on four datasets such as ACRIMA, Fundus, ORIGA, and Retinal image datasets. Key findings demonstrate that the proposed model achieves superior performance compared to other models for datasets considered, as evidenced by metrics for ACRIMA datasets such as precision (97%), specificity (99%), accuracy (99%), AUC (97%), recall (97%), and F1-score (97%). For fundus dataset, it obtains accuracy of 99%, precision of 92%, recall of 96%, specificity of 94%, F1-score of 95% and AUC of 90%; accuracy of 99%, precision of 95%, recall of 97%, specificity of 93%, F1-score of 94% and AUC of 88% for ORIGA dataset. For retinal datasets, it yields 97% of accuracy, 93% of precision, 93% of recall, 98% of specificity, 93% of F1-score and AUC of 93%. The study's uniqueness lies in its practical utility for addressing healthcare disparities in rural areas through IoT and machine learning, offering promising solutions for real-world applications.