Development of a Smart EfficientNetB0-Based Deep Learning Framework for Accurate Prediction and Stage Classification of Diabetic Macular Edema Using Fundus Images

Abstract

Introduction: Diabetic Macular Edema (DME) is a sight-threatening complication of diabetes mellitus that affects the macula, the central part of the retina responsible for detailed vision. DME arises due to fluid accumulation, causing swelling and visual distortion. Early detection and accurate classification of DME stages are crucial for timely intervention to prevent vision loss. Fundus imaging, a non-invasive retinal imaging technique, plays a critical role in diagnosing and monitoring DME.

Objectives: This study aims to develop a deep learning-based framework using the EfficientNetB0 architecture for accurate prediction and classification of DME stages from fundus images, thereby assisting in the early detection and management of DME in diabetic patients.

Methods: The proposed framework leverages the EfficientNetB0 model, known for its balance between computational efficiency and performance. The model was trained on a diverse and extensive dataset of annotated fundus images, encompassing various stages of DME. Preprocessing techniques, including image normalization and augmentation, were applied to enhance model robustness. Performance metrics such as accuracy, precision, recall, F1-score, and area under the receiver operating characteristic (ROC) curve (AUC) were used to evaluate the framework.

Results: The Smart EfficientNetB0 framework demonstrated high accuracy in predicting and classifying DME stages. The results highlighted the model’s capability to capture intricate retinal patterns indicative of DME progression. Additionally, EfficientNetB5 was evaluated and achieved an exceptional accuracy of 0.91, along with high precision, recall, F1-score, and AUC values, further validating the effectiveness of the approach.

Conclusions: The development of the EfficientNetB0-based framework addresses a critical need in diabetic care by enabling early and accurate assessment of DME stages. This innovation holds significant potential to revolutionize the diagnosis and management of DME, reducing the burden of vision loss among diabetic patients. The promising results underscore the value of leveraging advanced deep learning models like EfficientNetB0 and EfficientNetB5 in clinical applications.