VisionGuard: enhancing diabetic retinopathy detection with hybrid deep learning.

Abstract

Objectives: Early detection of diabetic retinopathy (DR) and timely intervention are critical for preventing vision loss. Recently, deep learning techniques have shown promising results in streamlining this process. The objective of this study was to develop a novel method, termed MobileFusionNet, which integrates the strengths of MobileNet and GoogleNet architectures to automate the detection of DR better using mobile devices.

Methods: The model is implemented in Python and trained on large-scale datasets of retinal images annotated with DR severity levels. The initial step involves pre-processing the images. Further, an advanced feature extraction technique named Histogram of Oriented Gradients (HOG) is utilized, which helps capture the shape/texture information. Finally, the methodology incorporates Linear Discriminant Analysis (LDA), a technique aimed at reducing the dimensionality of the extracted features.

Results: The proposed model displays low inference time and is highly energy efficient. The model exhibits high sensitivity and specificity in detecting DR, with an impressive accuracy of 98.19%.

Conclusions: The model with its modular architecture allows easy integration and holds great potential for revolutionizing DR detection by democratizing access to accurate and timely screening, particularly in resource-limited settings.