Optimised Hybrid Attention-Based Capsule Network Integrated Three-Pathway Network for Chronic Disease Detection in...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Chronic disease detection from retinal images is crucial for early intervention.
• Manual grading of retinal images is inefficient and error-prone.
• Existing Deep Learning (DL) methods face challenges like overfitting and high computational costs.

Purpose of the Study:

• To develop an optimized DL system for enhanced chronic disease detection in retinal images.
• To address limitations of current DL approaches in retinal image analysis.
• To improve the accuracy and efficiency of automated disease diagnosis.

Main Methods:

• Retinal image pre-processing using Normalization and HSI Colour Conversion.
• Feature extraction via Inception-V3, ResNet-152, and Convolutional Vision Transformer (Conv-ViT).
• Classification using an Optimized Hybrid Attention-based Capsule Network with performance enhancements.

Main Results:

• Achieved 99.05% accuracy on the Diabetic Retinopathy 2019 dataset.
• Attained 99.15% accuracy on the APTOS-2019 dataset.
• Demonstrated superior performance, highlighting the technique's effectiveness.

Conclusions:

• Automated methods significantly enhance chronic disease diagnosis efficiency and accuracy.
• The proposed DL system offers a patient-friendly and reliable diagnostic tool.
• Improved diagnostic capabilities benefit both healthcare providers and patients.