A lightweight deep learning model with attention mechanisms for hypertensive retinopathy classification

Area of Science:

• Ophthalmology and Medical Imaging
• Artificial Intelligence in Healthcare
• Computational Pathology

Background:

• Hypertensive Retinopathy (HR) is a significant complication of hypertension, necessitating prompt diagnosis to avert vision impairment.
• Current diagnostic methods for HR involve manual fundus image analysis, which is labor-intensive and prone to inter-observer variability.
• Developing automated, accurate diagnostic tools is crucial for efficient clinical decision support.

Purpose of the Study:

• To introduce MA-DNet, a novel lightweight deep learning model for automated Hypertensive Retinopathy classification.
• To enhance HR diagnostic accuracy by integrating DenseNet with channel and spatial attention mechanisms.
• To validate the efficacy of MA-DNet using a recognized clinical dataset.

Main Methods:

• Development of MA-DNet, a deep learning architecture combining DenseNet with channel and spatial attention modules.
• Implementation of feature enhancement and data balancing strategies to optimize model performance.
• Rigorous evaluation of MA-DNet on the OIA-ODIR dataset for Hypertensive Retinopathy classification.

Main Results:

• MA-DNet achieved a classification accuracy of 95.8% on the OIA-ODIR dataset.
• The proposed model demonstrated superior performance compared to existing state-of-the-art methods for HR detection.
• Attention mechanisms significantly contributed to the improved classification accuracy.

Conclusions:

• MA-DNet offers a highly accurate and efficient automated solution for Hypertensive Retinopathy diagnosis.
• The integration of attention mechanisms is effective in improving deep learning model performance for HR classification.
• This AI-driven approach holds promise for enhancing clinical decision-making and preventing vision loss in hypertensive patients.