Microscope-Assisted Hypertensive Retinopathy Diagnosis Using Deep Learning Models.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Hypertensive retinopathy (HR) is a complication of hypertension affecting the retina, potentially leading to blindness.
• Manual diagnosis of HR from retinal images is labor-intensive and prone to inaccuracies.
• Early detection and grading of HR are crucial for effective patient management and preventing vision loss.

Purpose of the Study:

• To develop and evaluate a novel, automated technique for the detection and grading of hypertensive retinopathy (HR) using deep learning models.
• To enhance the accuracy and efficiency of HR diagnosis compared to manual methods.
• To provide a tool for clinical application in identifying and classifying HR stages.

Main Methods:

• A novel method combining U-Net and Dense-Net architectures for automated HR detection and grading.
• Image preprocessing using Gabor filters to enhance retinal vasculature.
• Vessel segmentation using U-Net, followed by artery/vein classification with Dense-Net.
• Calculation of the arteriovenous ratio (AVR) from classified vessels for HR assessment.

Main Results:

• The automated system achieved high performance on the AVRDB dataset.
• Average accuracy for HR classification was 99.40%.
• Average accuracy for HR grading reached 99.77%.

Conclusions:

• The proposed U-Net and Dense-Net based method offers a highly accurate and efficient solution for automated HR detection and grading.
• This automated approach can significantly benefit clinical practice by providing reliable diagnostic support.
• The technique holds potential for early intervention and improved management of hypertensive retinopathy, reducing the risk of blindness.