Diabetic Retinopathy Detection Using Local Extrema Quantized Haralick Features with Long Short-Term Memory Network

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy is a leading cause of blindness.
• Early detection and treatment are crucial for preventing vision loss.
• Current diagnostic methods can be supplemented by automated techniques.

Purpose of the Study:

• To propose a novel approach for automatic diabetic retinopathy detection.
• To improve the accuracy and reduce false positives in diagnosing diabetic retinopathy symptoms.
• To analyze retina vasculature and hard-exudate using fundus images.

Main Methods:

• Feature extraction using local extrema information and quantized Haralick features.
• Utilizing Long Short-Term Memory (LSTM) network with local extrema patterns for precise image analysis.
• Employing a probabilistic approach to suppress false positives.

Main Results:

• The proposed method demonstrated promising performance on two public datasets.
• Evaluated using specificity, accuracy, and sensitivity, the approach showed significant indices.
• Comparative analysis with state-of-the-art methods confirmed the validity and superiority of the proposed approach.

Conclusions:

• The developed technique offers a robust and accurate method for diabetic retinopathy detection.
• This automated approach can aid in earlier diagnosis and timely treatment, mitigating blindness risk.
• The method's performance surpasses existing research, highlighting its clinical potential.