Level-set based adaptive-active contour segmentation technique with long short-term memory for diabetic retinopathy...

Area of Science:

• Ophthalmology and Medical Imaging
• Computer Vision and Machine Learning

Background:

• Diabetic Retinopathy (DR) is a leading cause of vision loss due to abnormal retinal blood vessels.
• Early automated detection of DR is crucial for preventing severe visual impairment.

Purpose of the Study:

• To develop and evaluate an effective segmentation approach for Diabetic Retinopathy (DR) detection.
• To improve the accuracy and efficiency of automated DR screening systems.

Main Methods:

• Developed a Level-set Based Adaptive-active Contour Segmentation (LBACS) method incorporating an Improved Boundary Indicator Function (LSMIBIF) and Adaptive-Active Counter Model (AACM).
• Pre-processed images using Gaussian filters, edge sharpening, contrast, and luminosity enhancement.
• Extracted features using Gray Level Co-occurrence Matrix (GLCM), Local ternary, and binary patterns.
• Classified DR using Long Short-Term Memory (LSTM) networks.

Main Results:

• The proposed LBACS-LSTM model achieved high accuracy rates of 99.43% on the Indian Diabetic Retinopathy Image Dataset (IDRiD) and 97.39% on the Diabetic Retinopathy Database 1 (DIARETDB 1).
• Demonstrated superior performance compared to existing methods like DSA-KL, KNN, and CTSA-SAE.
• The segmentation approach effectively improved boundary conditions and edge detection.

Conclusions:

• The LBACS-LSTM model offers a highly accurate and effective automated approach for Diabetic Retinopathy detection.
• This method shows significant potential for clinical application in early DR diagnosis and management.
• The research highlights the efficacy of advanced image processing and deep learning techniques in ophthalmology.