RP squeeze U-SegNet model for lesion segmentation and optimization enabled ShuffleNet based multi-level severity...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic Retinopathy (DR) causes retinal damage due to hypertension.
• Manual DR screening is labor-intensive and time-consuming.
• Automated methods offer a solution for efficient DR detection and grading.

Purpose of the Study:

• To propose a novel automated method for multi-level Diabetic Retinopathy severity classification.
• To enhance the accuracy and efficiency of DR screening using deep learning techniques.
• To develop a system capable of visualizing DR-affected regions and grading disease severity.

Main Methods:

• Fundus images undergo Non-local means Denoising (NLMD) for pre-processing.
• Recurrent Prototypical-squeeze U-SegNet (RP-squeeze U-SegNet) performs lesion segmentation.
• ShuffleNet, optimized by Fractional War Royale Optimization (FrWRO), classifies DR and its severity.

Main Results:

• The FrWRO-SqueezeNet model achieved high performance metrics: 97% sensitivity, 93.8% accuracy, 95.1% specificity, 91.8% precision, and 94.3% F-Measure.
• The system effectively visualizes abnormal areas in fundus images.
• Accurate identification of DR severity levels was demonstrated.

Conclusions:

• The proposed automated scheme accurately classifies Diabetic Retinopathy severity.
• This method can help prevent disease progression and vision loss.
• The system offers an efficient and effective tool for DR screening.