1：Experimental Design and Method Selection:

The study employs a hybrid approach combining image processing (histogram equalization and contrast limited adaptive histogram equalization) for enhancement and a convolutional neural network (CNN) for classification. The rationale is to improve diagnosis accuracy by preprocessing images to enhance features before deep learning classification.

2：Sample Selection and Data Sources:

400 retinal fundus images from the MESSIDOR database are used. The database includes images of normal retina, macular edema, proliferative diabetic retinopathy, and non-proliferative diabetic retinopathy, acquired with a color video 3CCD camera on a Topcon TRC NW6 non-mydriatic retinograph.

3：List of Experimental Equipment and Materials:

A computer system with Intel Core i5-3230M CPU, 2.60 GHz, 8 GB RAM, running Windows 10 64-bit; MATLAB r2017a software for implementation; MESSIDOR database images.

4：60 GHz, 8 GB RAM, running Windows 10 64-bit; MATLAB r2017a software for implementation; MESSIDOR database images. Experimental Procedures and Operational Workflow:

4. Experimental Procedures and Operational Workflow: Images are resized to 150x225 pixels, separated into RGB components, enhanced using HE and CLAHE techniques, concatenated back to colored images, and classified using a CNN with layers including image input, convolutional, ReLU, cross-channel normalization, max pooling, fully connected, softmax, and classification layers. Training uses 300 images, testing uses 100 images, with 20 independent runs and fine-tuning via stochastic gradient descent with momentum.

5：Data Analysis Methods:

Performance is evaluated using accuracy, sensitivity, specificity, precision, recall, FScore, and GMean, calculated from true positive, false positive, true negative, and false negative counts. Comparative evaluation with previous studies is done based on these metrics.