Diabetic Retinopathy (DR) is a leading cause of blindness worldwide. Early detection and accurate
grading are crucial for effective treatment. This study proposes a computer-aided diagnosis system
using image processing techniques to detect and grade DR from retinal fundus images. The
proposed approach involves image preprocessing and enhancement, followed by feature extraction
using texture analysis and vessel segmentation. A novel hybrid technique, referred to as Hybrid
GCAM (HGCAM), is introduced for feature extraction. This technique combines the Global
Channel Attention Mechanism (GCAM) with multi-scale feature refinement (MFR). It is
mentioned here as Hybrid GCAM or HGCAM. GCAM – Global channel attention mechanism will
extract global features along with local features. It increases robustness, it reduces computational
complexity by concentrating on critical features, GCAM enhances detection of microaneurysms,
hemorrhages, and other subtle lesions. By integrating GCAM and multiscale feature refinement
the model is supposed to improve diabetic retinopathy detection accuracy, enhance feature
representation and robustness, reduce computational complexity and can gain insights into feature
importance. The system uses Support Vector Machine (SVM) and Convolutional Neural Networks
(CNN) for classification. Early detection and treatment are critical to preventing vision loss, with
multi-label SVM classification (MLC) playing a key role in DR detection and grading. Here we
propose a system to diagnose diabetic retinopathy (DR) from colored fundus images, enabling
early detection and treatment. Our proposed deep learning-based computer-aided diagnosis (CAD)
system detects and analyzes retinal pathological changes without invasive procedures. The system
comprises: Image preprocessing (noise reduction, quality enhancement) Feature extraction and
classification. Our system demonstrates promising results, outperforming existing methods. It
enables accurate DR diagnosis, facilitating timely treatment and preventing complications.