Optimized Feature Selection Approach for Smartphone Based Diabetic Retinopathy Detection

Diabetic Retinopathy (DR) causes damage to the retina's blood vessels. It's also known as a silent illness because it causes only minor vision problems or no signs at all. Annual eye examinations are critical for early detection. As a result, it employs fundus cameras to capture retinal images, but it is an impractical method for widespread screening due to its size and expense. As a result, smartphones are being used to create lightweight, and inexpensive retinal imaging systems that can perform automated DR detection and screening. Preprocessing is done first, including green channel conversion and CLAHE (Contrast Limited Adaptive Histogram Equalization). Furthermore, the segmentation process begins with WT (watershed transform) optic disc segmentation and Triplet half band filter bank abnormality segmentation (exudates, micro aneurysms, hemorrhages, and IRMA) (THFB). Haralick and ADTCWT (Anisotropic Dual Tree Complex Wavelet Transform) methods are then used to remove the various features. Life choice-based optimizer (LCBO) algorithm selects the optimal features. The selected features are then put into an ML classifier, which divides the severity levels into average, mild DR, moderate DR, extreme DR, and proliferative DR. The proposed work is simulated in a Python environment, and datasets such as APTOS-2019-Blindness-Detection and EyePacs are used to test the efficiency of the proposed scheme.