Diabetic Retinopathy Classification Using C4.5

Early detection of diabetic retinopathy (DR) can prevent blindness and improve the quality of life. Practical detection requires a cost-effective screening over a large population. The presence of Microaneurysms (MAs) in a retinal image is the earliest sign of DR. This paper presents an efficient method to automatically detect MAs in a retinal image. The method is based on an advanced wavelet transform and the C4.5 algorithm (a categorization algorithm to distinguish DR and non-DR cases). It uses both the green and red channel data in RGB retinal images for detection of small sized MAs and obtains image feature parameters from the input image. A system was developed to implement the proposed method that displayed a sensitivity of 0.92 and a precision of 0.82.