Automated diagnosis of maculopathy stages using entropies and Hu's invariant moments

Diabetic maculopathy is characterized by distortion of central vision, because of prolonged diabetic retinopathy. It may lead to permanent central vision loss when the fluid rich fat and cholesterol leaks out of damaged vessels of retina to the macula. Routine eye check up of diabetic patients can help to diagnose diabetic maculopathy at the initial stage and hence can prevent the vision loss. It is very tedious and takes a lot of time of the ophthalmologists to screen retinal images using the naked eye. Hence, a cost-effective computerized detection of diabetic maculopathy screening system can significantly assist the doctors in their diagnosis. This work proposes an automated classification of retinal fundus images into normal, NCSME (Non clinically significant macular edema) and CSME (clinically significant macular edema). Grayscale features of fundus image were extracted using Entropy and Hu's invariant moments. The features which were statistically significant were then fed to the SVM (Support Vector Machine) classifier for automated diagnosis. The proposed technique was validated using 300 images, 100 images each of normal, NCSME and CSME. We have obtained the best results using SVM RBF (Radial basis function) classifier with an average accuracy, sensitivity and specificity of 74.66%, 94.50%, and 93% respectively using ten-fold cross validation. The clinicians can use this system as an adjunct tool in their diagnosis of diabetic maculopathy during the mass screening of diabetic subjects.