Detection of DME in OCT Images Based on Histogram Descriptor

Abstract

AbstractPurpose An approach for detection of Diabetic Macular Edema (DME) retinal diseases from Optical Coherence Tomography (OCT) in diabetic patients has been presented. Methods This is a study using a total of 45 retinal B-Scan OCT image datasets for detection and classification of DME. Initially in this work, the preprocessing was done in the OCT images. Then this was followed by using Histogram of Oriented Gradients (HOG) descriptor method for the detection of DME from retinal OCT images and then using Speeded-up robust feature (SURF) and HOG feature extraction methods the features are extracted from ROI portion of retinal OCT images. Based on extracted features SVM kernel classifiers were used to classify the OCT images into normal and abnormal with diabetic into two different categories. Result In this study a comparison of the overall performances of all the classifiers for detection of DME with respect to classifiers accuracy, sensitivity, specificity, precision and F-score was carried out. Our study shows that the classifier SVM-Polykernel detected DME disease using HOG feature descriptor better than SVM classifiers with SURF and SVM-RBF with HOG feature extraction method with sensitivity of 100%, specificity of 93%, accuracy of 98%, 97% of Precision and 98% of F-Score than of other SVM classifiers with SURF feature, respectively. Conclusion The proposed method helps to ophthalmologist in better detection of DME from retinal OCT images. This will help the eye doctor to initiate early and proper treatment.KeywordsDiabetic Macular Edema (DME)Optical Coherence Tomography (OCT)Support Vector Machine (SVM)