05-09-2016, 11:36 AM
ABSTRACT
In medical image processing a reliable means of microaneurysms detection in digital retinal images is still an open issue. In this paper, we propose a computerized scheme to improve microaneurysm detection. Image preprocessing is followed by the detection of microaneurysm regions using edge detection. Regions corresponding to blood vessels and bright lesions were removed by image segmentation from the fundus images. Unlike other well-known approaches of machine learning classifiers, we propose a combination of microaneurysm detection and diabetic retinopathy grading using SVM. Microaneurysm detection is decisive in diabetic retinopathy (DR) grading, so we evaluated our approach on four publicly available databases, where a promising AUC
≥ 0.96 is obtained in a “normal” or “abnormal”
-type classification based on the detected microaneurysms. The performance assessment of the automated system is based on Sensitivity, Specificity, and Accuracy together with the ROC curves.
In medical image processing a reliable means of microaneurysms detection in digital retinal images is still an open issue. In this paper, we propose a computerized scheme to improve microaneurysm detection. Image preprocessing is followed by the detection of microaneurysm regions using edge detection. Regions corresponding to blood vessels and bright lesions were removed by image segmentation from the fundus images. Unlike other well-known approaches of machine learning classifiers, we propose a combination of microaneurysm detection and diabetic retinopathy grading using SVM. Microaneurysm detection is decisive in diabetic retinopathy (DR) grading, so we evaluated our approach on four publicly available databases, where a promising AUC
≥ 0.96 is obtained in a “normal” or “abnormal”
-type classification based on the detected microaneurysms. The performance assessment of the automated system is based on Sensitivity, Specificity, and Accuracy together with the ROC curves.
