Automatic Detection of Microaneurysms in Retinal Fundus Images using Modified High Boost Filtering, Line Detectors and OC-SVM

Diabetic Retinopathy is a treatable disease throughout the disease progression, if diagnosed in time, to avoid the final vision loss. The relevance of microaneurysms counts in diabetic retinopathy and their close correspondence to the severity of the disease is well documented. As a result, the detection of Microaneurysms is vital to avoid further complications. Non- invasive way to analyze the retinal illness is fundus photographs. The factors like low contrast images, variable sizes of MA's and similarity to blood vessels awakes the need for reliable MA detection system. A novel method for the detection of MA's is proposed using a powerful combination of High Boost Filtering, Line detectors, and Single Class- SVM. Input Retinal image is pre-processed in terms of illumination Equalization, noise and object enhancement using unsharp masking, MA with Blood vessels are further enhanced by modified high boost filtering, to an extent such that single optimal threshold is sufficient to extract them all from image, further effective removal of major blood vessels has significantly reduced FPR. The geometric, intensity and region descriptors fed to OC-SVM for classification of objects as MA or outlier. The proposed approach is evaluated on the DiaretDB1 database and achieved a sensitivity of 72%. Results at MA enhancement and segmentation are better than methods in the literature.