Exudates Detection with DBSCAN Clustering and Back Propagation Neural Network

Diabetic Retinopathy (DR) is the third biggest cause of blindness in India. Hard exudates are the primary signs of DR. In this paper the authors propose a novel hybrid mechanism for the detection of Exudates based DBSCAN clustering algorithm. Unlike other clustering algorithms, DBSCAN clustering does not require the number of clusters to be specified. Classification of regions is being done using a system based on Back propagation Neural Network. The authors assessed the performance of algorithm using one of the publicly available databases DIARETDB1. Sensitivity of 90% and a specificity of 85% are achieved using a lesion based performance evaluation criterion and an accuracy of 100% is obtained on image based performance evaluation criterion.