Texture based back propagation neural networks for segmentation of arteriole and venule in fundus images

Segmentation of medical images plays an important role in the detection of abnormalities present in in it. In this paper, a combination of texture and Back Propagation Neural Networks is proposed for segmenting the arteries and veins present in the eye fundus images. To develop the proposed algorithm, the texture based back-propagation neural network (TBP-NN) is combined by a trained knowledge of textural properties for segmenting the arteriole and venule which can be used in early diabetic retinopathy (DR) detection. Our proposed algorithm is tested with a total of 200 fundus images, which has 100 early-stage DR and 100 normal images, which is used for further training. All retinal images were processed such that a priori knowledge in form of intensity values as features such as Energy. Homogeneity, Contrast and Correlation were automatically obtained. We compared our proposed algorithms efficiency with traditional BP-NN methods and support vector machine (SVM). The results show that the mean accuracy of a proposed algorithm which was higher than either that of the traditional BP-NN or that of the SVM classifier indicating that proposed algorithm could achieve a better segmentation results.