A Weighted Counter Propagation Neural Network for Abnormal Retinal Image Classification

Abstract

Artificial Neural Networks (ANN) is one of the primarily used computing techniques for medical image classification applications. However, ANN like Counter Propagation Neural Network (CPN) is less accurate which limits the usage for medical image analysis. In this work, this problem of low accuracy of CPN is tackled by performing suitable modifications in the conventional approach. The concept of weight assignment is used in the distance calculation procedure of Kohonen layer of CPN to enhance the performance of the overall system. The weight assignment procedure is based on the textural features estimated from the input images. This approach is called as Weighted Counter Propagation Neural network (WCPN) and the applicability of this approach is explored in the context of abnormal retinal image classification. Retinal images from four abnormal categories are used in this work. Suitable textural features are extracted from the input images and supplied as input for the ANN. The experiments are performed with the conventional CPN and the proposed WCPN. The performance measures used in this work are classification accuracy and convergence time. Experimental analysis shows promising results for the proposed approach.