Performance Enhanced Hybrid Kohonen-Hopfield Neural Network for Abnormal Brain Image Classification

Abstract

Image classification is one of the pattern recognition techniques which are often used for categorizing the abnormal medical images into different groups. Artificial Neural Networks (ANN) is widely used for this automated application owing to their numerous advantages. Despite the merits, one of the significant drawbacks of the ANN is the convergence problem. Specifically, in Kohonen Neural Networks (KN) & Hopfield Neural Networks (HN), the convergence to the stored patterns is not guaranteed which ultimately leads to misclassification on the input data. In this work, a hybrid approach namely, Kohonen-Hopfield neural network (KHN) is proposed to minimize the convergence problem for medical image classification applications. Experiments are conducted on KHN using abnormal Magnetic Resonance (MR) brain images from four classes. The performance of KHN is analyzed in terms of classification accuracy and convergence rate. Experimental results suggest promising results in terms of accuracy which indirectly indicates the minimization of convergence irregularities. A comparative analysis with other techniques is also performed to show the superior nature of the proposed approach.