Performance Improved Hybrid Intelligent System for Medical Image Classification

Kohonen neural networks are one of the commonly used Artificial Neural Network (ANN) for medical imaging applications. In spite of the numerous advantages, there are some demerits associated with Kohonen neural network which are mostly unexplored. Being an unsupervised neural network, they are mostly dependent on iterations which ultimately affect the accuracy of the overall system. Any iteration dependent ANN may have to face local minima problems also. In this work, this specific problem is solved by proposing a hybrid swarm intelligence- Kohonen approach. The inclusion of Particle Swarm Optimization (PSO) in the training algorithm of Kohonen network provides a convergence condition which eliminates the iteration-dependent nature of Kohonen network. The proposed methodology is tested on Magnetic Resonance (MR) brain tumor image classification. A comparative analysis with the conventional Kohonen network shows the superior nature of the proposed technique in terms of the performance measures.