Application Of Improved PSO Optimized ELM Algorithm In Medical Image Segmentation

Medical image segmentation has been a difficult problem because of the complex background, low SNR and no uniform standard. How to select an algorithm with high classification accuracy and low time cost to segment medical images is a question worthy of consideration.As a random search algorithm, Particle swarm optimization algorithm has the advantages of fast convergence speed, strong random search capability, but it is not suitable for solving complex multi-dimensional mapping problem. Neural network, as a complex network model, can deal with all kinds of complex mathematical mapping problems, and has a good generalization performance, but the algorithm is not efficient, and has the problem of over fitting.In the context of the above thought, the main ideas of this paper are as follows:1. The basic particle swarm optimization algorithm has the advantages of algorithm easy to implement and high search efficiency, but there are also some problems which are easy to fall into local optimum. In this paper, through the analysis of basic PSO is easy to fall into local optimum, take the particle swarm individual and global best value with the iterative process stops change number of steps(0T andgT) as the basis for the individual extremum and global extremum of gradient stochastic disturbance regulation method(RPSO) algorithm to improve random search performance and extend the search space. Meanwhile, through the optimization of complex multi-dimensional function experiment illustrated the search performance and efficiency of the improved algorithm.2. This paper introduces the basic theory of the extreme learning machine algorithm(ELM), explains the principle of the training and classification, analyzes the advantages of ELM which has the advantages of fast learning speed and complex problem of network mapping. Then RPSO-ELM algorithm are proposed by using the integrated advantages of RPSO and ELM algorithms. RPSO-ELM algorithm uses the multi-dimensional space random search capability of RPSO to optimize input weights and the hidden layer offset of the ELM model, it’s aim to establish the optimal ELM model to improve the generalization performance and the ability of classification.3. This paper through pre processing and feature extraction and other operations create a sample set, then the sample set will be input RPSO-ELM and classification experiments are carried out. After obtaining the classified results, image will be conducted morphological processing, return the image of sketching outline by algorithm, and by comparing the true positive rate, correct identification rate and background and recognition error rate index of background with gold standard results and results of basic ELM proves that RPSO-ELM applying to medical image segmentation has practical value.