The Pattern Recognition Of Brain Tumor Cell Images

Now, in the fields of the pattern recognition of brain tumor images, the pattern classifications which are used widely are mainly the artificial neural networks classifications. According to the way by which pattern features are selected, they can be divided into three classes: the texture-based approach, the cell-distribution approach and the cell-graph approach. The latter has higher recognition rates but larger computation than the former two. As the support vector machine has rather greater accuracy, particularly in the two class recognition, it has been applied gradually in the fields of pattern recognition. For the pattern recognition of the brain tumor cell images, this work uses the support vector machine classifications and the cell-graph approach.In the first step, this work uses k-means clustering algorithm and labels on the texture images, then distinguishes the cells from their background and gets the cell images. In the second step, it embeds a grid over the sample image, calculates the probability of a grid entry being a cell, applies a node threshold and then gets nodes of the cell graph. After connecting the nodes, it applies an edge threshold and gets edges of the graph. Finally it develops a cell graph model from the texture images. In the last step, by computing the metrics of the cell graphs, it extracts the pattern features of the texture images and then uses the support vector classification machine learning algorithm to do the pattern classification and recognition. The learning algorithm can be used to distinguish (i) gliomas from normal tissue and (ii) gliomas from inflammation.The experiments are conducted on the clinical data. The experiment results show that the cell-graph approach has better measurement and classification results than the texture-based approach. In the classification (i), the accuracy of the former and the latter are 97.15% and 92.35%, respectively. In the classification (ii), the accuracy of the former and the latter are 94.05% and 87.30%, respectively. For the contrast, it uses the artificial neural networks to do the pattern recognition. In the classification (i), the accuracy of the former and the latter are 96.35% and 87.75%, respectively. In the classification (ii), the accuracy of the former and the latter are 87.35% and 82.75%, respectively. This result shows that in the fields of pattern recognition of brain tumor images, the support vector machine has higher accuracy than the artificial neural networks. So the proposed approach is tested to be valid and reasonable.