Neural network classification algorithm based on feature space optimization in application of pulmonary nodules detection

This paper firstly preprocesses CT images using a dot enhance filter which can enhance some round-like nodule tissues and at the same time, restrain interference from tissues of other shapes(for example, linear vessels). Then, the paper adopts a feature space optimization design theory to select seven effective features from twelve original candidate features and regards the combinations of the seven features as input feature vectors of a classifier. Finally, the paper uses a BP neural network classifier to achieve pulmonary nodules classification. The experiment shows that the method presented here can effectively reduce false positive of nodule detection, obtaining a better classification result.