Lung nodule classification using curvelet transform, LDA algorithm and BAT-SVM algorithm

The lung nodule is the manifestation of lung cancer, which is of great significance for early detection and treatment. Traditional feature extraction vectors mainly consist of intensity features, shape features and texture features. A method which combines low and high frequency Curvelet coefficients with the feature vectors based on the traditional features to make up for contour and texture feature in details is proposed; Because PCA lacks supervision function in dimensionality reduction of multi-class data, thus the LDA algorithm is further used to deal with classification labels; Commonly used parameters optimization algorithms in SVM are cross validation grid search, genetic algorithm and PSO algorithm. In this paper, the new smart bat algorithm is used for parameters optimization, making it simple and rapid. The experimental results show that the proposed method is feasible and the recognition accuracy is higher.