Classification of pulmonary nodules on computed tomography scans. Evaluation of the effectiveness of application of textural features extracted using wavelet transform of image

In this paper, we present an efficient lung nodule classification system. The proposed system extracts various texture and morphological features from the pre-marked regions of interest containing lung nodules. The Gray Level Co-occurrence Matrix method is applied to extract the texture characteristics of the lung nodules. Additional features are extracted from the images transformed by discrete wavelet transform to extend the variety of texture features. Attribute selection is performed on the resulting feature vector to reduce the redundancy of the dataset. Resulting feature collection is used to train several classification models (Neural Network, Random Forest, K-nearest neighbors, Naive Bayes classifier, Support vector machine) which are used to distinguish regions of interest containing malignant nodules from those containing benign nodules.