Extraction Technique of Spicules-Based Features for the Classification of Pulmonary Nodules on Computed Tomography.

To avoid the deformation of spicules surrounding pulmonary nodules caused by the classic rubber band straightening transform (RBST), we propose a novel RBST technique to extract spicules-based features. In this paper, the run-length statistics (RLS) features are extracted from the RBST image, in which a smooth circumference with a suitable radius inside the nodule is proposed as the border of transformed object. An experimental sample set of 814 images of pulmonary nodules was used to verify the proposed feature extraction technique. The best accuracy, sensitivity and specificity achieved based on the proposed features were 79.4%, 66.5%, 89.2%, respectively, and the area under the receiver operating characteristic curve was 87.0%. These results indicate that the proposed method of feature extraction is promising for classifying benign and malignant pulmonary nodules.