Fast fully automatic detection, classification and 3D reconstruction of pulmonary nodules in CT images by local image feature analysis

Abstract

This study develops a computer-aided diagnosis (CAD) system for automatic detection and classification of pulmonary nodules in lung computed tomography (CT) images, which can simultaneously detect and classify ground-glass opacity (GGO), part-solid, and solid nodules. From the morphological feature and local image features of pulmonary nodules, a total of ten features were selected for artificial neural network (ANN) training and testing, to ensure the system can detect and classify pulmonary nodules. Then, the cross-sectional area of each slice of pulmonary nodule was extracted. The response evaluation criteria in solid tumor (RECIST) value was obtained using the Euclidean distance formula, and the number of pixels of the maximum cross-sectional area was counted for area computation. The nodule volume and 3D reconstruction was obtained using the marching cube algorithm and Riemann integral formula. The sensitivity of system detection and classification were 93.13% and 92.70%, respectively. The proposed system can detect GGO, part-solid, and solid nodules, and takes only 0.1 s to process a single image. This study used the clay model imitating pulmonary nodule morphology as physical samples. Each physical sample underwent three CT. The average difference between the physical volume and the volume derived from this study was 0.37%. The detection and classification results of the system enhanced the clinical detection of missing nodules. 3D reconstruction and volume information of the nodules can give their volume doubling time and growth rate when the patient undergoes CT for the second time, thus enhancing the effect of diagnosis and treatment.