The classification of pulmonary nodules using computed tomography (CT) and positron emission tomography (PET)/CT is often a hard task for physicians. To this end, in our previous study, we developed an automated classification method using PET/CT images. In actual clinical practice, in addition to images, patient information (e.g., laboratory test results) is available and may be useful for automated classification. Here, we developed a hybrid scheme for automated classification of pulmonary nodules using these images and patient information. We collected 36 conventional CT images and PET/CT images of patients who underwent lung biopsy following bronchoscopy. Patient information was also collected. For classification, 25 shape and functional features were first extracted from the images. Benign and malignant nodules were identified using machine learning algorithms along with the images’ features and 17 patient-information-related features. In the leave-one-out cross-validation of our hybrid scheme, 94.4% of malignant nodules were identified correctly, and 77.7% of benign nodules were diagnosed correctly. The hybrid scheme performed better than that of our previous method that used only image features. These results indicate that the proposed hybrid scheme may improve the accuracy of malignancy analysis.
A Combination of Shape and Texture Features for Classification of Pulmonary Nodules in Lung CT Images
