Diagnostic classification of solitary pulmonary nodules using support vector machine model based on 2-[18F]fluoro-2-deoxy-D-glucose PET/computed tomography texture features

Abstract

This study aimed to evaluate the diagnostic value of a support vector machine (SVM) model built with texture features based on standard 2-[F]fluoro-2-deoxy-D-glucose (F-FDG) PET in patients with solitary pulmonary nodules (SPNs) at a volume larger than 5 mL. The PET results of 82 patients diagnosed with SPNs between 2014 and 2018 were retrospectively analysed. The volumes of interest (VOIs) of the SPNs were automatically segmented using threshold techniques from the standard PET imaging. Then, a large number of texture features were extracted from the VOIs using texture-analysis software. Next, an optimized SVM machine-learning model that was trained on standard PET images using texture features was employed to identify the optimal discrimination between malignant and benign nodules. Diagnostic models based on the maximum standardized uptake value (SUVmax) and the metabolic tumour volume (MTV) were compared with the SVM model with regard to the SPN diagnostic power. Compared with the SUVmax and MTV models, the texture-based SVM model provided an improvement of approximately 20% in diagnostic accuracy, positive predictive value, negative predictive value and the area under the operating characteristic curve. The receiver operating characteristic curve of the SVM model showed a significant improvement compared with the MTV model (P = 0.0345 < 0.05) and the SUVmax model (P = 0.01 < 0.05). Standard F-FDG PET imaging can increase the differentiation of benign and malignant SPNs with volumes larger than 5 mL using an SVM model based on texture features.