Radiomics Analysis and Correlation With Metabolic Parameters in Nasopharyngeal Carcinoma Based on PET/MR Imaging

作者：Qi Feng, Jiangtao Liang, Luoyu Wang, Jialing Niu, Xiuhong Ge, Peipei Pang, Zhongxiang Ding · 发表于：Frontiers in Oncology · 年份：2020 · DOI：10.3389/fonc.2020.01619 · 被引用次数：34 · 研究领域：Radiomics and Machine Learning in Medical Imaging、MRI in cancer diagnosis、Advanced X-ray and CT Imaging

Objective: Accurate staging is of great importance in treatment selection for patients with nasopharyngeal carcinoma (NPC). The aim of this study was to construct radiomic models of NPC staging based on positron emission tomography (PET) and magnetic resonance (MR) images respectively, and investigate the correlation between metabolic parameters and radiomic features. Methods: A total of 100 consecutive cases of NPC (70 in training and 30 in testing cohort) with undifferentiated carcinoma confirmed pathologically were recruited. Metabolic parameters of the local lesions of NPC were measured. The total 396 radiomic features based on PET and MRI images were calculated (include Histogram, Haralick, Shape factor, Gray level cooccurrence matrix (GLCM), and Run-length matrix (RLM)) and selected (using maximum relevance and minimum redundancy (mRMR) and least shrinkage and selection operator (LASSO)) respectively. The logistic regression models were established according to these features. Finally, the relationship between metabolic parameters and radiomic features was analyzed. Results: We selected the most relevant nine radiomic features (6 from MR images, 3 from PET images) from local NPC lesions. In PET model, the area under the ROC curve (AUC), accuracy, sensitivity and specificity of the training group were 0.84, 0.75, 0.90 and 0.69, respectively. In MR model, those metrics were 0.85, 0.83, 0.75 and 0.86, respectively. Pearson correlation analysis showed that the metabolic par...