PT - JOURNAL ARTICLE AU - Hong, Xiaotong AU - Lv, Wenbing AU - Yuan, Qingyu AU - Wang, Quanshi AU - Feng, Qianjin AU - Chen, Wufan AU - Rahmim, Arman AU - Lu, Lijun TI - <strong>Prediction of local recurrence and distant metastasis using radiomics analysis of pretreatment nasopharyngea</strong><strong>l <sup>18</sup>F-FDG PET/CT images</strong> DP - 2018 May 01 TA - Journal of Nuclear Medicine PG - 249--249 VI - 59 IP - supplement 1 4099 - http://jnm.snmjournals.org/content/59/supplement_1/249.short 4100 - http://jnm.snmjournals.org/content/59/supplement_1/249.full SO - J Nucl Med2018 May 01; 59 AB - 249Objectives: Local recurrence and distant metastasis are important indications of poor prognosis for nasopharyngeal carcinoma (NPC). We aimed to investigate optimal radiomic features for prediction of recurrence and distant metastasis versus local control from pretreatment 18F-FDG nasopharyngeal PET/CT images. Methods: 87 NPC patients underwent pretreatment 18F-FDG PET/CT scans and received radiotherapy or chemoradiotherapy. During the follow-up period (mean: 30±8 months, range: 3-48 months), local control was achieved in 54 patients. 16 and 17 patients experienced recurrence and metastasis, respectively. A total of 116 radiomic features were extracted, including 19 intensity features, 9 shape features and 88 second- or higher-order textural features. First, one-way ANOVA was used to analyze whether there were significant differences between values of features between patients with different treatment outcomes. Next, we combined the sequential floating forward selection (SFFS) and support vector machine (SVM) classifier to select a combined feature set with improved prediction performance in terms of the area under receiver operating characteristic (ROC) curve (AUC). Subsequently, the above feature set was evaluated using binary stepwise logistics regression analysis to derive independent predictors. Results: The results showed that compactness1, compactness2, Solidity, LGRE_GLRLM, SRLGE_GLRLM and SGE_GLGLM had significant difference (p&lt;0.05) between patients achieving local control versus those with recurrence and metastasis. Among them, only SGE_GLGLM passed the normal distribution and homogeneity tests. Compared with tumors with local control, those with recurrence and distance metastasis showed a significantly higher SGE_GLGLM. The mean AUC was 0.8687±0.0210 when the SVM hyperparameter gamma ranged from 0.7-1.25. The best combined selected feature set consisted of: compactness1, Entropy_GLCM, Sumvariance_GLCM, Dissimilarity_GLCM, Clusterprominence_GLCM, Clustershade, Clustertendency, ZSN_GLSZM, SZLGE_GLSZM, LZLGE_GLSZM, GLV_GLSZM, SGE_GLGLM, LGHGE_GLGLM, and the corresponding AUC was 0.8883. Following binary stepwise logistic regression analysis, only SGE_GLGLM remained significant (p=0.021), and the classification accuracy was 71.3%. Conclusions: The radiomic feature SGE_GLGLM was a predictor of recurrence and metastasis from pretreatment 18F-FDG PET/CT nasopharyngeal carcinoma. Acknowledgments: This work was supported by the National Natural Science Foundation of China under grants 61628105, 81501541, the National key research and development program under grant 2016YFC0104003, the Natural Science Foundation of Guangdong Province under grants 2016A030313577, and the Program of Pearl River Young Talents of Science and Technology in Guangzhou under grant 201610010011.