TY - JOUR T1 - Prognostic potentials of radiomics analysis on the PET and CT components of PET/CT complementary to clinical parameters in patients with nasopharyngeal carcinoma JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 1734 LP - 1734 VL - 59 IS - supplement 1 AU - Wenbing Lv AU - Qingyu Yuan AU - Quanshi Wang AU - Jianhua Ma AU - Qianjin Feng AU - Wufan Chen AU - Arman Rahmim AU - Lijun Lu Y1 - 2018/05/01 UR - http://jnm.snmjournals.org/content/59/supplement_1/1734.abstract N2 - 1734Objectives: We aimed to investigate the prognostic performance of radiomic features extracted from both the PET and CT components of baseline 18F-FDG PET/CT complementary to clinical features in patients with nasopharyngeal carcinoma (NPC). Methods: 128 NPC patients (85 and 43 used for training and validation, respectively) were retrospectively enrolled. All patients underwent pre-treatment 18F-FDG PET/CT scans and clinical examination. The mean follow-up time was 32±9 months. 8 wavelet decompositions (LLL, LLH, LHL, LHH, HLL, HLH, HHL and HHH) of each original image were first obtained, and 4 discretizations (32, 64, 128 and 256 bins) were then applied to these 9 (8+1) types of images. Thus, 3348 radiomic features (9 shape-, 19 intensity- and 88 texture-based: 9+19[asterisk]9+88[asterisk]9[asterisk]4=3348) were extracted from the PET or CT components, and 13 clinical parameters (age, sex, initial T, N, M stage, AJCC stage, pretreatment plasma EBV DNA, VCA-IgA, LYM, NEUT, HGB, PLT and LDH) were used to predict progression-free survival (PFS). Following univariate screening for significant features by univariate Cox regression analysis, seven multivariate models were constructed by forward stepwise Cox regression analysis. This involved different combinations: (1-3) use of clinical parameters or PET or CT features alone, or (4-7) integrating PET features and/or CT features and/or clinical parameters (denoted as Clinical, PET, CT, Clinical+PET, Clinical+CT, PET+CT and Clinical+PET+CT, respectively). Patients were dichotomized into low- and high-risk groups by median value of prognostic score, and the significant difference between Kaplan-Meier curves of two groups was evaluated by log-rank test. The concordance index (C-index) was used to evaluate the prognostic performance of each model. Results: At the end of follow-up, 60 patients encountered events (28 recurrence, 17 metastasis and 15 death). M stage (hazard ratio-HR: 2.96-3.76, p: 0.000-0.007), SUVvar_HHL from PET (HR: 1.54-2.32, p: 0.000-0.003) and Contrast_TFCM_LHL_64 from CT (HR: 0.17-0.48, p: 0.000-0.015) were found to be consistently retained as independent prognostic factors in the seven models. In the training cohort, models integrating PET and/or CT features to clinical features (C-index: 0.80-0.89) improved the prognostic performance relative to those with PET or CT or clinical features alone (C-index: 0.68-0.83). Similar results were also found in the validation cohort (C-index: 0.68-0.81 vs 0.58-0.78). Conclusions: Radiomic features extracted from the PET and CT components of baseline 18F-FDG PET/CT provide complementary prognostic information and improve outcome prediction for NPC patients compared with clinical parameters. 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. ER -