The main objective of this study is to identify candidate imaging biomarkers from FDG PET/CT scans for progression-free survival (PFS) in patients with locally advanced head and neck squamous cell carcinoma (HNSCC) enrolled in the phase III RTOG 0522 study. It is hypothesized that quantitative FDG PET/CT imaging biomarkers increase prognostic value compared to the primary tumor and nodal staging. Patient selection RTOG 0522 is a completed phase III clinical trial that showed no improvement in progression-free survival (PFS) by adding cetuximab to chemoradiation in patients with HNSCC. This present study investigates the utility of FDG PET/CT scans for patients with unresected N2abc-N3 squamous cell cancers of the head and neck. Image processing FDG PET/CT scans were reviewed for suitability of analysis. Regions of interest (ROI), enclosing the primary tumor and lymph nodes, were drawn manually and then grown isotopically up to the boundary using semiautomatic segmentation algorithm. After processing, ROIs were visually verified. ROIs were separated into three: primary tumor, right, and left nodes. Post-treatment scans were deformably registered to pre-treatment scans. Pre-treatment, post-treatment and changes in the FDG uptake parameters were evaluated. Radiomics-based imaging feature extraction was performed using 5 x 5 x 5 voxel patches inside the ROIs. A total of 50 texture radiomics features were calculated for each one of the three groups of lesions. Statistical analysis Primary tumor and nodal FDG PET texture features were evaluated at pre-treatment (PRE), post-treatment (POS), and change between the two scans (CHG). Nodal features were evaluated as maximum (MN) and sum (SN) of left and right nodes. Features were analyzed as continuous (C) and dichotomized (D) by median for T, MN and SN. Variables were correlated to PFS using Cox models; models were compared by Akaike information criterion (AIC). RTOG 0522 randomized 940 patients (891 eligible); 116 participated in the PET/CT sub-study; 50 patients with both analyzable FDG PET/CT scans were included in analysis. For 22 texture features, models have better fit than T or N stage alone. In univariate analysis, the best fit was for primary PRE FDG PET dichotomized skewness [hazard ratio (HR) and 95% confidence interval (CI) 0.37 (0.16-0.84)]. Adjusted for T stage the best fit was for POS FDG PET MN continuous kurtosis [HR 0.75 (0.54-1.06) for 10% increase]. Adjusted for N stage the best fit was for dichotomized PRE FDG PET SN sum of squares variance [HR 3.29 (1.28-8.51)]. Twenty-two texture features derived from the pre- and post FDG PET/CT scans provided prognostic value beyond T or N stage alone for PFS in patients with advanced HNSCC. Among these variables, many were found to be highly correlated, so further analyses would reduce the number of potential biomarkers. Given the small sample and the large number of variables, these findings are potentially supportive of our primary hypothesis.
原文链接
分享
引用
