TY - JOUR T1 - <strong>A metabolic signature to predict the location of second tumor recurrence after re-irradiation in head and neck cancer </strong> JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 3113 LP - 3113 VL - 63 IS - supplement 2 AU - Arnaud BEDDOK AU - Valentin Calugaru AU - Christophe Nioche AU - Laurence Champion AU - Gilles Crehange AU - Irene Buvat Y1 - 2022/06/01 UR - http://jnm.snmjournals.org/content/63/supplement_2/3113.abstract N2 - 3113 Introduction: Curative reirradiation (reRT) is a promising alternative for the treatment of local recurrence (LR) of head and neck cancer (HNC). However, up to 50% of patients may experience a second LR within two years after the end of reRT. The prediction of the location of this second LR would allow adapting the reRT plan with a dose escalation in the highest risk areas for LR. The aim of our study was to evaluate whether radiomics from [18F]-FDG PET imaging could predict the location of this second LR.Methods: Among the 54 patients re-irradiated with curative intent from 2007 to 2019 in Curie Institute for advanced HNC, 31 had a [18F]-FDG PET before the start of the reRT, including 24 patients who experienced a second LR. We first divided this cohort into a discovery cohort (14 patients) and a validation cohort (10 patients). We first analyzed the discovery cohort data. For each patient, the reirradiated GTV was segmented based on the hypermetabolism on PET defined as SUV greater than 5. Twenty-nine features including three standardized uptake values (SUVs), five first-order statistics derived from the gray-level histogram, and 31 textural features were extracted from this GTV using the LIFEx software after spatial resampling: 2 x 2 x 2 mm, and fixed bin size of 0.157 SUV units for intensity discretization. The second LR, identified on PET, were categorized as “in-field”, “marginal”, or “outside” if 100%, ≤ 50%, or &lt; 20% of their volume was within the 95% isodose of the first recurrence GTV, respectively. Student t.tests were used to compare the feature values of the tumors that later had “in-field” and “outside” LR. Correlograms were calculated to identify potential associations between features that were significantly different between the two groups. Characterizing each patient by the values of features that were found to be significantly different between the two groups, principal component analysis (PCA) and Ascending Hierarchical Classification (AHC) were performed. Then, using the feature values of the patients from the validation cohort calculated with the same procedure as described for the discovery cohort, the site of LR was predicted in a blinded procedure based on the findings made on the discovery cohort, and verified afterward.Results: Among the 14 patients of the discovery cohort, seven had “in-field” and seven had “outside” second LR, respectively. The value of three histogram (SUV_min, SUV_Kurtosis, SUV standard deviation [SUV_std]) and two texture features (GLCM_Correlation, GLCM_Contrast) were significantly different between the two groups (p&lt;0.05). The five identified features were not highly correlated (Pearson coefficient &lt;0.80). In PCA, the in-field and outside LR patients were well separated in a 2D-space S spanned by the first two principal components (p&lt;0.02). “In-field” group corresponded to regions with high SUV_std and low Kurtosis, both features being associated with substantial signal heterogeneity in the reirradiated GTV. “Outside” group was associated with low GLCM_Correlation, std, and GLCM_Contrast, suggesting rather homogeneous signal in the reirradiated GTV. The AHC analysis also found a cluster with high std, GLCM_Contrast, GLCM_Correlation, and low Kurtosis, SUV_min, corresponding to patients with “in-field” second LR. The projection of new patients in 2D-space S allowed us to infer the location of the second LR for the validation cohort. Doing so, the location of the second LR was correctly predicted for 5/6 (83%) patients who had an outside recurrence and for 3/4 patients (75%) who had an in-field recurrence (figure 1). The probability to get such an accuracy randomly was 0.12. Conclusions: This study suggests that before reRT, [18F]-FDG PET radiomic features characterizing the signal heterogeneity in the GTV to be re-irradiated have different values in patients who will relapse “in-field” or “outside”, with “in-field” relapses associated with higher metabolic heterogeneity in the GTV than relapse occurring outside the GTV. ER -