ESTRO 2021 Abstract Book

S1040

ESTRO 2021

Distant metastasis is the main cause of treatment failure in locally advanced rectal cancer (LARC) patients (pts), despite the improvement in the treatment strategies obtained in the last years. This study aims to evaluate the “delta radiomics” approach in pts undergoing neoadjuvant chemoradiotherapy (nCRT) treated with magnetic resonance-guided radiotherapy (MRgRT), by developing a logistic regression model able to predict 2-years disease-free-survival (2yDFS) after nCRT in LARC pts. Materials and Methods All the pts enrolled in this retrospective study underwent MRgRT following a simultaneous integrated boost (SIB) technique with 45Gy in fractions of 1.8Gy delivered on the whole mesorectum and the drainage nodal stations (according to disease stage) and a boost of 55Gy in fractions of 2.2Gy on Gross Tumor Volume (GTV) plus corresponding mesorectum. Concomitant chemotherapy with capecitabine/5-fluorouracil or an intensification schedule with oxaliplatin was prescribed, in relation to clinical stage and high risk factors. For each pts, 6 T2*/T1 MR images, acquired during MRgRT at simulation and at fractions 5, 10, 15, 20 and 25 using the on-board 0.35T MR scanner, were considered for the analysis. Overall, 90 radiomic features (RF) belonging to three families (morphological, statistical, textural) were extracted from each MRI considering the GTV as region of interest. The variations of RF during treatment were quantified calculating the delta RF, which were the ratio between the values calculated at different dose levels and the one extracted at simulation. The Wilcoxon–Mann–Whitney test was performed to identify the features, which were predictive of 2yDFS at the univariate analysis. Pearson Correlation Coefficient was used to estimate the correlation among the significant features. A logistic regression model considering two delta RF was finally calculated, evaluating its performance in terms of Receiver Operating Characteristic curve (ROC). Results Data regarding 48 LARC patients were collected and reported in table 1. At the univariate analysis, a total of 65 image features resulted to be significant in identifying 2yDFS. The logistic regression model was built considering two textural features: the co-occurrence image contrast calculated at simulation and the variation of the run-length percentage feature calculated at fraction 15. The ROC curve obtained is shown in Figure 1: an Area under curve (AUC) value of 0.91 (0.81-1) was obtained, with a sensitivity of 0.88 and a specificity of 0.91 at the best discriminative threshold. At a median follow up of 31 (range 4-47) months, 10 pts developed metastases (20.8%). 990 image features were extracted for each patient (540 radiomic and 450 delta RF).