ESTRO 2020 Abstract Book

S299 ESTRO 2020

mg/m 2 /d from day -14 to the end of RT. Most patients (84%) received a 5-FU dose > 90%. The original protocol was seldom modified, based on oncologist’s preference: 59 pts received OXA three times, as planned, while in 36 pts the OXA dose at day +14, was not administered. For 82 patients, T2 weighed MRIs taken at planning and at half- RT were available and GTVs were respectively contoured (V pre and V half ): ERI TCP (equal to ERI TCP = -ln[(1 – (V half / V pre )) Vpre ], was calculated. The discriminative power of ERI TCP (in terms of AUC, sensitivity/specificity, positive/negative predictive value) was separately assessed for the two patients groups with different OXA regimen (2 vs 3 OXA cycles), by considering as end points both the clinical/pathological complete response (cCR&pCR) and the “limited response” (defined by a residual vital cells-RVC >10%). The impact of OXA and of selected clinical parameters on cCR&pCR was also investigated trough multivariate logistic regression. Results Complete data were available for 82 patients (21 pCR and 2 cCR not followed by surgery), 53 pts treated with 3 OXA cycles and 29 pts with 2 cycles: the percentage value of cCR&pCR was 32% and 14% in the two groups, respectively while ERI TCP was not significantly different. The discriminative power of ERI TCP was moderately high, both for cCR&pCR (AUC= 0.74; sensitivity=78%; PPV= 49%; NPV=89%) and for RVC>10% (AUC= 0.70; sensitivity=79%; PPV=55%, NPV=85%). The power increases by considering the patients treated with three OXA cycles (AUC=0.78/0.78; sensitivity=79%/92%, PPV=60/43%, NPV=86/96% for cCR&pCR/RVC>10%). In a logistic multivariate regression (p=0.0001; H&Ltest=0.70), ERI TCP (OR= 0.90, 95%CI=0.88-0.98) and the number of OXA cycles (OR= 4.3, 95%CI=1.22-15.3) were the only two independent variables predictive of cCR&pCR. In Figure 1 the effect of chemotherapy dose (3 vs 2 OXA cycle) on cCR&pCR prediction is plotted vs ERI TCP . Conclusion Current results confirmed ERI TCP as a promising index in predicting pathological response after neo-adjuvant RT for rectal cancer and clearly quantify the heavy detrimental effect of avoiding OXA in the last part of RT.

Z. Shi 1 , C. Zhang 1 , I. Compter 1 , M. Verduin 2 , A. Hoeben 2 , D. Eekers 1 , A. Dekker 1 , L. Wee 1 1 GROW School for Oncology and Developmental Biology- Maastricht University Medical Centre+, Department of Radiation Oncology MAASTRO CLINIC, Maastricht, The Netherlands ; 2 GROW School for Oncology and Developmental Biology- Maastricht University Medical Centre+, Department of Medical Oncology, Maastricht, The Netherlands Purpose or Objective We propose a pooling-based quantitative imaging feature selection method and showed how it would be applied to the clinical question of predicting two-year survival of glioma patients treated by radiotherapy (RT). Material and Methods Data from 130 patients with pathologically confirmed glioma treated at a single RT centre between January 2004 and December 2014 were collected. All patients only received a biopsy prior to high-dose RT with temozolomide or RT only. Follow-up consisted of quarterly clinical consultations including MRI examination, until death from any cause. MRI and CT images were co-registered and a gross tumour volume (GTV) was manually delineated by an experienced radiation oncologist. RT dose and radiomic features were calculated on helical CT (Siemens, Erlangen, Germany) with 0.98 mm by 0.98 mm pixels, 1mm reconstructed slice thickness and 120 kVp tube potential. A total of 1092 radiomic features were extracted from the GTV via an open-source radiomics package O-RAW that is an extension wrapper of PyRadiomics. The feature selection procedures consisting of 5 steps were shown in Figure 1: (1) split samples into in-train and in-test sets with label adjust by SMOTE approach. (2) feature dimension reduction by calculating correlation with tumour volume, feature pair-wise correlation, and Kolmogorov-Smirnov test. (3) LASSO was used to build feature pool. (4) recursive feature elimination was used to build signature pool using logistic regression. (5) optimal model selection. The AUC and Hosmer-Lemeshow test statistic were determined to assess the model discrimination and calibration, respectively.

Results At two year following RT, there was similar proportion of surviving to deceased patients on the in-train and in-test subsets). After implementing the proposed feature selection process for survival prediction, a signature was developed with 6 radiomic features. The signature achieved AUCs of 0.87 (95: CI:0.80-0.93) and 0.84 (95: CI:0.71-0.88) in the in-train and in-test subsets (Figure 2). The difference in AUCs between subsets were not statistically significant, as can be expected from the highly-overlapping confidence interval estimates. The Hosmer-Lemeshow test of the two-year survival prediction model yielded non-significant statistics (p=0.10 and p=0.22), indicating that deviation of model prediction from observed outcome was not statistically significant.

Poster discussion: PH: Radiobiological and predictive modelling, and radiomics 2

PD-0540 A pooling method for feature selection of quantitative imaging analysis: survival analysis in glioma