ESTRO 2021 Abstract Book

S672

ESTRO 2021

Figure. Regression curve showing association between tumor volume and the probability cCR.

Conclusion In circular cancers or with ≥7 cm length, cCR frequency and the sensitivity of cCR diagnosis are very low.

PD-0837 Delta radiomic features and early regression index for rectal cancer response prediction in MRgRT L. Boldrini 1 , D. Cusumano 1 , G. Chiloiro 1 , P. Yadav 2 , G. Yu 3 , A. Romano 1 , A. Piras 4 , L. Placidi 1 , S. Broggi 5 , L. Indovina 1 , M.A. Gambacorta 1 , M.F. Bassetti 2 , Y. Yang 6 , C. Fiorino 7 , V. Valentini 1 1 Fondazione Policlinico Universitario "A. Gemelli" IRCCS, Radiology, Radiation Oncology and Hematology, Rome, Italy; 2 School of Medicine and Public Health, University of Wisconsin-Madison, Department of Human Oncology, Madison, USA; 3 University of California , Department of Radiological Sciences, Los Angeles, USA; 4 Villa Santa Teresa - Centro San Gaetano, Radioterapia, Bagheria, Italy; 5 San Raffaele Scientific Institute , Medical Physics, Milan, Italy; 6 University of California, Department of Radiological Sciences, Los Angeles, USA; 7 San Raffaele Scientific Institute, Medical Physics, Milan, Italy Purpose or Objective Early Regression Index (ERI) is a radiobiological parameter that showed promising results in predicting pathological complete response (pCR) on MR images of patients affected by Locally Advanced Rectal Cancer (LARC). Its predictive performance ranges from 0.81 to 0.93 in terms of area under (AUC) Receiver Operating Characteristics (ROC) curve. The aim of this study is to investigate the delta radiomic role in improving discriminative power to ERI. Materials and Methods 59 patients were enrolled from 3 centres and treated with neoadjuvant chemoradiotherapy delivered with MR- guided Radiotherapy (MRgRT) units, followed by surgery. Centre 1 (40 pts) delivered 55Gy in 25 fractions (fx), centres 2 (8 pts) and 3 (7 pts) delivered 50.4 Gy in 28 fx. For each patient, a 0.35T T2*/T1 MRI was acquired during simulation and at each treatment fraction. Biologically effective dose (BED) conversion was used to compensate for the different schemes (α/β=10). pCR was defined as ypT0ypN0 at the pathological specimen. GTV was delineated on the MR images corresponding to BED levels of 0,13,26,40,53 and 59Gy and radiomic analysis was performed. 90 radiomic features were extracted from each MR, and delta radiomic features were calculated with respect the simulation MR(BED=0Gy). Univariate analysis was evaluated using Wilcoxon Mann Whitney test and Benjamini–Hochberg method to adjust for multiple comparisons. Pearson Correlation Coefficient (PCC) was used to estimate the correlation among the significant features. Two logistic regression models were finally generated: the first one considering the most significant feature at the univariate analysis and the second one combining this feature with the one showing the PCC closest to 0. ROC curves were calculated for both models and a direct comparison was performed using the Delong test. The reliability of the models was evaluated using a 5 folds cross-validation analysis with 3 iterations. Results pCR was observed in 14 cases. A total of 540 radiomic and 450 delta radiomic features were calculated for each case. The most significant feature was ERI with a p=1.33∙10 -4 and it was used for the first logistic model. The feature showing the PCC closest to 0 was the skewness calculated at the 3 rd week: this feature was combined with ERI to obtain the 2-variables model. DeLong test reported p=0.15 in comparing the 2 ROC curves. Figure 1 reports the ROC curves obtained and table 1 their predictive performance. Figure 1.