ESTRO 2020 Abstract Book

S406 ESTRO 2020

generated by first eliminating redundant RFs, then by backward selection of best predictive parameters and finally by bootstrap-based validation. Predictive indices based on the COX models for OS, LRFS and DRFS were derived as the cut-off values best discriminating pts with the event (death or relapse) at the time corresponding to the 1 st quartile (6, 5, and 2,5 months from RT start for OS, LRFS and DRFS respectively).

Results Two-hundred and five RF were extracted by using a previously validated semi-automatic SUV gradient-based method and 133 out of them were excluded due to poor repeatability/reproducibility, mostly due to delineation uncertainty. Among 72 selected robust RF only 7 were conserved after redundancy filtering; the resulting COX multivariable models for OS, LRFS and DRFS are reported in Table 1 with/without including clinical variables; Kaplan Mayer curves including only RF are shown in Figure 1. Models based on few (i.e.: one-two) independent RF were able to clearly stratify pts in risk classes for all end-points. In particular, a surprisingly high ability in stratifying pts according to the risk to develop distant metastasis (HR= 0.24) or local relapse (HR=0.38) was found, as reported in Figures 1B-C: two first order RF (COM_shift and percentile10) and one texture RF (small-zone-emphasis) were the best RF predictors for DRFS and LRFS respectively. As shown by p values and hazard ratios in Table 1, the addition of clinical factors (pre-RT GICA/stage for OS/LRFS) did not significantly improve the models performances. Bootstrap based validation showed that the results were sufficiently robust. Conclusion Few robust PET-RF carefully predicted the outcome of LAPC pts after chemo-RT. External validations of the models are in progress on an independent group of patients treated in the same Institute and in another large group treated in additional three Institutes. If confirmed, current results could have dramatically positive impact on LAPC personalized treatment. PH-0718 Quantitative MRI in prognosticating clinical outcomes in carcinoma cervix treated with Radiotherapy S. Mitra 1 , A. Jajodia 2 , V.P.B. Koyyala 3 , V. Mahawar 4 , A. Dewan 5 , S. Aggarwal 6 , I. Singh Wahi 6 , S. Barik 7 , K. Dobriyal 8 , J. Mukhee 8 , H. Khurana 9 , R. Tripathy 9 , A. Rao 10 , A. Chaturvedi 11 1 Rajiv Gandhi Cancer Institute & Research Centre, Senior Consultant Radiation Oncologist, Rohini- Delhi, India ; 2 Rajiv Gandhi Cancer Institute & Research Centre, Senior Resident Radiologist, Rohini- Delhi, India ; 3 Rajiv Gandhi