ESTRO 2021 Abstract Book

S413

ESTRO 2021

Conclusion Deep learning DM time-to-event models showed good overall accuracy and could be used for RT treatment personalization. The binary masking experiment suggested that image texture features are less relevant for good CNN performance. OC-0527 Tumour Repopulation Varies Between Sequential vs. Concurrent Chemoradiation in Lung Cancer H. Huang 1 , M.G. Nix 2 , J.D. Fenwick 3,4 , M.A. Hawkins 1 1 University College London, Faculty of Engineering, Department of Medical Physics and Biomedical Engineering, London, United Kingdom; 2 Leeds Teaching Hospitals NHS Trust, Leeds Cancer Centre, Department of Medical Physics and Engineering, Leeds, United Kingdom; 3 University of Liverpool, Institute of Translational Medicine, Department of Molecular and Clinical Cancer Medicine, Liverpool, United Kingdom; 4 Clatterbridge Cancer Centre, Department of Physics, Wirral, United Kingdom Purpose or Objective Chemoradiotherapy is the standard of care for treating inoperable locally-advanced non-small cell lung cancer (LA-NSCLC). We aim to investigate new models of survival following treatment for LA-NSCLC incorporating the effects of chemotherapy. Materials and Methods In the baseline tumour control probability (TCP) model, chemotherapy effects had been included via a difference in maximum overall survival ( OS Max _by Treatment ) between radiotherapy and any form of chemoradiotherapy, and a radiosensitisation factor for concurrent chemoradiotherapy (cCRT) ( RS_cCRT ). In our improved model, we separated tumour repopulation parameters ( λ , repopulation rate ; T k , repopulation onset time) by treatment-type to test for differences in repopulation between radiotherapy alone, sequential chemoradiotherapy (sCRT), and cCRT. Model fitting was done using maximum-likelihood estimation, comparing fits using the Akaike Information Criteria and likelihood ratio test. LA-NSCLC dose-response curves for radiotherapy, sCRT, and cCRT were plotted for a fixed treatment duration of 40 days. Results Population data were collated from prospective trials and institutional series with >20 cases published between 2000 – 2016. The data comprised 4866 NSCLC patients receiving radiotherapy (1192 patients), sCRT (1792 patients), or cCRT (1882 patients), with doses varying from 1.26Gy/Fx - 3.00Gy/Fx over 16 - 60 days. In the optimised model, accelerated tumour repopulation began at 16 days for radiotherapy (95% CI: 16 - 26), 25 days for sCRT (95% CI: 16 - 34), and 34 days (95% CI: 16 - 39) for cCRT, and respectively accounted for 0.37Gy (95% CI: 0.20 - 0.49), 0.34Gy (95% CI: 0.24 - 0.58) and 0.92Gy (95% CI: 0.54 - 1.43) radiation dose lost per day of treatment protraction.