ESTRO 2021 Abstract Book

S1531

ESTRO 2021

Conclusion Our results suggest that deep learning methods can predict pCR after NACRT with an accuracy of 68%. External validation is needed before clinical adoption in the clinic.

PO-1803 Voxel-wise quantification of anatomical tumor lung location is associated with overall survival D. Vuong 1 , M. Bogowicz 1 , L. Wee 2 , O. Riesterer 3 , E. Vlaskou Badra 1 , L.A. D'Cruz 4 , P. Balermpas 1 , J.E. van Timmeren 1 , S. Burgermeister 1 , A. Dekker 2 , D. de Ruysscher 2 , J. Unkelbach 1 , S. Thierstein 5 , E.I. Eboulet 5 , S. Peters 6 , M. Pless 7 , M. Guckenberger 1 , S. Tanadini-Lang 1 1 University Hospital Zurich and University of Zurich, Department of Radiation Oncology, Zurich, Switzerland; 2 GROW School for Oncology and Developmental Biology, Maastricht University Medical Centre+, Department of Radiation Oncology (MAASTRO), Maastricht, The Netherlands; 3 Kantonsspital Aarau KSA, Radioonkologie-Radiotherapy, Aarau, Switzerland; 4 Universitätsklinikum Frankfurt, Strahlentherapie und Onkologie, Frankfurt, Germany; 5 Swiss Group for Clinical Cancer Research (SAKK), Coordinating Center, Bern, Switzerland; 6 Centre Hospitalier Universitaire Vaudois (CHUV), Department of Oncology, Lausanne, Switzerland; 7 Kantonsspital Winterthur, Department of Medical Oncology, Winterthur, Switzerland Purpose or Objective The anatomical location and extent of the primary lung tumor (PT) has shown prognostic value for outcome prediction; however, its manual assessment is prone to interobserver variability. This study aims to evaluate the feasibility to predict overall survival (OS) using the PT distance to anatomical regions based on cumulative status maps for locally advanced non- small cell lung cancer (NSCLC) patients. Materials and Methods Five stage IIIA/IIIB NSCLC patient cohorts were retrospectively collected, treated either with radiochemotherapy (RCT): RCT1* (n=107), RCT2 (n=85), RCT3 (n=32) or surgery: S1* (n=134), S2 (n=37). Pre-treatment CT scans were registered to a reference patient using an ipsilateral lung contour-only based deformable image registration and the ipsilateral main bronchus as an anatomical landmark (MIM Vista, 6.9.2.). An in-house software was developed to transfer each PT to the reference patient while maintaining the original PT shape. A frequency-weighted cumulative status (fwCS) map was created for both exploratory cohorts (indicated with asterisk), where PT location was uni-labeled with 2 years OS patient status (Fig.1).

Fig.1: Identification of decreased survival areas and extraction of the primary tumor closest distance. Based on the frequency weighted cumulative status (fwCS) map, a permutation test was performed to identify areas with significantly worse OS, from which the closest distance of a primary tumor (blue) was calculated. To identify regions with significantly worse OS, a permutation test (500 repetitions with resampling) was performed by randomly assignment of survival/death to PT of exploratory cohorts. A voxel was considered significant if its test statistic (mean/standard deviation) was larger than 95% of the test statistics based on the permutation test (referred to as decreased survival areas (DSA)). The closest PT-DSA Euclidean distance was extracted (negative distance was assigned in