ESTRO 2021 Abstract Book

S1162

ESTRO 2021

of R1 resection in those patients receiving adjuvant RT who required reintervention, with the frequency distribution of toxicities being similar. Further phase III studies are needed to determine the differences in efficacy of the treatments.

PO-1415 Association of radiomic features with aggressive phenotypes in soft tissue sarcomas A. Datta 1,2 , L. Forker 1,3 , A. McWilliam 1 , H. Mistry 1 , J. Zhong 4 , J. Wylie 3 , C. Coyle 3 , D. Saunders 3 , S. Kennedy 3 , J. O'Connor 1,2,5 , P. Hoskin 1,3,6 , C. West 1 , A. Choudhury 1,3 1 University of Manchester, Division of Cancer Sciences, Manchester, United Kingdom; 2 The Christie Hospital, Clinical Radiology, Manchester, United Kingdom; 3 The Christie Hospital, Clinical Oncology, Manchester, United Kingdom; 4 The University of Leeds, Leeds Institute of Medical Research, Leeds, United Kingdom; 5 Institute of Cancer Research, Division of Radiotherapy and Imaging, London, United Kingdom; 6 Mount Vernon Cancer Centre, Clinical Oncology, Northwood, United Kingdom Purpose or Objective Soft tissue sarcomas(STS) are rare and heterogeneous tumours with variable outcomes. Improving survival requires identifying and targeting aggressive phenotypes. Novel ways of stratifying patients include a clinical prognostic nomogram(Sarculator; includes tumour size) and a gene expression derived hypoxia score(HS). We investigate the feasibility of non-invasive and repeatable imaging to assess clinically/biologically relevant and Retrospective analysis of 43 extremity STS patients with matched diagnostic biopsy-imaging data was performed. Imaging was acquired at several different hospitals in the region using various scanners/protocols. Patients underwent curative-intent surgery±adjuvant radiotherapy. HS(24-gene signature) were measured using NanoString. Sarculator predicted 10yr OS. Treatment naïve T 1 (n=41) and T 2 (n=28) weighted sequences were segmented by two radiologists using Raystation. Histogram normalisation and gray-level intensity discretisation steps were performed. PyRadiomics v3.0.1 was used to extract features and robustness was assessed using intra-class correlation(ICC; threshold >0.9). Features with a high degree of association within their classes were further selected using Spearman’s rank correlation. Associations with Sarculator and HS were determined using rank correlation matrices and principal component analysis(PCA). Significance levels were set at p<0.05. Results ICC identified 63(T 1 ) and 68(T 2 ) features. Further selection resulted in 4(T 1 ) and 14(T 2 ) exploratory features. Sarculator correlated strongly with T 1 ( ρ=-0.75) and T 2 ( ρ=-0.84) volume features ( Fig 1 ). T 1 size( ρ=0.44) correlated strongest with HS . Top T 2 features, gray-level non-uniformity(GLN) and zone entropy(ZE), correlated with Sarculator( ρ=-0.57 , ρ=-0.56 respectively) and with hypoxia ( ρ=-0.37,ρ=0.39 respectively ). GLN is a gray-level run length matrix (GLRLM) feature quantifying variability of gray-level intensity. ZE is a gray- level size zone matrix feature quantifying randomness in distribution zone sizes and gray levels. High GLN values indicate more heterogeneity in intensity; high ZE values indicate more heterogeneity in texture. PCA identified clusters using the patient radiomics values, and box plots highlight differences ( Fig 2 ). T 1 derived features were significantly different between the 3 groups for Sarculator(p=0.013) but not HS(p=0.156). There were no significant differences identified by T 2 derived features for Sarculator(p=0.088) or HS(p=0.676). targetable phenotypes. Materials and Methods

Fig 1:Rank correlation matrix of selective T 2

-features, along with HS and Sarculator.