ESTRO 2020 Abstract Book

S362 ESTRO 2020

across packages was assessed using the intraclass correlation coefficient (ICC). The ICC and confidence intervals (CI) were stratified to indicate poor (ICC CI < 0.5), moderate (0.5 < ICC CI < 0.75), good (0.75 < ICC CI < 0.9) and excellent (ICC CI > 0.9) agreement. The agreement between features using default package parameters versus setting the same parameters (64 grey levels, Chebyshev distance 1) for the IBSI-compliant packages was also assessed. The impact of different packages on the prognostic value of the 17 radiomic features was investigated by comparing univariable cox models predicting survival for patients with H&N cancer. The Benjamini-Hochberg method was applied. Results The agreement of features between all four packages was only excellent for volume, and the maximum and mean intensity in both datasets (Table 1). Stratification of agreement was consistent across H&N and lung cancer for 10/17 features. Where disagreement occurred the agreement tended to be worse in the H&N dataset. When comparing IBSI-compliant packages only, the agreement was improved only when the same parameters were used per package. A calculation error for sphericity was identified in one package which is in the process of being corrected by the vendor following these results.

Fig.2: Boxplots for the deviations in the DVH metrics, excluding outliers. x-axis: HMM classification, y-axis: change in DVH metric, *: p<0.05, **: p<0.01. Conclusion The HMM performs well on an external dataset considering accuracy, showing that it can be transferred between institutes. However, underestimation of categories can lead to relevant fractions not being flagged, potentially missing anatomical changes (false negatives), while overestimation leads to unnecessary flagging (false positives), thus increasing workload. Model fine-tuning may resolve this. Relating HMM classification based on γ features to increasing DVH differences is possible for some OARs, but not for the target volumes. PH-0652 Standardization influences repeatability and prognostic value of radiomic features I. Fornacon-wood 1 , J. O'Connor 1 , C. Faivre-Finn 1 , G. Price 1 1 The University of Manchester, Division of Cancer Sciences, Manchester, United Kingdom Purpose or Objective Radiomic features have shown potential as prognostic biomarkers for personalizing radiotherapy treatment. However, standardization issues hinder clinical translation. The image biomarker standardization initiative (IBSI) makes recommendations to address these issues. In this work we assessed the benefit of IBSI compliance for radiomic feature calculation packages in lung and head and neck (H&N) cancer datasets. The impact of package choice on predicting survival for patients with H&N cancer was also investigated. Material and Methods Four widely used radiomics software packages were evaluated: LIFEx, PyRadiomics and CERR (all of which are IBSI compliant) and IBEX (not IBSI-compliant). 17 radiomic features common to all packages were calculated on the planning CT scan for the GTV of small cell lung cancer (N=37) and H&N cancer (N=111) patients treated with radiotherapy (Table 1). Agreement between features

6 texture features were significant (p<0.05) for survival when calculated in LIFEx, PyRadiomics and CERR, but not IBEX. Figure 1 presents Kaplan-Meier analysis for the feature GLCM joint entropy calculated in each package, split on the median. The risk stratification is significant for GLCM joint entropy calculated in LIFEx, PyRadiomics and CERR (p=0.01, 0.013, 0.0095 respectively), but not in IBEX (p=0.64).

Conclusion IBSI-compliance improves feature repeatability between software packages, given the same parameters are used, and this improves prognostic model quality. It is important