ESTRO 2020 Abstract Book

S842 ESTRO 2020

treatment, were enrolled as the training and validation cohorts. PET/CT images were first transformed into PM images by the proposed algorithm, then the single modality based radiomics workflow was applied. A radiomics signature was constructed with the least absolute shrinkage and selection operator based Cox regression from the training cohort and verified in the validation cohort. Then a nomogram combining the radiomics signature and clinical factors was built on the whole cohort. Finally, the performance of the radiomics signature and nomogram to predict survival status was evaluated on receiver operating characteristic (ROC) The radiomics signature built from the training cohort was associated with OS in the validation cohort ( p <0.001), and its value showed no significant difference in Mann-Whitney U test ( p =0.471). The C-indexes of the radiomics signature to predict OS in the training and validation cohorts were 0.767(95% CI, 0.759 to 0.773) and 0.743 (95% CI, 0.737 to 0.752) respectively. Due to the combination of clinical factors, the nomogram on the whole cohort showed a strong association with OS, which achieved a C-index of 0.814 (95% CI, 0.809 to 0.821). The ROC curves of survival status estimation further demonstrated that radiomics signature and nomogram could be used in clinical. Conclusion The PM image-based radiomics workflow could obtain a stable and reproducible result in multimodal images and has potential in clinical practice. curves. Results PO-1556 4D-PET radiomics-features for radiotherapy treatment monitoring in locally advanced NSCLC patients M. Carles 1 , T. Fechter 1 , M. Benndorf 2 , G. Radicioni 3 , T. Schimek Jasch 3 , S. Adebahr 3 , D. Baltas 1 , A.L. Grosu 3 , M. Mix 4 , U. Nestle 5 , E. Gkika 3 1 Medical Center- Faculty of Medicine- University of Freiburg, Division of Medical Physics- Department of Radiation Oncology, Freiburg, Germany ; 2 Faculty of Medicine- University of Freiburg, Department of Radiology, Freiburg, Germany ; 3 Medical Center- Faculty of Medicine- University of Freiburg, Department of Radiation Oncology, Freiburg, Germany ; 4 Medical Center- Faculty of Medicine- University of Freiburg, Department of Nuclear Medicine, Freiburg, Germany ; 5 Kliniken Maria Hilf GmbH Moenchengladbach, Department of Radiation Oncology, Moechengladbach, Germany Purpose or Objective The use of PET radiomics-features (RF) in treatment monitoring requires the identification of RF changes large enough to be considered statistically and clinically relevant. For patients with non-small cell lung cancer (NSCLC) the evaluation of the primary tumor using PET/CT RF presents challenges due to respiratory movement. Respiratory gated (4D)-PET/CT improves image quality and quantification accuracy by compensation of respiratory motion effects. Our primary aim was to assess the use of PET RF changes along the treatment as a predictor of the local recurrence (LR) in NSCLC patients. For this purpose and in order to minimize the impact of RF variability due to the noise, resolution and lesion movement inherent in PET, we evaluated the implementation of the RF variability observed across of 4D-PET phases (standard deviation, σ RF ), as a patient individualized correction factor which emphasizes statistically relevant RF changes across treatment. Material and Methods Patients with histologically proven locally advanced NSCLC were prospectively treated in the PET-Plan Trial (ARO 2009-09; NCT00697333) using concurrent chemoradiotherapy. A total of 47 patients additionally

underwent a PET/CT during radiotherapy (RT) treatment, resulting in 3 PET-images: a 4D and a conventional (3D) before and a 3D during RT. In this analysis patients (n=15) for which all PETs were performed with GEMINI-TF-TOF-64 system and with 4mm voxel resolution were involved. Primary lesions were delineated on PET using an automatic contrast-oriented-algorithm. For each PET, 135 RF were computed. The study involved (figure 1): (i) the determination of σ RF for the RF following a normal distribution across 4D-PET phases; (ii) the identification of the RF robust to the PET/CT reconstruction protocol, i. e. comparable between 3D and 4D; (iii) the assessment of the RF for which their variation (δ RF =Δ RF /σ RF ) was able to discriminate patients with and without LR and (iv) the development of LR model. Comparisons were analysed with Wilcoxon-signed-rank-test in (ii) and two-tailed Mann- Whitney-U-test for non-pairwise testing in (iii). Model development involved imbalance-adjusted bootstrap resampling for RF selection, prediction performance estimation and model coefficients computation. Results 47 RF followed a normal distribution across breathing phases. Their mean µ and σ RF were dependent on RF and patient: σ RF /µ ranging from 0.01 to 176%. 68 RF were comparable between 3D and 4D images. The best performance for the discrimination of patients with LR was obtained for 4 strong linear correlated texture features. The model for the prediction of LR resulted in an area- under-the-curve of 0.90 and specificity of 0.83 for the δ RF of the texture feature Zone-percentage after uniform- quantization.

Conclusion We presented the first evaluation of 4D-PETRF variability as individualized criteria to identify clinically relevant RF changes along RT-treatment for NSCLC. An enlarged cohort is required to confirm the LR prognosis findings. PO-1557 Findable, Accessible, Interoperable, Reusable (FAIR) Quantitative Imaging Analysis Workflow Z. Shi 1 , A. Fedorov 2 , A. Hosny 3 , C. Parmar 3 , H. Aerts 2,3 , L. Wee 1 , A. Dekker 1 1 GROW School for Oncology and Developmental Biology-