ESTRO 2020 Abstract Book

S845 ESTRO 2020

PO-1561 Staging of non-small cell lung cancer using random forest classifiers based on radiomics S. Aouadi 1 , R. Hammoud 1 , T. Torfeh 1 , N. Al-Hammadi 1 1 National Center for Cancer Care & Research, Radiation Oncology, Doha, Qatar Purpose or Objective To predict automatically the stage of the tumor, for patients with Non-Small Cell Lung Cancer (NSCLC), based on radiomics analysis approach. Material and Methods 290 patients with pretreatment CT, manually contoured GTV, and identified stage (I, II, IIIa, IIIb) were collected from the cancer imaging archive and divided randomly into a training set (232 patients) and an external validation set (58 patients). Data were classified into two categories: stage I-II and stage IIIa-IIIb. CT scans were resampled to 1×1×1 mm³ voxels and intensities were divided into bins of 10HU from -1000 to 3000HU. 851 radiomic features were extracted from GTV’s VOI, defined on each patient CT, using Pyradiomics (v2.2.0). Four types of imaging features were identified (shape, first order, texture, wavelet). Shape (14) features describe the 3D geometric properties of the tumor. The first order (18) features describe tumor intensity distribution. Texture features, which describe the intra- tumor heterogeneity, were extracted from the gray level co-occurrence (24), gray level run length (16), gray level size zone (16), neighboring gray tone difference (5), and gray level dependence (14) matrices. Eight wavelet decompositions were obtained using a coiflet wavelet transformation; each decomposition was used as an input image to calculate the first-order statistics and the textural features described above. To tackle the problem of unbalanced training set, Synthetic Minority Over-sampling Technique (SMOTE) was used to create synthetic data making the two categories with equal number of samples. Given the high number of features in comparison to the number of patients, feature preselection was performed to reduce oversampling. Three algorithms were tested: the minimum redundancy maximum relevance (MRMR), mutual information score (MI) and Anova F-value score (AF).42 features were preselected. The classification approach was Random Forest (RF).RF’s parameters were optimized by cross- validated grid-search over a parameter grid. Three machine learning models (MLM) were built: (MRMR, RF), (MI, RF), and (AF, RF). The performance of MLM was assessed by the computation of the ROC curve, the area under the ROC curve (AUC- measure of separability between classes for any threshold), the accuracy (ACC- the fraction of correct predictions), the precision (PR- proportion of positive identifications that was actually correct) and the recall (RE- proportion of actual positives that was identified correctly). Results Table 1 displays AUC, ACC, PR and RE, on the validation set, of optimized RF for each feature selection approach. AUC, ACC, PR, RE were close to 1 for the training set. Fig 1 shows the roc curves for each combination (MRMR, RF), (AF, RF) and (MI, RF). The best performance of RF was obtained with MRMR feature selection.

[2] Ken et al, Radiat Oncol 2013 [3] Tensaouti et al, ESTRO 38, 2019 [4] Zhang et al, Medical Physics 2015 [5] Zhou et al, J. R. Statist. Soc. B (2005)

PO-1560 Contrast-enhanced CT-based radiomics nomogram predicts esophageal cancer survival after radiotherapy C. Zeng 1 , T. Zhai 1 , J. Chen 1,2 , L. Guo 1 , B. Huang 1 , G. Liu 3 , T. Zhuang 1 , W. Liu 1 , T. Luo 1 , Y. Wu 1 , G. Peng 1 , C. Chen 1 1 Cancer Hospital of Shantou University Medical College, Department of Radiation Oncology, Shantou City, China ; 2 CRUK/MRC Oxford Institute for Radiation Oncology, Department of Oncology- University of Oxford, Oxford, United Kingdom ; 3 Zhongshan City people's Hospital, Department of Radiation Oncology, Zhongshan City, China Purpose or Objective The heterogeneous outcomes of esophageal squamous cell carcinoma (ESCC) after definite chemoradiotherapy (CRT) remain a major challenge for the optimal management of this disease. To address this issue, it is pivotal to be able to predict the survival of patients after treatment. Plain CT-based radiomics has been shown to correlate with patient outcomes in a number of cancer types, including ESCC. However, the most commonly available imaging for patients underwent definitive CRT were not plain CT scans but contrast-enhanced ones which were obtained during radiation treatment planning before the start of treatment. Whether these CT scans could be used for radiomics analysis and provide a tool of predictive value in ESCC is not very clear. This study aims to evaluate the prognostic potentiality of contrast-enhanced CT-based radiomics in ESCC patients treated with definitive CRT. Material and Methods A total of 154 ESCC patients treated with definitive CRT were enrolled in this retrospective study and randomly divided into the training cohort (n=103) and the validation cohort (n=51). Pre-treatment contrast-enhanced CT scans were obtained from all patients and used for extraction of image biomarkers (IBMs). The prognostic values of pre- selected IBMs together with patient’s clinical characteristics were evaluated by univariate and multivariable Cox regression analysis. Independent factors identified from these analyses were used to generate a prognostic nomogram model. The performance of this model was assessed using the concordance index (C-index) and receiver operation characteristic (ROC) curve. Results A total of 134 IBMs were extracted from pretreatment contrast-enhanced CT scans. Of them, 3 IBMs were found to be independent predictors of survival in this cohort of patients, including Q975, Major_axis_length and Short_run_emphasis. On the basis of these IBMs, a nomogram prognostic model was developed, which had a C-index of 0.700 (95%CI, 0.636- 0.784) in the training cohort and 0.661 (95% CI, 0.565- 0.757) in the validation cohort. This model exhibited moderate, but statistically significant discrimination potential of survival status in both the training cohort (AUC 0.753, 95% CI, 0.661-0.846, p < 0.001) and the validation cohort (AUC 0.722, 95% CI, 0.580-0.864, p = 0.007). Patients divided into high risk versus low risk groups according to the nomogram model displayed significantly different 5-year overall survivals in the training cohort (89.2% vs 41.3%, P < 0.001) as well as in the validation cohort (70.9% vs 37.0%, P = 0.006). Conclusion In this study, we developed a radiomics nonogram model based on pre-treatment contrast-enhanced CT scans. This model had the ability to predict long-term survival of patients underwent definitive CRT for ESCC. Further multicenter studies with larger sample sizes are warranted.