ESTRO 2021 Abstract Book

S985

ESTRO 2021

indication of SBRT for GGA tumors should be considered adequately.

PO-1187 Overall survival prediction in NSCLC: a radiomic approach with a multi-VOI analysis S. Silipigni 1 , M. Miele 1 , S. Gentile 1 , E. Molfese 1 , P. Soda 2 , G. Iannello 2 , R. Sicilia 2 , M. Fiore 1 , S. Ramella 1 1 Campus Biomedico University, Radiation Oncology, Roma, Italy; 2 Campus Biomedico University, Engineering, Roma, Italy Purpose or Objective This study aims to predict the Overall Survival (OS) of patients with lung cancer who underwent a chemo- radiotherapy treatment using a radiomic approach. Moreover, we evaluate how an accurate segmentation impacts the results using three incremental segmentations of the same lesion. Materials and Methods For this retrospective study 97 CT images of patients with Non Small Lung Cancer (NSCLC) treated with chemo- radiotherapy were included. Images were acquired with a Siemens Somatom Emotion, with 140 Kv, 80 mAs, and 3mm slice thickness. Patients were divided into two classes according to their OS (53 patients died, 44 survived after median follow-up of 18.55 months). For each patient three segmentations were considered: the Gross Tumor Volume (GTV), the Clinical Target Volume (CTV) and the Planning Target Volume (PTV). All volumes were manually segmented by a radiation oncologist who included in the GTV all the macroscopic disease seen at CT, in the CTV the GTV plus a margin for sub-clinical disease and in the PTV, the CTV plus of a 0.5 cm safety margin. Totally 242 radiomic features: First order, 3D Gray Level Co-occurrence Matrix and Three Orthogonal Planes-Local Binary Patterns, were extracted with an in-house developed MATLAB code. For features selection, we ran a wrapper based approach using three machine learning (ML) algorithms: AdaBoost (Ada), Decision Tree (DT) and Random Forest (RF). In the classification stage, the same ML algorithms were used. Results Totally 9 different experiment were performed, combining three ML algorithms and three segmentations for each patient. Performance was evaluated through accuracy and AUC. We obtained the following results: Ada- GTV: 70.1%, 70.1%, Ada-CTV: 83.5%, 82.8%, Ada-PTV: 71.1%, 70.1%, DT-GTV: 70.1%, 70.1%, DT-CTV: 60.8%, 61.1%, DT-PTV: 63.9%, 64.1%, RF-GTV: 73.2%, 71.8%, RF-CTV: 78.4%, 77.3%, RF-PTV: 77.3%, 76.6%. Conclusion Results reported an accuracy of 83.5% and an AUC of 82.8% in predicting the OS. The best result was obtained with the intermediate segmentation, which includes margin for sub-clinical disease and increases the prediction accuracy. This result could be the first try to change the knowledge about the segmentation step in radiomics. The result showed that it is possible to predict the OS of patients giving the opportunity to change treatment options. Moreover, it showed some aspects of relevance for segmentation. PO-1188 Adjuvant Involved Field versus Elective Nodal Irradiation for N2 NSCLC: a dosimetric study M. Mohammed 1 , A. Mkanna 2 , P. Ramia 3 , D. Mahmoud 4 , L. Hilal 4 , F. Olleik 5 , M. Dagher 4 , Z. Ayoub 4 , W. Jalbout 4 , B. Shahine 6 , F. Geara 6 , B. Youssef 4 1 American university of Beirut Medical Center, Radiaton Oncology, Beirut, Lebanon; 2 Nabatyeh General Hospital, Radiation Oncology, Nabatyeh, Lebanon; 3 McGill University, Radiation Oncology, Montreal, Canada; 4 American University of Beirut Medical Center, Radiation Oncology, Beirut, Lebanon; 5 Lankenau Medical Center, Internal Medicine, Wynnewood, USA; 6 American University of Beirut Medical Center, Radiation Oncology, Beirut, Lebanon Purpose or Objective Adjuvant radiation therapy for N2 positive non-small cell lung cancer (NSCLC) has been widely used as a standard treatment up until the recent publication of the lung ART trial abstract. Results from the above trial showed no survival benefit of postoperative radiation therapy (PORT) in the adjuvant setting for N2 disease. However, PORT decreased the risk of mediastinal recurrence by around 50% but the benefit was offset by more toxic deaths due to cardiopulmonary (CP) toxicity. The lung ART protocol used wide RT fields that could possibly account for the increased CP toxicity. Conversely, involved field irradiation (IFI) is the standard treatment for unresectable stage III NSCLC with <5% risk of out of field regional recurrence. The aim of our dosimetric study was to compare elective nodal irradiation (ENI) (as per ART fields) vs. IFI for N2 NSCLC in the adjuvant setting. Materials and Methods We retrospectively reviewed the charts of 15 patients with NSCLC, treated with PORT for N2 disease in a single tertiary care center. For each patient, we generated two sets of contours for the clinical target volumes (CTV). In the first set, the CTV included the bronchial stump, and the mediastinal lymph node (LN) stations at risk as defined by the lung ART protocol (ENI). The second set CTV included the bronchial stump and the involved LN stations only (IFI). 3D-Conformal Radiation Therapy treatment plans were generated for the two sets using Panther software (Prowess Inc, Concorde, CA) version 5.10. All patients received a total dose of 50.4 Gy in 28 fractions. Dose volume histogram (DVH) data were extracted. Paired t-test was used to compare the following among IFI vs. ENI groups: heart mean dose, esophagus mean and max dose, and lung mean dose, V5 and V20. Results Results : Mean age at diagnosis was 61 years. 87 % had adenocarcinoma. 93 % of patients underwent lobectomy with lymph node sampling. Mean doses to the lungs, esophagus and heart as well as lung V5 and V20 were significantly lower with IFI (Table 1).