ESTRO 2022 - Abstract Book

S1581

Abstract book

ESTRO 2022

PO-1775 Associations of CT-based radiomics data with disease recurrence in early stage Lung cancer patients

A. Giraldo Marin 1 , M. Ligero 2 , A. Seoane 3 , M. Ramos 1 , M. Simo 4 , J. GIralt 5 , M. Escobar 6 , P. Raquel 7

1 Vall D´Hebron Hospital, Radiation Oncology, Barcelona, Spain; 2 Vall D’Hebron Institute of Oncology, Radiomics Grroup, Barcelona, Spain; 3 Vall D´Hebron Hospital, Medical Physics, Barcelona, Spain; 4 Vall D´Hebron Hospital, Nuclear Medicine, Barcelona, Spain; 5 Vall D´Hebron, Radiation Oncology, Barcelona, Spain; 6 Vall D´Hebron, Radiology, Barcelona, Spain; 7 Vall D’Hebron Institute of Oncology, Radiomics Group, Barcelona, Spain Purpose or Objective To develop an association between CT-based radiomics score and clinical prognostic factors capable of predicting recurrence free survival (RFS) in patients with lung lesions <5cm Materials and Methods We analysed data from 62 patients and 66 lesions treated with SABR from January 2015 to February 2019 in our institution. 4D PET-CT images were used for treatment planning, however, only CT images were used for extracting Radiomics features. The dose and number of fractions were selected according to our institution protocol. Retrospectively, a RFS predictive signature was derived. LASSO-Cox regression was implemented for feature selection and univariate and multivariate Cox- proportional Hazard regression were implemented with radiomics features and prognostic clinical factors. Kaplan- Meier (KM) curves were implemented to evaluate the prognostic value of the signature. A nomogram was developed for the clinical radiomics model.

Results Patient’s characteristics are displayed in Table 1. The LASSO-Cox selected three radiomics features as relevant for RFS prediction from GLCM and GLSZM matrices. The radiomics model associated with RFS with a Concordance index (CI) of 0.63[0.55- 0.72]. Four clinical factors were selected by LASSO (body mass index, Karnofsky status, chronic obstructive pulmonary disease, and previous thorax irradiation). The clinical model showed a similar predictive capacity as radiomics only (CI = 0.68[0.57- 0.78]). The integration of radiomics with clinical data improved the predictive value to a CI=0.84[0.78-0.9]. KM curves showed significant differences in RFS between low, medium and high clinical radiomics score (Figure 1).