Determination of image features in cone beam computerized tomography used in lung stereotactic radiotherapy: A phantom study

Abstract

Introduction The extraction of features from different medical images can provide information on the changes induced in tumor tissue when patients undergo radiotherapy treatments [1]. One of the imaging modalities in daily use to ensure accurate positioning in lung cancer patients undergoing stereotactic body radiation therapy (SBRT) is cone beam computed tomography (CBCT). In this study we propose a method to extract potentially useful quantitative features from CBCT images in a phantom chest. We also analyze the degree of repeatability and reproducibility that can be achieved in each of the features. Materials and method After the implementation of a CBCT quality control protocol [2] to the On-Board Imaging System (OBI) of a Novalis Tx (Varian Medical Systems) medical linear accelerator at the American British Cowdray Medical Center in the City of Mexico, CBCT images of a phantom of the thorax were acquired with different inserts made of thermoplastic material. Using open access software, 2D and 3D shape features, features based on a statistic of first order and texture features (features based on a statistic of second order) were extracted. Results and discussion From the analyzed CBCT images, a total of 107 features were extracted. The methodology was selected according to semi-automatic segmentation procedures to reduce user-dependent results, starting with threshold Hounsfield unit values in a window range for lung tissue. The degree of reproducibility and repeatability of the extracted features and the possible viability of CBCT images for possible radiomics studies in patients with lung cancer and undergoing radiosurgery are analyzed. Conclusions This work describes a methodology for the extraction of reproducible and repeatable features in CBCT images that can potentially be used to asses lung radiotherapy treatment effect in tumor tissue when the methodology is extrapolated to real images of patients. However, the prognostic value of the reported characteristics should be investigated in further studies.