Validating the Radiomics approaches on a population of Non Small Cell Lung Cancer (NSCLC) patients treated by (chemo)-radiotherapy at the Cliniques Universitaires Saint-Luc

Details

Supervisors

Xavier Geets

Faculty

Faculté de médecine et médecine dentaire

Degree label

Master [180] en médecine

Abstract

Introduction: Radiomics is the high-throughput extraction of quantitative image features from medical images (watch the video : https://youtu.be/Tq980GEVP0Y and visit www.radiomics.world). These image features can be divided into four groups depending on the tumour characteristic that they describe: tumour intensity, tumour shape, tumour texture or wavelets. In this study we present a radiomic analysis of the 18 026 features extracted from standard of care, pre-treatment, 4D CT images of a cohort of 44 NSCLC patients treated by chemo-radiotherapy at the Cliniques Universitaires St-Luc. A Radiomic signature is created using a small group of around 2 to 15 radiomic features that were selected based on their correlation to the outcome being studied (survival, histological types...). The signature is created by analysing the features of one cohort of patients and must ideally then be tested and validated on an external dataset in order to see if the features have the same correlation to the selected outcome. Aims and objectives: First, we hypothesized that we could differentiate adenocarcinomas from squamous cell carcinomas using a new radiomic signature, in doing so, this would show that radiomic features extracted from pre-treatment CTs contain information about the tumours histological type. We further hypothesized that we could validate two previously published prognostic radiomic signatures (Aerts et al [3] and Tunali et al [61]) on the UCL cohort. Material and method: Our study was divided into four branches. For the first branch we trained a radiomic signature to differentiate patients with histologically confirmed adenocarcinoma versus squamous cell carcinoma on a cohort of 422 NSCLC patients from MAASTRO and validated the signature on the cohort from the UCL. For the second branch we validated the Aerts et al prognostic signature published in 2013 in Nature Communications on the UCL cohort, this signature uses 4 radiomic features that describe the tumour, to divide the patients into two different prognostic groups. Thirdly we validated the published signature of Tunali et al. on the UCL cohort, this signature is also a prognostic signature that uses only two radiomic features: radial gradient (RG) and radial deviation (RD) to divide the patients into a group with a better prognostic and a more indolent tumour versus a group with a more aggressive disease and a poorer prognostic. And lastly, we also used two different international scoring systems to evaluate the methodology of our approach: the Radiomics Quality Score (RQS) and the TRIPOD recommendations. Results: The three studies showed promising although not statistically significative results, highlighting the potential that radiomic features can have in providing added non-invasive information on the tumour’s proliferation and histological type, this could have a clinical impact, improving treatment decision making and reducing complications linked to invasive diagnostic procedures. This study would need to be carried out on a larger and more balanced cohort of patients in the future.