RADIOMICS FOR OUTCOME PREDICTION IN EARLY-STAGE NON-SMALL CELL LUNG CANCER PATIENTS TREATED WITH STEREOTACTIC BODY RADIOTHERAPY (SBRT): METHODOLOGICAL CHALLENGES AND CLINICAL APPLICATIONS

ABSTRACT In recent years, Radiation Oncology has taken advantage of progresses in the fields of biology, computational sciences, and medical engineering. This has laid the foundation for the incorporation of high-throughput data, deriving from different sources, including medical images, histopathological samples, and blood. Of these, quantitative imaging-derived biomarkers represent an appealing and actionable source of information in addition to classical qualitative and semi-quantitative parameters used in clinical practice. This field of study, referred to as radiomics, is the subject of several investigations, on multiple disease sites. Of these, Non-Small Cell Lung Cancer (NSCLC) is a good model of study, not only for its relevance- being one of the so-called “big killers” in Oncology- but also for the availability of preliminary results, suggesting the potentials of radiomics in this clinical setting. Specifically, focus will be given to early-stage NSCLC, a stage of limited disease involvement which may be treated with either surgery, or radiotherapy. On these premises, one of the main scopes of this thesis is to investigate currently unsolved methodological questions in the radiomic pipeline for NSCLC; focusing on common issues, such as the impact of image preprocessing and the stability of radiomic features across the respiratory phases of the simulation computed tomography. To do so, statistics and machine learning methods will be explored and discussed with an eye to their scalability to current clinical practice. Furthermore, the construction of clinical, radiomic and clinico-radiomic models on a retrospective highly-curated dataset will be presented. Other than presenting and commenting these results, space will be given to current limitations of clinico-radiomic models, including their explainability and generalizability. Finally, a Chapter will be dedicated to the presentation of a prospective observational trial- which has recently been funded by the AIRC (Associazione Italiana per la Ricerca sul Cancro), of which I am honored to be the Principal Investigator. This study- named MONDRIAN (Multi-omics integrative modelling for stereotactic body radiotherapy in early-stage non-small cell lung cancer)-will provide the unprecedented opportunity to characterize the disease, and to elucidate the interaction between determinants of radiosensitivity and radioresistance. Taken together, this work aims to provide a compressive outline of the state-of-art of radiomics, including promises and pitfalls, to delve into largely unexplored methodological aspects, and- in its conclusion- to provide a perspective of what the future of advanced outcome modeling could be in the upcoming years.