Computational methods to boost radiomics

In the field of medicine, radiomics is a method that extracts a large number of quantitative features from medical images with the ultimate goal to enhance the prognosis of patients. Since the term radiomics was coined in 2012, its research efforts has been growing exponentially, fuelled by the ambition to move towards more personalised medicine and thanks to technological development in the hardware and software of medical scanners, as well as to advances in artificial intelligence. This thesis explores different aspects of the radiomics workflow with the aim of finding techniques that improve the results and stability of this method. In details, we investigated here: the development and introduction of new features, available solutions to cope with imbalanced learning, the combination of deep learning and machine learning techniques, and the influence of segmentation on model performance. The results shed light on ways to improve the standard radiomics workow, by modifying the standard procedures.