Imaging-based prediction models

Abstract

The modern personalized medicine approaches in oncology deal with a huge and barely manageable number of patient-specific variables, from genetic data to simple blood tests, trying to ensure a fully individualized and tailored therapy. In this context, the recent research trends towards imaging biomarkers, quantitative imaging analysis and radiomics applications have gained an ever-increasing interest, leading to the proposal of imaging-based predictors for clinical decision support systems (DSS). The application of advanced machine learning solutions to manage always larger databases of patient imaging-derived variables is becoming increasingly necessary and opens new frontiers in the field of clinical outcome prediction. The different types of predictive modelling techniques, together with their strengths and pitfalls, are highlighted in this chapter which offers a brief overview about the state of the art as well as on the future developments of this fascinating topic.