Combining deep learning and radiomics-based machine learning to optimize predictions on medical images

In 2040, it is estimated that 28 million people will be diagnosed with cancer, an increase of almost 50% in comparison to figures from 2020 (GLOBACAN 2020). This will increase cancer’s burden on society and healthcare. Moreover, the lack of clinicians is already a worldwide issue, thus increasing the demand for tools to reduce their workload. There is therefore a need to keep improving and developing clinical decision support systems, which is the focus of this thesis. This thesis consists of two parts, both of which aim to explore the combined value of feature-based and deep-learning models for medical image analysis in cancer. The first part of this thesis investigates the combined predictions obtained from feature-based and deep-learning models. This could potentially lead to more accurate and robust frameworks. The second part of this thesis explores the use of feature-based models to augment the predictions of deep-learning models