Machine learning for imaging in colorectal liver metastases

Medical imaging gives valuable information for diagnosis and treatment planning of cancer patients. Routinely, at primary diagnosis imaging is used for diagnosis, staging and treatment planning and basic metrics are extracted from these images for prognostication, or to assess treatment response. However, there is much more information captured in these images, which is not visible by a radiologist. Radiomics is a collection of analytical methods to convert images into high dimensional data via a set of quantitative phenotypic descriptors called “features”. Radiomic features offer quantitative measurements of tumors including texture, intensity, heterogeneity, and morphology information allowing a comprehensive analysis of the tumor phenotype. These features can also be used to develop diagnostic or prognostic models that may serve as a tool for personalized diagnosis and clinical decision support systems. This thesis proposes different radiomics analyses based on medical images on patient cohorts of colorectal cancer with liver metastasis