Fractionated radiotherapy for cancer treatment is a field of constant innovation. Developments in dose delivery techniques have made it possible to precisely direct ionizing radiation at complicated targets. In order to further increase tumour control probability (TCP) and decrease normal-tissue complication probability (NTCP), margins used to account for interfraction and intrafraction anatomical changes and motion need to be reduced. This can only be achieved through proper treatment plan adaptations and intrafraction motion management. This thesis describes methods in support of image-based treatment replanning and real-time intrafraction guidance techniques. The selected contributions detail a number of findings and techniques, in particular: - For ten head & neck cancer patients, changes in tumour density were well described by linear functions with patient-specific slope and intercept. This is of particular interest for proton therapy as delivered dose to a tissue and calculated dose distributions rely on density. Furthermore, tumour density changes might be indicative of treatment response. - It is demonstrated how spatially varying elasticity parameters can be employed in image registration to encourage bone rigidity and local tissue volume change only in the gross tumour volume and the lungs. This is highly relevant in adaptive radiotherapy when modelling significant tumour volume changes. - It is described how cone beam CT reconstruction can be modelled as a deformation of a planning CT scan of the same patient, using a non parametric diffusion based deformation model, opening the door to the use of a number of advanced non-parametric algorithms. An advantage of reconstruction by deformation is that no subsequent image registration is needed in order to obtain the deformation which can be employed for contour propagation in adaptive radiotherapy. - MRI-radiotherapy devices have the potential to offer near real-time intrafraction imaging without any additional ionising radiation. It is detailed how the use of multiple, orthogonal slices can form the basis for reliable 3D soft tissue tracking.