This thesis develops the role of contrast enhanced MRI in recurrent malignant glioma. This form of imaging is integral to the management of malignant glioma as it is used for diagnosis, treatment planning and following response to therapy. The aim of this thesis is to apply quantitative image analysis techniques to measure parameters of potential prognostic significance and improve methods for assessing radiological response to therapy. Following evaluation of conventional quantitative analysis techniques, novel techniques are developed to facilitate automated, objective assessment of images. A large (n=90) serial image data set is used to evaluate these quantitative methods using Cox proportional hazards model and the best endpoint of survival. The most common parameter measured using medical imaging is tumour size. Conventional measurement systems (1D, 2D and 3D) are described and methods developed facilitating objective comparison of these techniques. Using survival data, the prognostic value of these measures at baseline is assessed. 1D and 2D techniques, advocated by current international response assessment criteria, are shown not to be comparable with the more accurate 3D technique raising questions about their validity. Only 3D measurement of baseline tumour size is shown to be associated with overall survival. Tumour shrinkage during therapy is assumed to be indicative of patient benefit. International radiological response assessment protocols, based on this assumption, define response using percentage change in tumour size from baseline. These protocols are heavily relied upon in the evaluation of new therapy. Measurement systems (1D, 2D and 3D) and response definitions are described and the effect of using different techniques assessed. The validity of such response assessment protocols is assessed using survival data. 1D measurement is shown to underestimate change in tumour size casting further dubiety about the use of this measurement. With regards to predicting outcome however, percentage change in tumour size (measured using 1D, 2D and 3D techniques) is shown to be significantly associated with survival. Assessing the validity of actual response protocols, current criteria are shown to accurately define patients with shorter survival as progressive disease. However, no difference in survival is detected between patients defined as stable disease or partial response demonstrating limitations of current response protocols. Tumour location may also affect patient outcome. Assessment of this parameter is difficult however, as measurement usually relies upon subjective description. An automated method is developed to objectively describe tumour location. This involves spatial normalisation of tumour images to standard space, tumour segmentation and automated anatomical labelling of tumour voxels using a three dimensional database. The prognostic value of these objective measures of tumour location is assessed using survival data. Tumour predominantly affecting the occipital lobe is shown to be significantly associated with shorter survival. Brain tumours cause complex 3D distortion and displacement of major brain structures. As the brain is enclosed in the skull, an increase in volume due to a growing tumour can cause intracranial distortion which may also affect patient outcome. The extent or pattern of tumour induced distortion is difficult to assess however and clinical practice tends to rely upon qualitative description or a simple measure of horizontal or vertical displacement. An automated method is developed to objectively assess tumour induced distortion. This involves spatial normalisation of tumour images to standard space, followed by non-linear high dimensional warping to match images to a normal reference image. Resultant deformation fields are used to measure volumetric change in a series of brain structures reflecting the extent of tumour induced distortion. The prognostic value of these objective measures is assessed using survival data. Asymmetry of the lateral ventricles is shown to be significantly associated with survival indicating that tumour induced distortion is prognostic in recurrent malignant glioma. Following progression of brain tumours over time is particularly problematic, both qualitatively and quantitatively, due to the complexity of the tumour itself and the variable nature of the effect it can have on the remainder of the brain. An automated method is developed to objectively assess temporal change in tumour and secondary structures. This involves rigid body registration, followed by non-linear high dimensional warping of serial image sets. Resultant deformation fields are used to measure temporal volumetric change of tumour and secondary brain structures. The prognostic value of these objective measures of temporal change is assessed using survival data. With the current implementation of this technique, temporal measures are not shown to be predictive of survival. Despite this finding however, the feasibility and potential of this automated method for monitoring tumour progression/regression is demonstrated.