Adaption in der dynamischen kontrastmittelverstärkten Magnetresonanztomographie

In breast DCE MRI, dynamic data are acquired to assess signal changes caused by contrast agent injection in order to classify lesions. Two approaches are used for data analysis. One is to fit a pharmacokinetic model, such as the Tofts model, to the data, providing physiological information. For accurate model fitting, fast sampling is needed. Another approach is to evaluate architectural features of the contrast agent distribution, for which high spatial resolution is indispensable. However, high temporal and spatial resolution are opposing aims and a compromise has to be found. A new area of research are adaptive schemes, which sample data at combined resolutions to yield both, accurate model fitting and high spatial resolution morphological information. In this work, adaptive sampling schemes were investigated with the objective to optimize fitting accuracy, whilst providing high spatial resolution images. First, optimal sampling design was applied to the Tofts model. By that it could be determined, based on an assumed parameter distribution, that time points during the onset and the initial fast kinetics, lasting for approximately two minutes, are most relevant for fitting. During this interval, fast sampling is required. Later time points during wash-out can be exploited for high spatial resolution images. To achieve fast sampling during the initial kinetics, data acquisition has to be accelerated. A common way to increase imaging speed is to use view-sharing methods, which omit certain k-space data and interpolate the missing data from neighboring time frames. In this work, based on phantom simulations, the influence of different view-sharing techniques during the initial kinetics on fitting accuracy was investigated. It was found that all view-sharing methods imposed characteristic systematic errors on the fitting results of Ktrans. The best fitting performance was achieved by the scheme ``modTRICKS'', which is a combination of the often used schemes keyhole and TRICKS. It is not known prior to imaging, when the contrast agent will arrive in the lesion or when the wash-out begins. Currently used adaptive sequences change resolutions a fixed time points. However, missing time points on the upslope may cause fitting errors and missing the signal peak may lead to a loss in morphological information. This problem was addressed with a new automatic resolution adaption (AURA) sequence. Acquired dynamic data were analyzed in real-time to find the onset and the beginning of the wash-out and consequently the temporal resolution was automatically adapted. Using a perfusion phantom it could be shown that AURA provides both, high fitting accuracy and reliably high spatial resolution images close to the signal peak. As alternative approach to AURA, a sequence which allows for retrospective resolution adaption, was assesses. Advantages are that adaption does not have to be a global process, and can be tailored regionally to local sampling requirements. This can be useful for heterogeneous lesions. For that, a 3D golden angle radial sequence was used, which acquires contrast information with each line and the golden angles allow arbitrary resolutions at arbitrary time points. Using a perfusion phantom, it could be shown that retrospective resolution adaption yields high fitting accuracy and relatively high spatial resolution maps