One of the current limitations of dynamic contrast-enhanced MR angiography is the requirement of both high spatial and high temporal resolution. Several undersampling techniques have been proposed to overcome this problem. However, in most of these methods the tradeoff between spatial and temporal resolution is constant for all the time frames and needs to be specified prior to data collection. This is not optimal for dynamic contrast-enhanced MR angiography where the dynamics of the process are difficult to predict and the image quality requirements are changing during the bolus passage. Here, we propose a new highly undersampled approach that allows the retrospective adaptation of the spatial and temporal resolution. The method combines a three-dimensional radial phase encoding trajectory with the golden angle profile order and non-Cartesian Sensitivity Encoding (SENSE) reconstruction. Different regularization images, obtained from the same acquired data, are used to stabilize the non-Cartesian SENSE reconstruction for the different phases of the bolus passage. The feasibility of the proposed method was demonstrated on a numerical phantom and in three-dimensional intracranial dynamic contrast-enhanced MR angiography of healthy volunteers. The acquired data were reconstructed retrospectively with temporal resolutions from 1.2 sec to 8.1 sec, providing a good depiction of small vessels, as well as distinction of different temporal phases. Magn Reson Med 64:514-526, 2010. V C 2010 Wiley-Liss, Inc.Key words: DCE-MRA; undersampling; golden-radial phase encoding; SENSE; regularization.The simultaneous high spatial and temporal resolution required in dynamic MRI is limited due to time acquisition constraints. This is especially important in applications such as dynamic contrast-enhanced MR angiography (DCE-MRA), where for example high spatial resolution is required to visualize small vasculature structures, whereas high temporal resolution is needed to achieve adequate distinction between arterial and venous vessels (1). To attain this goal, several undersampling acquisition and reconstruction techniques have been proposed. These techniques speed up the acquisition of dynamic MRI by reducing the number of acquired samples in k-t space. The missing data can be estimated by exploiting the high spatial-temporal correlation of dynamic process (2-9), by using prior information (10,11), or by employing spatial sensitivity encoding in parallel imaging acquisitions (12)(13)(14).For most of those techniques, the tradeoff between spatial and temporal resolutions is constant for all the time frames and must be specified prior to data acquisition. However, in DCE-MRA a time-varying retrospective adaptation of this tradeoff would be of great benefit. For instance, a high temporal resolution is desirable to visualize the arrival of the contrast agent to the vessels of interest, though a lower spatial resolution is bearable for this part of the dynamic process. Conversely, for arterial and venous signal peak phases, a high spatial ...