INTRODUCTION: The reconstruction of sensitivity–encoded non–Cartesian undersampled MRI has been facilitated by the use of iterative techniques [1]. However, the ill-conditioning of the associated inverse problem produces residual aliasing and noise amplification. A proven approach to stabilize the reconstruction and to diminish these effects is the use of explicit regularization methods [2-3], by including prior information. In dynamic images, especially in Dynamic Contrast-Enhanced MRI (DCE-MRI), the use of an inappropriate regularization image can highly affect the ability to remove artifacts and the temporal resolution of the reconstruction [4]. To overcome this problem, we propose a new dynamic undersampled acquisition scheme, which combines a 3D Radial Phase Encoding (RPE) trajectory [5] with the golden angle profile order [6], allowing regularized reconstruction of dynamic images with different temporal and spatial resolutions. This approach takes advantage of the flexibility and properties of the trajectory to provide explicit regularization information. A compound image, reconstructed from many profiles, is used to regularize the dynamic reconstruction from a few profiles. This compound image is used to stabilize the iterative reconstruction rather than used as a direct constraint, like in the non-iterative back-projection approach (HYPR, [7]). Moreover, since the acquisition is based in the golden angle profile order, an adaptive regularization can be obtained for different dynamic phases.