The effect of deep learning-based compressed sensing on the image quality of contrast-enhanced 3D T1-weighted images of the maxillofacial region