Spatio-Temporal Undersampling Artefact Reduction with Neural Networks for Fast 2D Cine MRI with Limited Data