Selecting cardiac magnetic resonance images suitable for annotation of pulmonary arteries using an active-learning based deep learning model