CEST is a new contrast mechanism in MRI. However, a successful application of CEST is hampered by its slow acquisition. This work investigates accelerating parallel CEST imaging using dynamic convolutional recurrent neural networks. This work is the first try to apply recurrent neural networks to accelerate CEST imaging, which jointly learns the spatial and Z-spectral features. The in vivo brain results show that the proposed method demonstrates a much better reconstruction quality of the human brain MTRasym maps than the traditional dynamic compressed sensing method, while the reconstruction time is one hundred times shorter.