A deep-learning based synthesized T2 weighted imaging with multi-modality information and k-space correction