Echo-planar readout is often used in hyperpolarized imaging because of its rapid sampling of the k-space. This scheme, however, suffers from Nyquist ghosting artifacts. We present a data-driven approach to correct Nyquist ghosting in hyperpolarized 13C-EPSI and EPI . The method exploits the sparsity of the spectra in hyperpolarized 13C imaging spectroscopy, and estimates the phase discrepancies from the odd and even data in 13C-EPSI. The method was demonstrated in phantoms and in vivo. We show that it performs better than the 1H reference-based reconstruction, and it eliminates the need for reference scans, which may be an unnecessary source of error.