Cardiac diffusion tensor imaging (cDTI) sequences inherently suffer from low signal-to-noise (SNR) ratios. Although high field strength systems improve SNR, long single-shot readout trains such as echo-planar imaging experience detrimental effects due to changes in magnetic susceptibility at tissue boundaries. Using synthetic and in vivo free-breathing cDTI data, an iterative time-segmented off-resonance correction methodology was implemented and evaluated. Using this approach, the cDTI data was geometrically restored to the original shape, and underlying tensors metrics were corrected. The framework holds potential to aid geometrically accurate in vivo cDTI for multi-contrast and multi-modal imaging studies.