Cardiac diffusion tensor imaging (cDTI) is an MRI technique for assessing the microstructural architecture of cardiac tissue. However, wider adoption in the research community is hindered by the tedious and inconsistent reproducibility of left ventricle (LV) segmentation and identification of the right ventricular insertion points (RVIP) needed for regional segmentation. Manual segmentation typically relies on multiple image contrasts, including diffusion-weighted images (DWI), mean diffusivity (MD) maps, primary eigenvector maps, and fractional anisotropy maps (FA). Currently, there is also no consensus or standard approach, which complicates comparisons across studies. In this study, we developed an automated LV segmentation and RVIP identification approach using nnUNet, a self-configuring deep-learning model. We targeted a standardized workflow to reduce inter- and intra-observer variability and showed highly consistent results. Our best model produces a Dice Similarity Coefficient (DSC) of 0.91 for LV segmentation, a Hausdorff Distance (HD) of 2.15 mm for anterior RVIP identification, a HD of 1.72 mm for inferior RVIP identification, and insignificant differences in median MD and FA values between automated and expert segmentations. By addressing these challenges, our nnUNet-based approach enhances reproducibility and facilitates broader adoption of cDTI in research.