Background: This study aims to investigate voxel-level dose calculation methods and improve its calculation efficiency in nuclear medicine that can consider animal-specific heterogeneous tissue compositions and radiopharmaceutical biodistributions simultaneously. Materials and Methods: The voxelized mouse phantom was constructed from real mouse CT images and simulated using the Monte Carlo GEANT4 code. According to the dynamic PET images of real mouse, the real distribution of radiopharmaceutical activity was set in the Monte Carlo simulation. The sampling method to improve the calculation efficiency was proposed. Two voxel-level dose calculation methods were implemented in this study. The average absorbed dose in vital target organs and the tumor was calculated by the proposed voxel-level dose calculation methods and the traditional MIRD method respectively. The results of the average absorbed dose calculated by the two methods were compared. Based on the voxel-level dose calculation method, the three-dimensional dose distribution in organs and the tumor was obtained and evaluated. Results: The relative difference of average absorbed dose between the two voxel-level dose calculation methods was mostly less than 10%. The sampling method proposed to improve calculation efficiency for the voxel-level dose calculation can decrease the calculation time by ~34% with less deviation. Conclusion: The results confirmed that the voxel-level dose calculation methods proposed in this study allow for more accurate and efficient assessment of the internal radiation dose.