Higher image reconstruction with excellent structural detail allows experts to perform accurate analysis, especially on the smallest organ details. The interpolation method that approaches the problem of medical image reconstruction, especially 3D, still causes serious problems. The medical image produced by the interpolation method produces blurred or smooth lines on some parts of the organ. This can cause errors in the medical analysis that will be carried out if the reconstruction results are problematic. For this reason, a method is needed that can reconstruct images well without producing blur but does not require very large computer resources. This study aims to evaluate and compare the quality of 3D magnetic resonance imaging medical images reconstructed using interpolation methods and artificial neural network architectures in the DICOM data format. This study evaluates and compares the quality of 3D magnetic resonance imaging medical images reconstructed using interpolation methods and artificial neural network architectures. The test scenario was performed using images from the ADNI dataset and comparing the output results using a variational autoencoder and a multi-level densely connected super-resolution network on 3D data with existing interpolation methods. The evaluation was done using two metrics, i.e., SSIM and PSNR. The results showed that the variational autoencoder method has the highest SSIM and PSNR values, indicating it has the highest image quality among the three methods, while the mDCSRN method has the lowest SSIM and PSNR values, meaning it has the lowest image quality.