Medical image translation plays a pivotal role in bridging the gap between diverse imaging modalities and enhancing diagnostic capabilities in healthcare. Magnetic Resonance (MR) and Computer Tomography (CT) are two different medical image modalities that provide doctors with the information that complements each other in clinical applications. However, to obtain both images occasionally is cost-consuming and prone to unavailability. Medical image translation using GANs is often applied to address challenges such as the limited availability of certain imaging modalities as well as enhancing the visual quality of images. To this end, in order to supplement the existing GAN architecture this paper presents a novel generative network, called VAE-CycleGAN which integrates the advantages of Variational Auto-Encoders and Generative Adversarial Networks. Through rigorous comparisons with other conventional networks it is demonstrated that this new VAE-CycleGAN model is superior to other prevalent image translation methods and yields higher quality synthetic CT images. Experimental results indicate that the proposed VAE-CycleGAN model produces better results than other conventional generative models, achieving the PSNR and MAE values of 20.61364 And 0.06935 respectively.