Magnetic Resonance Imaging (MRI) and Computed Tomography (CT) are two fundamental imaging modalities in clinical practice, each offering distinct advantages. MRI provides superior soft tissue contrast, while CT is particularly effective in visualizing bone structures and detecting acute conditions. Despite their complementarity, the broader clinical adoption of these modalities is constrained by practical challenges: MRI suffers from lengthy acquisition times, high equipment and operational costs, and sensitivity to patient motion artifacts, whereas CT poses concerns due to its relatively high radiation exposure. To address the issues of incomplete or unavailable imaging modalities, deep learning-based medical image translation techniques have gained increasing attention. While Generative Adversarial Networks (GANs) have shown potential in cross-modal translation, they often suffer from instability and mode collapse. In contrast, Denoising Diffusion Probabilistic Models (DDPMs) have emerged as a more stable and high-fidelity alternative, leveraging a progressive denoising process. To further advance the performance of unsupervised medical image translation, we propose TFTrans, a novel adversarial diffusion-based framework that synergistically integrates time-frequency feature extraction, perceptual supervision, and adversarial diffusion mechanisms. TFTrans enhances structural consistency, detail preservation, and training stability, providing a robust and efficient solution for high-quality cross-modal medical image translation. Experimental results on multiple public datasets demonstrate the model’s superior performance and strong potential for clinical applications.