In medical diagnostics, multi-modality medical image fusion is essential for providing comprehensive pathological information. However, traditional fusion techniques often fail to adequately combine high and low-frequency information from source images, resulting in the loss of crucial details. To address this issue, we introduce an innovative medical image fusion model, Diff-MF. Based on an explicit denoising diffusion probabilistic model (DDPM) and integrated with a feature fusion module, this model effectively utilizes the semantic features of each modality to distinguish and adaptively merge high and low-frequency information, thereby preserving rich structural and detailed information. Experimental results on public Harvard datasets (MRI-CT images) demonstrate our model’s exceptional performance in maintaining the integrity of modal structures and richness of details, highlighting its effectiveness and potential for application in the medical image fusion field.