DMF-LP: Enhancing image quality through multimodal medical image fusion using diffusion and Laplacian techniques

Multimodal medical image fusion techniques are of great significance in medical diagnosis and treatment. Existing fusion methods are often difficult to effectively retain and integrate the texture information and features of the source images when dealing with medical images of different modalities, leading to obvious deficiencies in the visual performance and quantitative assessment of the fusion results. To address this problem, this paper proposes a multimodal medical image diffusion fusion model, referred to as DMF-LP. DMF-LP designs a LP-F module before the forward diffusion stage of the model, which performs pre-fusion of the source image pairs, and is used to extract and retain the texture information and features of the source images. In addition, in order to optimize the diffusion model training process, this paper incorporates the information entropy theory into the traditional loss function and constructs a multi-objective loss function, which is used to improve the model’s utilization of medical information and optimize the fusion effect. Experiments demonstrate that DMF-LP outperforms state-of-the-art methods across key metrics: SSIM (0.912 ± 0.011 vs. 0.897 ± 0.063 for DDFM), PSNR (34.416 ± 0.568 dB vs. 34.166 ± 1.093 dB for DDFM), and CC (0.980 ± 0.001 vs. 0.976 ± 0.006 for DDFM). Notably, DMF-LP achieves a 15.6% improvement in SSIM over traditional CNN-based approaches (0.912 vs. 0.869). At the same time, the final fused image maintains visual details and has high structural clarity, which proves the effectiveness of the DMF-LP method. Code is available at https://gitee.com/tan-xueping/DMF-LP.