A multiscale residual dense fusion network for nuclear medical image fusion

Introduction: Multimodal medical image fusion technology generates new images containing more accurate disease information by fusing different modal images. It not only improves the accuracy and efficiency of diagnosis but also provides strong support for the formulation of treatment plans. Meanwhile, it also shows great potential value in disease monitoring, personalized medicine, and clinical research. Although different multimodal medical image fusion methods have been presented, most of them are hindered by information loss, blurred edges, and low fusion efficiency. Methods: To solve these problems, this paper proposes a multiscale residual dense fusion network (MRDFN) for multimodal medical image fusion. MRDFN integrates the strengths of both the multiscale residual network and dense network to achieve feature extraction and fusion. Results: Experiments show that the fusion images of the proposed method are superior to the reference methods in terms of edge intensity, detail definition, and objective metrics. The comparative analysis of these fusion metrics proves that the fusion image quality of MRDFN is better than that of the reference methods. The suggested method achieves higher values in average gradient, standard deviation, spatial frequency, and visual information fidelity for fusion, with average gradient reaching 2.0 times the average of the comparison algorithms, standard deviation 1.2 times, spatial frequency 2.3 times, and visual information fidelity for fusion 1.3 times. Conclusions: The findings in this study demonstrate that MRDFN outperforms other approaches discussed in the analysis, particularly in objective metrics and detailed information, and the average fusion time of MRDFN is lower than that of most reference methods, demonstrating effective multimodal medical image fusion.