SMDFnet: Saliency multiscale dense fusion network for MRI and CT image fusion

Abstract

MRI-CT image fusion technology combines magnetic resonance imaging (MRI) and computed tomography (CT) imaging to provide more comprehensive and accurate image information. This fusion technology can play an important role in medical diagnosis and surgical planning. However, there are several issues with current MRI-CT image fusion, such as the presence of artifacts in both MRI and CT images, which may affect the quality and accuracy of the images during the fusion process. The current fusion strategy of MRI and CT images is complex and prone to losing a large amount of information, which still needs further research and improvement. This article proposes a saliency multi-scale dense fusion network for MRI and CT image fusion to address existing issues. The proposed method first uses a pretrained network to extract depth information from MRI and CT images, which can effectively overcome the noise and artifacts caused by directly training and extracting features and improve the saliency information in the original images. Then, a multi-scale dense network is used to further enhance the extracted pre training features and achieve fusion, and multiple loss functions are used to optimize the network and improve the fusion quality. The experimental results show that the fusion results of the proposed method perform better than the objective indicators of the reference method, while retaining more significant information.

Keywords: Fusion strategy; Image fusion; Pretrained features; Saliency multiscale dense fusion network.