Medical image fusion method based on dense block and deep convolutional generative adversarial network

  • PDF / 2,416,500 Bytes
  • 16 Pages / 595.276 x 790.866 pts Page_size
  • 101 Downloads / 234 Views

DOWNLOAD

REPORT


(0123456789().,-volV)(0123456789(). ,- volV)

ORIGINAL ARTICLE

Medical image fusion method based on dense block and deep convolutional generative adversarial network Cheng Zhao1 • Tianfu Wang1 • Baiying Lei1 Received: 15 May 2020 / Accepted: 5 October 2020  Springer-Verlag London Ltd., part of Springer Nature 2020

Abstract Medical image fusion techniques can further improve the accuracy and time efficiency of clinical diagnosis by obtaining comprehensive salient features and detail information from medical images of different modalities. We propose a novel medical image fusion algorithm based on deep convolutional generative adversarial network and dense block models, which is used to generate fusion images with rich information. Specifically, this network architecture integrates two modules: an image generator module based on dense block and encoder–decoder and a discriminator module. In this paper, we use the encoder network to extract the image features, process the features using fusion rule based on the Lmax norm, and use it as the input of the decoder network to obtain the final fusion image. This method can overcome the weaknesses of the active layer measurement by manual design in the traditional methods and can process the information of the intermediate layer according to the dense blocks to avoid the loss of information. Besides, this paper uses detail loss and structural similarity loss to construct the loss function, which is used to improve the extraction ability of target information and edge detail information related to images. Experiments on the public clinical diagnostic medical image dataset show that the proposed algorithm not only has excellent detail preserve characteristics but also can suppress the artificial effects. The experiment results are better than other comparison methods in different types of evaluation. Keywords Medical image fusion  Deep convolutional GAN  Dense block  Encoder–decoder  Loss function

1 Introduction In the wake of the continuous advance of computer vision and sensor equipment, medical imaging plays an irreplaceable role in different clinical applications such as treatment planning and surgical navigation [1]. According

Tianfu Wang and Baiying Lei have contributed equally to this work. & Tianfu Wang [email protected] & Baiying Lei [email protected] Cheng Zhao [email protected] 1

School of Biomedical Engineering, Health Science Center, Shenzhen University, National-Regional Key Technology Engineering Laboratory for Medical Ultrasound, Guangdong Key Laboratory for Biomedical Measurements and Ultrasound Imaging, Nanhai Ave 3688, Shenzhen 518060, Guangdong, China

to the extensiveness of imaging mechanisms, the medical images of different modes have unique information capture capabilities for human tissues or organs. For instance, computed tomography (CT) images can often be used to provide an accurate location for dense structures such as bones or grafts with less distortion [2]. Similarly, magnetic resonance imaging (MRI) images can provide soft