A new scheme of medical image fusion using deep convolutional neural network and local energy pixel domain

In this paper, a new multimodal medical image fusion method based on deep convolutional neural networks (CNN) and local spatial domain modification is proposed. First, the source image is fed to Siamese CNN to obtain the weight map and then processed by the Weighted Sum of Eight neighbourhood-based Modified Laplacian (WSEML) to obtain a new image-based WSEML. Next, CT and MRI input images are fed to Weighted Local Energy (WLE). Finally, the activity level measurement based on local energy is dedicated to combining each of the new WLE images and new WSEML images to retrieve useful information at the reconstruction stage. Simulation results demonstrate that the proposed method extracted more useful information with higher visibility from source images, and at the same time reduce fused image artefacts