Medical image Synthesis with improved Deep Convolutional Bi-Generative Adversarial Network Aided Genetic Algorithm

Abstract

In different clinical applications, important role is played by Medical imaging. Acquisition of various image models are limited by radiation lose and cost considerations. Desired image modality are estimated by medical image synthesis without original scan. To address this issues, generative adversarial method is proposed in this research. For a given source image, target image is generated by training fully convolutional network (FCN).

Better model of FCN is obtained by using adversarial learning method. Mapping of source to target imaging is done in an effective way by this and highly accurate target images are produced by this. Generation of target images with blurring is avoided by incorporating loss function with image-gradient-difference in the design of FCN. Network is trained using Long-term residual unit. Deep convolutional adversarial network with context awareness is created by applying Auto-Context Model (ACM).

Deep Convolutional GA connected with bi-generative adversarial network (DC-Bi-GAN) is implemented for synthesizing images. From source images, target images are accurately synthesized by proposed model and it is more robust as shown by results of experimentation. From MRI, CT is generated and from 3T MRI, 7T MRI are generated using this model with three datasets. In all task and dataset, superior results are produced by proposed model.