Cyclegan based Deep Learning for Synthesis of Dynamic Contrast Enhanced Computed Tomography from Computed Tomography
Functional imaging plays a very important role in medical diagnosis and radiotherapy. Dynamic contrast enhanced computed tomography (DCE-CT) is one of the most popular functional imaging techniques. Recently, many researchers have focused on the translation from MRI to CT, while DCT-CT has not been addressed. In this paper, we develop a CycleGAN based deep learning method to synthetize dynamic contrast enhanced computed tomography from plain computed tomography image. We employ U-Net as the generator in the presented translation architecture. 30 patients images comprised of head, thorax and abdomen region are collected for building our model. Each patient has both plain CT and corresponding DCE-CT. The initial result suggests the potential of deep learning for addressing the synthesis of DCE-CT. In addition, U-Net outperforms ResNet as the generator in the model.
Keywords - Dynamic Contrast enhanced, Computed Tomography, Deep Learning, Cycle Generative Adversaral Network