Title:Pseudo Dual Energy CT Imaging using Deep Learning Based Framework: Initial Study

Abstract:Dual energy computed tomography (DECT) has become of particular interest in clinic recent years. The DECT scan comprises two images, corresponding to two photon attenuation coefficients maps of the objects. Meanwhile, the DECT images are less accessible sometimes, compared to the conventional single energy CT (SECT). This motivates us to simulate pseudo DECT (pDECT) images from the SECT images. Inspired by recent advances in deep learning, we present a deep learning based framework to yield pDECT images from SECT images, utilizing the intrinsic characteristics underlying DECT images, i.e., global correlation and high similarity. To demonstrate the performance of the deep learning based framework, a cascade deep ConvNet (CD-ConvNet) approach is specifically presented in the deep learning framework. In the training step, the CD-ConvNet is designed to learn the non-linear mapping from the measured energy-specific (i.e., low-energy) CT images to the desired energy-specific (i.e., high-energy) CT images. In the testing step, the trained CD-ConvNet can be used to yield desired high-energy CT images from the low-energy CT images, and then produce accurate basic material maps. Clinical patient data were employed to validate and evaluate the presented CD-ConvNet approach performance. Both visual and quantitative results demonstrate the presented CD-ConvNet approach can yield high quality pDECT images and basic material maps.