Deep Learning Methods in Dual Energy CT Imaging

Dual-energy computed tomography (DECT) utilizes X-rays of two distinct spectra to visualize the anatomical structures of the patient. By employing material decomposition technology, DECT images can be transformed into material-specific images, enabling quantitative analysis that is not available with traditional single-energy CT (SECT) imaging. Although DECT has achieved success in certain clinical scenarios, it still faces challenges including deployment cost, radiation dose, imaging quality, etc., which hinder its further utilization. To address these issues, researchers implemented deep learning (DL) methods for DECT imaging and produced significant advancements. In this chapter, we go over the applications of deep learning (DL) in DECT imaging, specifically focusing on data synthesis, image reconstruction, and material decomposition. Data synthesis methods aim at synthesizing DECT images from data collected with SECT systems, thereby reducing costs. CD-ConvNet, DL-DECT, FLESHQ1-DECT, and several GAN-based methods developed for this purpose are introduced. Image reconstruction methods focus on high-quality DECT reconstruction under different imaging geometries. We introduce DL methods for image reconstruction in scenarios with sparse-view, limited-angle, and truncated-data conditions. Additionally, many deep-learning-based techniques for material decomposition have been presented to effectively reduce artifacts and enhance accuracy. While the present research on DL-based DECT imaging is in its early stages, these methods demonstrate a robust capability to enhance the quality of DECT imaging and hold significant promise to revolutionize the field of medical imaging.