[Accuracy of Virtual Non-contrast Image Reconstruction Using Material Decomposition for Fast kV-switching Dual-energy CT].

Researchers

Journal

Modalities

Models

Abstract

Dual-energy computed tomography (DECT) system can generate virtual non-contrast (VNC) images. Although several reconstruction algorithms are defined, there are not many researches using deep learning image reconstruction (DLIR) algorithm. In this study, we evaluated the accuracy of the VNC image reconstruction under various conditions using DLIR algorithm.At first, each iodine insert with variable concentrations (2.0, 5.0, 10.0, 15.0 mg/ml) or diameters (2.0, 5.0, 10.0, 28.5 mm), or mixed insert including blood-mimicking material with iodine (iodine concentrations: 2.0, 4.0 mg/ml) was put in the center of the multi-energy CT phantom (Gammex, USA). This phantom was placed in the isocenter of DECT, and it scanned and reconstructed the VNC images. In addition, the VNC images were reconstructed with various display field of view (DFOV) sizes (240, 350 mm) or reconstruction algorithms (filtered back projection, advanced statistical iterative reconstruction, deep learning image reconstruction) for each iodine diameter. Attenuation values of these images (CTVNC) were measured and assessed by placing a circular region of interest (ROI) on each insert.CTVNC form iodine inserts increased with iodine concentration became lower, whereas CTVNC form blood plus iodine inserts were stable regardless of low iodine concentration. As iodine diameter became smaller, CTVNC increased remarkably. CTVNC remained steady even though reconstruction parameters were varied.In our study, the VNC image reconstruction using DLIR algorithm was affected by various conditions such as iodine concentration and size. In particular, its accuracy was reduced by the size of target.