Assessing the influence of kernel selection on chest computed tomography image quality across varying dose levels using TrueFidelity reconstruction.
Authors
Affiliations (4)
Affiliations (4)
- Department of Medical Radiation Sciences, Institute of Clinical Sciences, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden.
- Region Västra Götaland, Sahlgrenska University Hospital, Department of Biomedical Engineering and Medical Physics, Gothenburg  Sweden.
- Department of Radiology, Institute of Clinical Sciences, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden.
- Region Västra Götaland, Sahlgrenska University Hospital, Department of Radiology, Gothenburg, Sweden.
Abstract
Deep learning image reconstruction (DLIR) utilizes neural networks to generate high-quality computed tomography (CT) images. One commercially available DLIR software is TrueFidelity from GE Healthcare. The Standard kernel was the only available reconstruction kernel previously, but recently other kernels, including the Lung kernel, have been introduced by GE. This study aimed to evaluate the image quality of chest CT scans acquired at full-dose (FD, 2.5Â mSv) and ultra-low-dose (ULD, 0.05Â mSv) when reconstructed using TrueFidelity with both Standard and Lung kernels. Twenty-five patients underwent chest CT scans at Sahlgrenska University Hospital. The images were reconstructed and then evaluated by four radiologists in two different studies, one including ULD CT axial images and the other one the FDCT. Visual Grading Characteristics (VGC) analysis was applied, using the Standard kernel as reference and the area under the VGC curve (AUCVGC) for comparison. At FD, the Standard kernel yielded better results regarding the visualization of six structures and the general image quality. However, in ULD scans, the differences between kernels were not statistically significant. The FD images were mostly rated as acceptable, while ULD images were often rated as probably acceptable or unacceptable, especially for emphysema assessment. Overall, TrueFidelity seems to perform better with the Standard kernel than with the Lung kernel in FD protocols, but no reliable conclusions can be drawn for the ULD protocol.