Experimental Study
Comparison of image quality based on deep-learning image reconstruction and iterative reconstruction algorithm for dual-energy CT: a phantom and animal-model study
Jiang Jiang, Chen Yong, Shi Xiaomeng, Lu Wei, Wang Baisong, Shi Bowen, Wang Wenfang, Zhu Lan, Pan Zilai, Zhang Huan
Published 2023-12-10
Cite as Chin J Radiol, 2023, 57(12): 1361-1367. DOI: 10.3760/cma.j.cn112149-20230824-00126
Abstract
ObjectiveTo investigate the impact of the deep learning reconstruction algorithm TrueFidelityTM for Gemstone Spectral Imaging (TF-GSI) and the adaptive statistical iterative reconstruction algorithm (ASiR-V, hereinafter referred to as ASiR-V) based on phantom and animal models on the image quality of dual-energy CT images.
MethodsGE Revolution Apex CT was used to scan the ACR 464 phantom and a mouse model of gastric cancer with lymph node metastasis (n=16). TF-GSI and ASiR-V were separately used to reconstruct middle and high-grade images (TF-GSI-M, TF-GSI-H, ASiR-V-50%, and ASiR-V-100%) on the phantom and mouse based on virtual monoenergetic images at 70 keV. The task transfer function (TTF) of bone and acrylic, image noise power spectrum (NPS), and detectability index (d′) of the phantom images were evaluated. One-way ANOVA analysis was used to compare the image noise, signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) for brain and liver on images of mice. The consistency of the two reconstruction-algorithm images (TF-GSI-H and ASiR-V100%) in the detection of small lesions by two radiologists (A and B) was evaluated using kappa test.
ResultsIn terms of the phantom, the TF-GSI-H group had the best performance in TTF, NPS, and d′. Compared to ASiR-V-100%, the TTF50% of bone and acrylic in the TF-GSI-H group increased by 2.4% and 8.9%, respectively; the NPS peak decreased by 54.1%, compared to ASiR-V-100%; the d′ of bone and acrylic in the TF-GSI-H group relative to ASiR-V-100% increased by 52.7% and 59.5%, respectively. The TF-GSI group had reduced image noise compared to the ASiR-V group, and both SNR and CNR of the two tissues increased, but the differences between the groups were not statistically significant (all P>0.05). The two reconstruction-algorithm images showed good consistency in image evaluation by the two radiologists (A, Kappa=0.875, P<0.001; B, Kappa=0.625, P=0.012). In terms of the detection of micro-metastases in mice, the TF-GSI group outperformed the ASiR-V group (average accuracy: 83.5% vs 71.9%; average sensitivity: 77.8% vs 61.2%; average specificity: 85.7% vs 85.7%).
ConclusionCompared with iterative reconstruction algorithm, the DLIR algorithm showed improved spatial resolution, reduced image noise, and enabled detectability of micro-lesion for images from dual-energy CT.
Key words:
Tomography, X-ray computed; Deep learning; Iterative reconstruction; Image quality
Contributor Information
Jiang Jiang
Department of Radiology, Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China
Chen Yong
Department of Radiology, Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China
Shi Xiaomeng
CT Imaging Research Center, GE Healthcare China, Shanghai 201203, China
Lu Wei
CT Imaging Research Center, GE Healthcare China, Shanghai 201203, China
Wang Baisong
Department of Biological Statistics, Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China
Shi Bowen
Department of Radiology, Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China
Wang Wenfang
Department of Radiology, Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China
Zhu Lan
Department of Radiology, Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China
Pan Zilai
Department of Radiology, Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China
Zhang Huan
Department of Radiology, Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China
