摘要
目的 探讨基于深度学习的图像重建算法(DLIR)对能谱CT单能量图像及能谱曲线图像质量的影响。方法 将9支装有不同管径及不同浓度碘造影剂及水和钙溶液的聚丙烯试管放置在1个直径为20 cm的圆柱形聚丙烯体模(QSP)内,采用Revolution APEX CT对体模进行能谱CT成像,利用能谱分析软件重建出40~140 keV单能量图像及能谱曲线,选取碘造影剂浓度为3.75 mgI/mL(可模拟延迟期或实质脏器增强等)、15 mgI/mL(可模拟动脉期的腹主动脉)及Water(可模拟平扫期及囊肿、肌肉等用于图像背景的非增强物质)3支试管进行数据测量,分别在FBP、40%ASIR-V(常规临床检查参数)及True FidelityTM(DLIR-L、DLIR-M、DLIR-H)5组图像测量单能量图像(40 keV、70 keV、100 keV)的CT值,计算图像的信噪比(SNR),对比5组图像质量的差异。结果 低浓度碘造影剂(3.75 mgI/mL)、高浓度碘造影剂(15.00 mgI/mL)及水试管内FBP、40%ASIR-V及True FidelityTM(DLIR-L、DLIR-M、DLIR-H)5组图像40 keV、70 keV、100 keV的CT值比较,差异无统计学意义(P>0.05)。40 keV、70 keV、100 keV图像噪声及图像信噪比5组比较,差异均有统计学意义(P<0.05)。True FidelityTM下的噪声值均较FBP及40%ASIR-V降低,图像信噪比提高(P<0.05),True FidelityTM-DLIR-H噪声最小,信噪比最高。结论 在能谱CT成像中,True FidelityTM较FBP及40%ASIR-V在单能量图像噪声降低,信噪比提高。
Objective[HJ1.3mm]To explore the effects of image reconstruction algorithm based on deep learning on the image quality of spectral CT monochromatic image.Methods Nine polypropylene test tubes with different diameters and concentrations of iodine contrast media and water and calcium solutions were placed in a cylindrical plastic phantom(QSP)with a diameter of 20 cm.Spectral CT imaging was performed on the phantom,and the 40~140 keV monochromatic image and energy spectrum curve were reconstructed using the energy spectrum analysis software.Three test tubes with concentrations of 3.75 mgI/mL(which simulated delayed phase or parenchymal organ enhancement),15 mgI/mL(which simulated abdominal aorta in arterial phase)and water(which simulated non enhancement substances used for image background such as plain scanning phase,cyst and muscle)were selected for the data measurement.CT values of monochromatic images(40 keV,70 keV and 100 keV)and image noise were measured in five groups of images,namely FBP,40%ASIR-V(routine clinical examination parameters)and True Fidelity(low,medium and high level),the signal-to-noise ratio(SNR)of each image was calculated,and the differences in the quality of the five groups of images were compared.Results There was no significant difference in CT values of 40 keV,70 keV,100 keV in low-concentration contrast agent(3.75 mgI/mL),high-concentration contrast agent(15.00 mg/mL)and water among five groups of images(P>0.05).There was significant difference in image noise and SNR of single energy images(40 keV,70 keV and 100 keV)(P<0.05).Compared with FBP and 40%ASIR-V,the image noise of True Fidelity TM was lower than that of FBP and 40%ASIR-V,and the image signal-to-noise ratio is improved(P<0.05).True Fidelity TM-DLIR-H had the lowest noise and the highest signal-to-noise ratio.Conclusion In spectral CT imaging,True Fidelity TM has lower noise and higher signal-to-noise ratio than FBP and 40%ASIR-V in single-energy image.
作者
赵艳红
马保龙
町田治彦
沈云
张晓文
石骁萌
苏治祥
张涛
ZHAO Yanhong;MA Baolong;Haruhiko Machida;SHEN Yun;ZHANG Xiaowen;SHI Xiaomeng;SU Zhixiang;ZHANG Tao(Medical Imaging Center,People′s Hospital of Ningxia Hui Autonomous Region,Yinchuan 750002,China;CT Imaging Research Center,GE Healthcare China,Beijing 100176,China;Department of Radiology,Tokyo Women′s Medical University Adachi Medical,Tokyo 123-8558 Japan;Northern University for Nationalities,Yinchuan 750002,China)
出处
《宁夏医学杂志》
CAS
2024年第1期16-19,共4页
Ningxia Medical Journal
基金
宁夏自然科学基金项目(2021AAC03316)
北方民族大学一般科研项目(2022XYZYX03)。
关键词
体模
能谱
体层摄影术
X线计算机
深度学习
图像质量
Phantom
Energy spectroscopy
Tomography
X-ray computer
Deep learning
Image quality
