深度学习重建算法在低管电压冠状动脉CT血管成像中的应用

doi:10.16150/j.1671-2870.2022.03.014

摘要/Abstract

摘要：

目的：比较深度学习重建算法（deep learning-based image reconstruction,DLIR）、自适应统计迭代重建算法（adaptive statistical iterative reconstruction-veo,ASiR-V）和滤波反投影重建算法（filtered back-projection,FBP）对低管电压冠状动脉CT血管成像（coronary computed tomographic angiography,CCTA）图像质量的优化效果。方法：前瞻性纳入100例行CCTA扫描的患者,根据其身体质量指数（body mass index,BMI）,选择使用70 kVp（50例,BMI≤26 kg/m²）或80 kVp（50例,BMI>26 kg/m²）管电压扫描,每例患者的图像分别用FBP（A组）、ASiR-V 50%（B组）、中级DLIR（DLIR-Medium,DLIR-M,C组）和高级DLIR（DLIR-High,DLIR-H,D组）进行重建,比较4组重建算法图像的CT值、噪声、信噪比（signal-to-noise ratio,SNR）和对比噪声比（contrast-noise ratio,CNR）,并采用李克特5级评分法对图像质量进行主观评价。结果：在客观图像质量评价中, A、B、C、D 4组两两组间比较,图像噪声、SNR、CNR差异均有统计学意义（P<0.05）,其中D组的图像噪声最低,而SNR和CNR最高;在主观图像质量评价中,C组与D组间差异无统计学意义,但均明显高于A组及B组（P<0.05）。结论：在低管电压CCTA扫描中,使用DLIR重建的图像质量优于ASiR-V 50% 和FBP,提示DLIR适用于临床低管电压CCTA扫描。

关键词: 冠状动脉CT血管成像, 深度学习, 迭代重建, 辐射剂量, 图像质量

Abstract:

Objective: To compare the quality of low tube voltage coronary CT angiography (CCTA) images reconstructed with deep learning-based image reconstruction (DLIR) and with filter back projection (FBP) and with adaptive statistical iterative reconstruction-veo (ASiR-V). Methods: One hundred patients who underwent CCTA were included. The CCTA tube voltage were set as 70 kVp (n=50, BMI≤26 kg/m²) and 80 kVp (n=50, BMI>26 kg/m²) according to body mass index(BMI). The images were reconstructed with FBP (Group A), ASIR-V 50% (Group B), DLIR at medium (DLIR-M, Group C) and high DLIR(DLIR-H, Group D) levels, respectively. Objective evaluation indice including CT attenuation, noise, signal-to-noise ratio (SNR), and contrast-to-noise ratio(CNR) were measured or calculated between groups, and Likert 5-point scale was adopted for subjective image quality assessment. Results: There were significant differences in image noise, SNR, CNR among the 4 groups(P<0.05), and Group D had the highest SNR and CNR, and lowest noise. There was no significant difference between Group C and Group D in subjective scores, but Group C and D both had higher subjective scores than those of Group A and B (P<0.05). Conclusions: For low tube voltage CCTA, images reconstructed with DLIR generate higher quality,and DLIR may be suitable to apply in low tube voltage CCTA.

Key words: Coronary CT angiography, Deep learning, Iterative reconstruction, Radiation dose, Image quality

中图分类号:

R445.4

范婧, 杨文洁, 王梦真, 陆伟, 石骁萌, 朱宏. 深度学习重建算法在低管电压冠状动脉CT血管成像中的应用[J]. 诊断学理论与实践, 2022, 21(03): 374-379.

FAN Jing, YANG Wenjie, WANG Mengzhen, LU Wei, SHI Xiaomeng, ZHU Hong. The application of deep learning algorithm reconstruction in low tube voltage coronary CT angiography[J]. Journal of Diagnostics Concepts & Practice, 2022, 21(03): 374-379.

图/表 5

表1

表2

表3

图1

图2

参考文献 13

[1]	沈洁云, 王忠敏, 陈克敏. 冠状动脉斑块的CT血管造影评价[J]. 中国医学计算机成像杂志, 2016, 22(1):97-100.
	Shen JY, Wang ZM, Chen KM. CT Angiographic Evaluation of Coronary Artery Plaque[J]. Chin Comput Med Imag, 2016, 22(1):97-100.
[2]	Einstein AJ, Henzlova MJ, Rajagopalan S. Estimating risk of cancer associated with radiation exposure from 64-slice computed tomography coronary angiography[J]. JAMA, 2007, 298(3):317-323. pmid: 17635892
[3]	Small GR, Kazmi M, Dekemp RA, et al. Established and emerging dose reduction methods in cardiac computed tomography[J]. J Nucl Cardiol, 2011, 18(4):570-579. doi: 10.1007/s12350-011-9400-1 pmid: 21630110
[4]	Willemink MJ, Noël PB. The evolution of image reconstruction for CT-from filtered back projection to artificial intelligence[J]. Eur Radiol, 2019, 29(5):2185-2195. doi: 10.1007/s00330-018-5810-7 pmid: 30377791
[5]	McCollough CH. CT dose: how to measure, how to reduce[J]. Health Phys, 2008, 95(5):508-517. doi: 10.1097/01.HP.0000326343.35884.03 pmid: 18849683
[6]	Hsieh J, Liu E, Nett B, et al. A new era of image reconstruction: True Fidelity? Technical white paper on deep learning image reconstruction[R/OL]. https://www.gehealthcare.com/-/jssmedia/040dd213fa89463287155151fdb01922.pdf.
[7]	Greffier J, Hamard A, Pereira F, et al. Image quality and dose reduction opportunity of deep learning image reconstruction algorithm for CT: a phantom study[J]. Eur Radiol, 2020, 30(7):3951-3959. doi: 10.1007/s00330-020-06724-w pmid: 32100091
[8]	Kim JH, Yoon HJ, Lee E, et al. Validation of deep-learning image reconstruction for low-dose chest computed tomography scan: emphasis on image quality and noise[J]. Korean J Radiol, 2021, 22(1):131-138. doi: 10.3348/kjr.2020.0116 URL
[9]	Kim I, Kang H, Yoon HJ, et al. Deep learning-based image reconstruction for brain CT: improved image quality compared with adaptive statistical iterative reconstruction-Veo(ASIR-V)[J]. Neuroradiology, 2021, 63(6):905-912. doi: 10.1007/s00234-020-02574-x URL
[10]	Cao L, Liu X, Li J, et al. A study of using a deep lear-ning image reconstruction to improve the image quality of extremely low-dose contrast-enhanced abdominal CT for patients with hepatic lesions[J]. Br J Radiol, 2021, 94(1118):20201086. doi: 10.1259/bjr.20201086 URL
[11]	Benz DC, Benetos G, Rampidis G, et al. Validation of deep-learning image reconstruction for coronary compu-ted tomography angiography: Impact on noise, image quality and diagnostic accuracy[J]. J Cardiovasc Comput Tomogr, 2020, 14(5):444-451. doi: 10.1016/j.jcct.2020.01.002 URL
[12]	中华医学会放射学分会质量控制与安全管理专业委员会. 心血管CT成像辐射剂量优化中国专家共识[J]. 中华医学杂志, 2016(7):510-516.
	Quality Control and Safety Management Committee of Radiology branch of Chinese Medical Association. Chinese experts′ consensus on radiation dose optimization of cardiovascular CT imaging[J]. Natl Med J China, 2016(7):510-516.
[13]	Yin WH, Lu B, Gao JB, et al. Effect of reduced x-ray tube voltage, low iodine concentration contrast medium, and sinogram-affirmed iterative reconstruction on image quality and radiation dose at coronary CT angiography: results of the prospective multicenter REALISE trial[J]. J Cardiovasc Comput Tomogr, 2015, 9(3):215-224. doi: 10.1016/j.jcct.2015.01.010 URL

参数	测量部位	FBP（A组）	ASIR-50%（B组）	DLIR-M（C组）	DLIR-H（D组）
CT值（HU）	主动脉根部	721.7±102.4	723.2±108.1	724.1±102.6	723.8±102.3
	左主干	691.1±113.2	692.2±115.9	693.5±113.2	693.0±111.6
	右冠近段	643.4±115.7	643.0±117.4	642.5±113.7	642.3±111.8
噪声（HU）	主动脉根部	59.6±11.7^b,c,d	50.9±10.0^a,c,d	33.1±8.3^a,b,d	24.6±6.7^a,b,c
	左主干	47.2±11.2^b,c,d	42.2±10.2^a,c,d	30.2±9.5^a,b,d	25.1±8.6^a,b,c
	右冠近段	44.4±12.2^b,c,d	39.1±11.7^a,c,d	32.9±11.4^a,b,d	28.7±10.8^a,b,c
SNR	主动脉根部	12.7±4.1^b,c,d	15.0±4.0^a,c,d	23.2±4.8^a,b,d	31.3±6.8^a,b,c
	左主干	16.5±3.8^b,c,d	18.8±5.7^a,c,d	25.5±4.9^a,b,d	31.2±6.5^a,b,c
	右冠近段	16.3±7.50^b,c,d	18.9±8.9^a,c,d	23.6±11.9^a,b,d	27.4±14.2^a,b,c
CNR	主动脉根部	21.7±8.1^b,c,d	34.5±10.8^a,c,d	50.2±15.3^a,b,d	64.5±15.4^a,b,c
	左主干	20.1±8.6^b,c,d	31.5±12.3^a,c,d	44.8±16.4^a,b,d	60.2±18.3^a,b,c
	右冠近段	17.1±8.4^b,c,d	18.3±14.8^a,c,d	39.8±18.5^a,b,d	55.5±19.8^a,b,c

参数	测量部位	FBP（A组）	ASIR-50%（B组）	DLIR-M（C组）	DLIR-H（D组）
CT值（HU）	主动脉根部	595.1±118.6	596.9±119.4	596.8±119.5	597.24±119.5
	左主干	541.7±115.5	542.3±114.3	543.2±112.8	544.8±110.5
	右冠近段	527.5±117.4	527.3±119.3	528.7±122.6	528.9±116.7
噪声（HU）	主动脉根部	57.2±7.4^b,c,d	46.3±6.9^a,c,d	30.6±4.1^a,b,d	21.3±3.9^a,b,c
	左主干	41.6±18.2^b,c,d	34.1±17.8^a,c,d	25.3±14.6^a,b,d	20.2±13.9^a,b,c
	右冠近段	39.0±16.3^b,c,d	32.5±15.2^a,c,d	27.1±13.1^a,b,d	24.6±12.3^a,b,c
SNR	主动脉根部	11.2±2.5^b,c,d	13.4±2.3^a,c,d	21.5±4.6^a,b,d	28.4±5.8^a,b,c
	左主干	16.1±7.5^b,c,d	20.2±9.4^a,c,d	28.9±12.6^a,b,d	35.5±18.2^a,b,c
	右冠近段	16.3±7.5^b,c,d	19.9±9.3^a,c,d	24.9±12.3^a,b,d	28.1±14.6^a,b,c
CNR	主动脉根部	14.4±6.0^b,c,d	23.0±10.2^a,c,d	35.6±15.8^a,b,d	45.2±19.7^a,b,c
	左主干	12.8±6.0^b,c,d	20.6±10.2^a,c,d	31.7±15.90^a,b,d	41.2±18.7^a,b,c
	右冠近段	13.1±5.5^b,c,d	20.1±9.6^a,c,d	29.8±14.9^a,b,d	39.3±18.8^a,b,c

管电压	FBP（A组）	ASiR-50%（B组）	DLIR-M（C组）	DLIR-H（D组）	t值	P值
70 kVp	4/19/22/3/2^{b）,c）,d）}	12/26/10/1/1^{a）,c）,d）}	29/17/3/0/1^a）,b）	42/6/1/0/1^a）,b）	108.3	<0.05
80 kVp	1/22/16/10/1^{b）,c）,d）}	10/24/12/4/0^{a）,c）,d）}	24/21/5/0/0^a）,b）	34/23/3/0/0^a）,b）	112.6	<0.05