超低剂量平扫CT深度学习图像重建评价肺部病灶的可行性

doi:10.3969/j.issn.1674-8115.2022.08.011

上海交通大学学报（医学版） ›› 2022, Vol. 42 ›› Issue (8): 1062-1069.doi: 10.3969/j.issn.1674-8115.2022.08.011

• 论著 · 临床研究 • 上一篇

超低剂量平扫CT深度学习图像重建评价肺部病灶的可行性

赵珂珂(), 蒋蓓蓓(), 张璐, 王凌云, 张亚平, 解学乾()

上海交通大学医学院附属第一人民医院放射科，上海 200080

收稿日期:2022-04-17 接受日期:2022-07-25 出版日期:2022-08-28 发布日期:2022-10-08
通讯作者: 解学乾 E-mail:1797673460@qq.com;jennifer.chiang@hot mail.com;xiexueqian@hotmail.com
作者简介:赵珂珂（1996—），女，硕士生；电子信箱：1797673460@qq.com
蒋蓓蓓（1995—），女，博士生；电子信箱：jennifer.chiang@hot mail.com第一联系人：（赵珂珂、蒋蓓蓓并列第一作者）
基金资助:
国家自然科学基金面上项目(81971612);科技部国际合作项目(2016YFE0103000);上海市教育委员会高峰高原学科建设计划(20181814)

Feasibility of ultra-low-dose noncontrast CT based on deep learning image reconstruction to evaluate chest lesions

ZHAO Keke(), JIANG Beibei(), ZHANG Lu, WANG Lingyun, ZHANG Yaping, XIE Xueqian()

Department of Radiology, Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai 200080, China

Received:2022-04-17 Accepted:2022-07-25 Online:2022-08-28 Published:2022-10-08
Contact: XIE Xueqian E-mail:1797673460@qq.com;jennifer.chiang@hot mail.com;xiexueqian@hotmail.com
Supported by:
National Natural Science Foundation of China(81971612);International Scientific Alliance Program of Ministry of Science and Technology of China(2016YFE0103000);Shanghai Municipal Education Commission—Gaofeng Clinical Medicine Grant Support(20181814)

摘要/Abstract

摘要：

目的·探讨使用超低剂量CT（ultra-low-dose CT，ULDCT）平扫评价基于实体瘤疗效评价标准（response evaluation criteria in solid tumors，RECIST）定义的肺部靶病灶和磨玻璃结节的可行性。方法·2020年4月—6月纳入接受了胸部ULDCT平扫（0.07~0.14 mSv）和低剂量增强CT检查（2.38 mSv），而且有RECIST标准定义的可测量肺部靶病灶或直径≤1 cm磨玻璃结节的患者。每例患者均重建了4组图像，包括3组ULDCT图像，分别为80%强度的多模型自适应统计迭代重建（adaptive statistical iterative reconstruction-V with an 80% strength level，ASIR-V-80%）图像、中等强度的深度学习重建（deep learning image reconstruction of moderate strength，DLIR-M）图像和高强度的深度学习重建（deep learning image reconstruction of high strength，DLIR-H）图像，以及1组作为参考标准的增强CT图像。结果·80例患者符合入组标准，平均年龄（62±11）岁，共80个靶病灶和27个磨玻璃结节。3组ULDCT图像的肺部靶病灶测量值（r分别为0.988、0.987和0.990）、≤1 cm磨玻璃结节的直径测量值（r分别为0.905、0.906和0.969）、非肺门淋巴结靶病灶测量值（r分别为0.969、0.957和0.977）、肺门淋巴结靶病灶测量值（r分别为0.972、0.994和0.994）与增强CT有很高的相关性。Bland-Altman分析显示，DLIR-H重建图像中肺部靶病灶测量值的大小与参考值的差异为4.3%（95%一致性界限：-5.7%~14.3%），非肺门淋巴结靶病灶测量值的大小与参考值的差异为5.1%（-9.1%~19.3%），优于ASIR-V-80%［8.5%（-3.3%~20.3%），9.7%（-6.0%~25.3%）］和DLIR-M［8.5%（-4.2%~21.3%，8.8%（-9.9%~27.5%）］。DLIR-H重建图像中的肺门淋巴结病灶测量值大小与参考值的差异为18.3%（8.8%~27.9%），优于ASIR-V-80%［20.2%（-1.2%~41.5%）］和DLIR-M［23.4%（13.5%~33.2%）］。DLIR-H重建图像中磨玻璃结节测量值与参考值的差异为7.0%（-5.7%~19.7%），优于ASIR-V-80% ［14.4%（-4.4%~33.2%）］和DLIR-M［16.3%（-4.1%~36.7%）］。结论·基于DLIR-H重建的ULDCT平扫图像中的肺部靶病灶和直径≤1 cm磨玻璃结节的测量值与增强CT图像有很高的相关性和较低的差异性，有利于在大幅度降低辐射剂量下对肿瘤和磨玻璃结节进行重复扫描。

关键词: 肺癌, RECIST, 深度学习, 超低剂量CT

Abstract:

Objective ·To explore the feasibility of using noncontrast ultra-low-dose CT (ULDCT) to evaluate chest target lesions based on response evaluation criteria in solid tumors (RECIST) and ground glass nodules (GGNs). Methods ·From April to June 2020, patients who underwent noncontrast chest ULDCT (0.07?0.14 mSv) and low-dose enhanced CT (2.38 mSv) and had measurable target lesions defined by RECIST and GGNs ≤1 cm in diameter were included. Four sets of CT images were reconstructed for each patient, including 3 sets of ULDCT images, i.e., adaptive statistical iterative reconstruction-V with an 80% strength level (ASIR-V-80%), deep learning image reconstruction of moderate strength (DLIR-M) and deep learning image reconstruction of high strength (DLIR-H), and one set of enhanced CT images as the reference. Results ·Eighty patients who had 80 target lesions and 27 GGNs met the inclusion criteria, and the average age was (62±11) years old. Between the ULDCT images (3 sets of image reconstruction) and enhanced CT, the measured values of target lesions (r=0.988, 0.987 and 0.990, respectively), GGNs≤1 cm in diameter (r=0.905, 0.906 and 0.969, respectively), mediastinal lymph node target lesions (r=0.969, 0.957 and 0.977, respectively), and hilar lymph node target lesions (r=0.972, 0.994 and 0.994, respectively) were highly correlated. Bland-Altman analysis showed that the difference between the measured size of lung target lesion and reference value in DLIR-H reconstruction image was 4.3% (95% limits of agreement: -5.7%?14.3%), and the difference between the measured size of mediastinal lymph node target lesion and reference value was 5.1% (-9.1%?19.3%), which was better than that of ASIR-V-80% [8.5% (-3.3%?20.3%), 9.7% (-6.0%?25.3%)] and DLIR-M [8.5% (-4.2%?21.3%), 8.8% (-9.9%?27.5%)]. The difference between the measured size of lymph node target lesions in DLIR-H images and the reference value was 18.3% (8.8%?27.9%), better than ASIR-V-80% [20.2% (-1.2%?41.5%)] and DLIR-M [23.4% (13.5%?33.2%)]. The difference between the measured size of GGNs in DLIR-H images and the reference value was 7.0% (-5.7%?19.7%), better than ASIR-V-80% [14.4% (-4.4%?33.2%)] and DLIR-M [16.3% (-4.1%?36.7%)]. Conclusion ·Noncontrast ULDCT based on DLIR-H is highly correlated and consistent with the traditional enhanced CT in evaluating chest target lesions and GGNs (≤1 cm in diameter), which is conducive to repeated scanning of tumors and GGNs at a significantly reduced radiation dose.

Key words: lung cancer, RECIST, deep learning, ultra-low-dose CT

中图分类号:

R445.4

赵珂珂, 蒋蓓蓓, 张璐, 王凌云, 张亚平, 解学乾. 超低剂量平扫CT深度学习图像重建评价肺部病灶的可行性[J]. 上海交通大学学报（医学版）, 2022, 42(8): 1062-1069.

ZHAO Keke, JIANG Beibei, ZHANG Lu, WANG Lingyun, ZHANG Yaping, XIE Xueqian. Feasibility of ultra-low-dose noncontrast CT based on deep learning image reconstruction to evaluate chest lesions[J]. Journal of Shanghai Jiao Tong University (Medical Science), 2022, 42(8): 1062-1069.

图/表 7

表 1 入组患者基本特征

Tab 1 Basic characteristics of the included patients

Variable	Included patient		P value
Variable	0.07 mSv （n=40）	0.14 mSv （n=40）	P value
Age/year	64±9	61±12	0.165
Gender/n （%）
Male	28 （52）	26 （48）	0.633
Female	12 （46）	14 （54）
BMI/（kg·m^-2）	23.14±3.61	22.30±3.03	0.173
<18.5	3	4	0.243
≥18.5 and <25.0	26	31
≥25.0	11	5
Lung target tumor lesion/n
Malignant	13	15	0.377
Benign or no histological result	9	17
Mediastinal lymph node/n
Malignant	8	3	0.793
Benign or no histological result	4	2
Hilar lymph node/n
Malignant	2	3	0.294
Benign or no histological result	3	1

图1 可测量肺部靶病灶在不同重建图像上的测量值Note： A 57-year-old man had a target lesion in the lower lobe of the right lung and the pathological result was lung adenocarcinoma. A. The long diameter measured on enhanced CT image was 26.5 mm. B. The measured long diameter of ASIR-V-80% reconstructed image was 27.8 mm. C. The measured long diameter of DLIR-M reconstructed image was 28.3 mm. D. The measured long diameter of DLIR-H reconstructed images was 27.2 mm. The long diameter of B, C and D was overestimated by 4.9%, 6.8%, and 2.6% compared with that of A (enhanced CT), respectively.

Fig 1 Measured values of lung target lesions in different image reconstruction kernels

图2 可测量淋巴结靶病灶在不同重建图像上的测量值Note： A 64-year-old male had a lymph node target lesion in the superior mediastinum and the pathological result was infiltrating adenocarcinoma. The lymph node target lesion was metastatic. A. The short diameter of the target lesion measured on enhanced CT image was 17.1 mm. B. The measured short diameter of ASIR-V-80% reconstructed image was 17.7 mm. C. The measured short diameter of DLIR-M reconstructed image was 18.9 mm. D. The measured short diameter of DLIR-H reconstructed images was 17.5 mm. The short diameter of B, C and D was overestimated by 3.5%, 10.5%, and 2.3% compared with that of A (enhanced CT), respectively.

Fig 2 Measured values of lymph node target lesions in different image reconstruction kernels

表2 超低剂量CT靶病灶测量值与增强CT测量值的Pearson相关系数

Tab 2 Pearson's correlation coefficients of target lesions measured on ultra-low-dose CT and enhanced CT images

Item	r
Item	ASIR-V-80% vs enhanced CT	DLIR-M vs enhanced CT	DLIR-H vs enhanced CT
All lung target tumor lesion	0.988	0.987	0.990
Malignant	0.985	0.980	0.986
Benign or no histological result	0.991	0.993	0.994
GGN （diameter≤1 cm）	0.905	0.906	0.969
Mediastinal lymph node	0.969	0.957	0.977
Malignant	0.952	0.930	0.955
Benign or no histological result	0.999	0.997	1.000
Hilar lymph node	0.972	0.994	0.994

表 3 超低剂量CT靶病灶测量值与增强CT测量值差异性的Bland-Altman分析

Tab 3 Bland-Altman analysis of the variability of measured values of target lesions on ultra-low-dose CT and enhanced CT images

Item	Arithmetic mean （95% limits of agreement）
Item	ASIR-V-80% vs enhanced CT	DLIR-M vs enhanced CT	DLIR-H vs enhanced CT
All lung target tumor lesion	8.5% （-3.3%‒20.3%）	8.5% （-4.2%‒21.3%）	4.3% （-5.7%‒14.3%）
Malignant	8.7% （-2.4%‒19.7%）	10.3% （-3.7%‒24.3%）	4.8% （-5.5%‒15.2%）
Benign or no histological result	8.3% （-4.4%‒21.0%）	6.7% （-3.6%‒17.0%）	3.8% （-5.9%‒13.5%）
GGN （diameter≤1 cm）	14.4% （-4.4%‒33.2%）	16.3% （-4.1%‒36.7%）	7.0% （-5.7%‒19.7%）
Mediastinal lymph node	9.7% （-6.0%‒25.3%）	8.8% （-9.9%‒27.5%）	5.1% （-9.1%‒19.3%）
Malignant	11.8% （-6.1%‒29.7%）	10.5% （-11.3%‒32.4%）	6.3% （-10.9%‒23.6%）
Benign or no histological result	5.8% （-0.4%‒11.9%）	5.7% （-3.9%‒15.2%）	2.8% （-0.2%‒5.8%）
Hilar lymph node	20.2% （-1.2%‒41.5%）	23.4% （13.5%‒33.2%）	18.3% （8.8%‒27.9%）

图3 超低剂量CT靶病灶测量值与增强CT测量值差异性的Bland-Altman点图

Fig 3 Bland-Altman analysis of the variability of measured values of target lesions on ultra-low-dose CT and enhanced CT images

表 4 超低剂量CT靶病灶测量值与增强CT测量值差值影响因素的多元线性回归分析

Tab 4 Multiple linear regression analysis of the influential factors on the differences between the measured values of ultra-low-dose CT and enhanced CT of target lesions

Influential factor	Difference of measured values （ASIR-V-80% and enhanced CT）		Difference of measured values （DLIR-M and enhanced CT）		Difference of measured values （DLIR-H and enhanced CT）
Influential factor	β	P value	β	P value	β	P value
BMI	-0.003	0.976	-0.013	0.913	-0.042	0.717
Age	-0.099	0.392	-0.014	0.631	-0.049	0.675
Gender	-0.135	0.251	-0.095	0.416	-0.105	0.370
CT dose	-0.084	0.463	-0.073	0.525	-0.077	0.506
Lesion type	0.256	0.034	0.257	0.033	0.287	0.018
Lesion type （without hilar lymph node）	0.013	0.919	-0.049	0.702	-0.027	0.839
Histological result	0.175	0.143	0.203	0.088	0.142	0.233

参考文献 18

1	EISENHAUER E A, THERASSE P, BOGAERTS J, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1)[J]. Eur J Cancer, 2009, 45(2): 228-247.
2	SHI L, TASHIRO S. Estimation of the effects of medical diagnostic radiation exposure based on DNA damage[J]. J Radiat Res, 2018, 59(suppl_2): ii121-ii129.
3	WOOD D E, KAZEROONI E A, BAUM S L, et al. Lung cancer screening, version 3.2018, NCCN clinical practice guidelines in oncology[J]. J Natl Compr Canc Netw, 2018, 16(4): 412-441.
4	MAZLOUMI M, VAN GOMPEL G, KERSEMANS V, et al. The presence of contrast agent increases organ radiation dose in contrast-enhanced CT[J]. Eur Radiol, 2021, 31(10): 7540-7549.
5	PERISINAKIS K, SEIMENIS I, TZEDAKIS A, et al. Radiation burden and associated cancer risk for a typical population to be screened for lung cancer with low-dose CT: a phantom study[J]. Eur Radiol, 2018, 28(10): 4370-4378.
6	KIM Y, KIM Y K, LEE B E, et al. Ultra-low-dose CT of the thorax using iterative reconstruction: evaluation of image quality and radiation dose reduction[J]. AJR Am J Roentgenol, 2015, 204(6): 1197-1202.
7	JIANG B, LI N, SHI X, et al. Deep learning reconstruction shows better lung nodule detection for ultra-low-dose chest CT[J]. Radiology, 2022, 303(1): 202-212.
8	SHIRI I, AKHAVANALLAF A, SANAAT A, et al. Ultra-low-dose chest CT imaging of COVID-19 patients using a deep residual neural network[J]. Eur Radiol, 2021, 31(3): 1420-1431.
9	CABALLERO B. Humans against obesity: who will win?[J]. Adv Nutr, 2019, 10(suppl_1): S4-S9.
10	SUN J, LI H, WANG B, et al. Application of a deep learning image reconstruction (DLIR) algorithm in head CT imaging for children to improve image quality and lesion detection[J]. BMC Med Imaging, 2021, 21(1): 108.
11	KIM J H, YOON H J, 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.
12	PARAKH A, CAO J, PIERCE T T, et al. Sinogram-based deep learning image reconstruction technique in abdominal CT: image quality considerations[J]. Eur Radiol, 2021, 31(11): 8342-8353.
13	NAM J G, AHN C, CHOI H, et al. Image quality of ultralow-dose chest CT using deep learning techniques: potential superiority of vendor-agnostic post-processing over vendor-specific techniques[J]. Eur Radiol, 2021, 31(7): 5139-5147.
14	JENSEN C T, LIU X, TAMM E P, et al. Image quality assessment of abdominal CT by use of new deep learning image reconstruction: initial experience[J]. AJR Am J Roentgenol, 2020, 215(1): 50-57.
15	BENZ D C, BENETOS G, RAMPIDIS G, et al. Validation of deep-learning image reconstruction for coronary computed tomography angiography: impact on noise, image quality and diagnostic accuracy[J]. J Cardiovasc Comput Tomogr, 2020, 14(5): 444-451.
16	NODA Y, KAGA T, KAWAI N, et al. Low-dose whole-body CT using deep learning image reconstruction: image quality and lesion detection[J]. Br J Radiol, 2021, 94(1121): 20201329.
17	JENSEN C T, GUPTA S, SALEH M M, et al. Reduced-dose deep learning reconstruction for abdominal CT of liver metastases[J]. Radiology, 2022, 303(1): 90-98.
18	蒋蓓蓓, 张亚平, 张琳, 等. 深度卷积神经网络对≤3 cm的亚实性肺腺癌CT图像病理学分型预测的可视化研究[J]. 上海交通大学学报(医学版), 2019, 39(9): 1045-1051.
	JIANG B B, ZHANG Y P, ZHANG L, et al. A visualization study of deep convolutional neural network to classify the pathological type of sub-soild pulmonary adenocarcinoma of ≤3 cm based on CT images [J]. J Shanghai Jiao Tong Univ (Med Sci), 2019, 39(9): 1045-1051.

[1]	邹菁华, 宫淼淼, 沈瑛. KRAS4A^G12C和KRAS4B^G12C对人肺上皮细胞生长和运动的作用差异及机制研究[J]. 上海交通大学学报（医学版）, 2022, 42(8): 1016-1023.
[2]	刘子杨, 王小文, 陈力. lncRNA GK-IT1通过调控醛缩酶A影响非小细胞肺癌细胞的恶性进展[J]. 上海交通大学学报（医学版）, 2022, 42(5): 591-601.
[3]	林嘉希, 汪盛嘉, 赵鑫, 高欣, 殷民月, 朱锦舟. 基于深度卷积神经网络的Barrett食管内镜图片分类模型的建立[J]. 上海交通大学学报（医学版）, 2022, 42(5): 653-659.
[4]	陆文清, 孟周文理, 虞永峰, 陆舜. 非小细胞肺癌第三代表皮生长因子受体-酪氨酸激酶抑制剂的耐药机制及治疗策略[J]. 上海交通大学学报（医学版）, 2022, 42(4): 535-544.
[5]	陈立奇, 薛卓维, 吴氢凯. 基于磁共振成像的女性盆底器官三维数字模型重建的研究进展[J]. 上海交通大学学报（医学版）, 2022, 42(3): 381-386.
[6]	张宇, 吴晓渊, 管丽华, 刘译远, 彭星月, 谢海燕, 胡玮, 郝可可, 夏宁, 陆国军, 侯志波. 高通量药物敏感性筛选系统在非小细胞肺癌伴恶性胸腔积液治疗中的应用[J]. 上海交通大学学报(医学版), 2022, 42(1): 82-89.
[7]	周冰倩, 韩丽, 陈哲逸, 陈诗宇, 郑英霞. 蛋白精氨酸甲基转移酶5在肺癌中的表达及其促进肺癌的作用机制[J]. 上海交通大学学报(医学版), 2021, 41(8): 1009-1016.
[8]	凌徐心仪, 张瑶, 钟华. 非小细胞肺癌免疫治疗获益人群筛选的研究进展[J]. 上海交通大学学报(医学版), 2021, 41(8): 1114-1119.
[9]	贺艳琪, 迟锐, 陈梦萍, 陈思, 刘春良, 刘云霞, 孙海鹏. 支链氨基酸分解代谢在肺癌细胞中的功能[J]. 上海交通大学学报(医学版), 2021, 41(7): 858-864.
[10]	马韵芳, 潘丽娜, 李圳, 高蓓莉, 胡家安, 徐志红. 司美替尼下调KRAS G12V突变型非小细胞肺癌细胞PD-L1水平的探索性研究[J]. 上海交通大学学报(医学版), 2021, 41(6): 741-748.
[11]	徐建华, 江萍, 邓炯. ATP结合盒蛋白G超家族成员2在肺癌中的表达及意义[J]. 上海交通大学学报(医学版), 2021, 41(6): 830-833.
[12]	李小敏, 曲扬, 张少霆, 赵亮, 刘畅, 谢帅宁, 戴尅戎, 艾松涛. 人工智能技术在骨肌系统影像学方面的应用[J]. 上海交通大学学报(医学版), 2021, 41(2): 262-266.
[13]	邹琛1, 2，徐润灏1, 3，张泓2，马展2，陈黎2，张洁3，李敏3，张舒林1. 小分子代谢物在肺癌和肺炎鉴别诊断中的潜在作用[J]. 上海交通大学学报(医学版), 2020, 40(8): 1041-1047.
[14]	何春明，尹航，郑佳杰，唐健，傅于捷，赵晓菁. 肺癌免疫治疗：免疫抑制细胞和肺内免疫[J]. 上海交通大学学报(医学版), 2020, 40(8): 1137-1142.
[15]	陆言巧，沈兰，何奔. 人工智能在心血管疾病的辅助诊疗中的应用[J]. 上海交通大学学报（医学版）, 2020, 40(2): 259-.

超低剂量平扫CT深度学习图像重建评价肺部病灶的可行性

Feasibility of ultra-low-dose noncontrast CT based on deep learning image reconstruction to evaluate chest lesions

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 7

参考文献 18

相关文章 15

编辑推荐

Metrics