深度学习图像重建在虚拟平扫CT尿路成像中的应用价值

doi:10.16150/j.1671-2870.2024.02.007

摘要/Abstract

摘要：

目的：探究深度学习图像重建（deep learning image reconstruction， DLIR)在基于能谱CT虚拟平扫(virtual non-contrast scan， VNC)的CT尿路成像（CT urography， CTU）中的图像质量和肾结石测量精度。方法: 回顾性分析2022年9月至2023年4月期间于我院行腹盆平扫和CTU的90例患者的临床和影像学资料。所有患者均在常规行腹盆CT平扫后，进行能谱模式的多期CTU扫描。真实平扫采用ASIR-V 70%权重进行重建（TNC-AR70组）。基于基于实质期及排泌期数据分别获得2组VNC图像，再分别结合DLIR中档和高档权重重建得到4组VNC图像，即实质期-VNC-DLIR中档（venous phase-VNC-DLIR medium， VP-VNC-DM）组、实质期-VNC-DLIR高档（venous phase-VNC-DLIR high， VP-VNC-DH）组、排泌期-VNC-DLIR中档（delay phase-VNC-DLIR medium， DP-VNC-DM）组、排泌期-VNC-DLIR高档（delay phase-VNC-DLIR high，DP-VNC-DH）组。记录平扫、实质期及排泌期的辐射剂量。在5组图像上分别测量CT值、噪声（SD）、信噪比(signal-to-noise ratio， SNR)和对比噪声比(contrast-to-noise ratio， CNR)，并进行组间比较。由2位资深放射诊断医师，用李克特五级量表法(Likert Scale)计分方法对图像质量和病灶显示度进行主观评价。此外，以TNC作为标准，采用Bland-Altman分析VNC上肾结石的CT值和体积的测量差异。结果：在客观图像质量评价上，VNC-DH组图像质量优于TNC-AR70，且5组图像间的CT值差异无统计学意义(P>0.05)；DP-VNC-DH组的图像噪声最低，SNR、CNR最高。在主观评价方面，DP-VNC-DH组的图像质量评分最高，而VP-VNC-DH组在病灶显示度方面表现最佳。在结石的CT值和体积测量上，4组VNC重建图像与真实平扫之间均无统计学差异(P>0.05)。 结论：在CTU检查中，基于DLIR重建技术的VNC图像质量优于基于ASIR-V 70%重建的真实平扫，推荐结合使用实质期和排泌期的DLIR-H重建VNC图像代替真实平扫，以减少CTU扫描的辐射剂量。

关键词: CT尿路成像, 能谱CT, 深度学习, 虚拟平扫, 肾结石

Abstract:

Objective To investigate the effect of dual-energy CT (DECT) virtual non-contrast (VNC) images reconstructed by deep learning image reconstruction (DLIR) on the image quality and measurements of renal calculus in CT urography (CTU). Methods The clinical and imaging data of 90 patients who underwent abdominal and pelvic non-contrast CT examination followed by a nephrographic-phase DE CTU during September 2022 to April 2023 were retrospectively analyzed. The non-contrast CT images were reconstructed with ASIR-V with 70% weight (TNC-AR70). Four groups of VNC images were reconstructed based on medium level and high level DLIR for venous phase and delay phase, namely venous phase-VNC-DLIR medium (VP-VNC-DM), venous phase-VNC-DLIR high (VP-VNC-DH), delay phase-VNC-DLIR medium (DP-VNC-DM), and delay phase-VNC-DLIR high (DP-VNC-DH). The radiation doses of TNC and VNC in venous phase and delay phase were recorded. The mean CT value, image noise (SD), signal-to-noise ratio (SNR) and contrast-to-noise (CNR) were recorded and compared among the five groups. Two radiologists independently assessed the overall image qua-lity and lesion visibility with 5-point Likert scale. Additionally, according to results of TNC, Bland-Altman was used to analyze the measurement differences between VNC and TNC in mean CT value and mean size of renal calculus. Results In the objective assessments, the image quality of the VNC-DH group was better than that of TNC-AR70, and there was no statistically significant difference in CT value among the five groups of images (P>0.05). DP-VNC-DH showed the lowest SD and the highest SNR and CNR values. In the subjective assessments, DP-VNC-DH achieved the best subject scores on image qua-lity, and VP-VNC-DH achieved the best subject scores on lesion visibility. Furthermore, Bland-Altman analysis showed that there was a strong overall agreement between VNC and TNC for renal calculus characterization (all P>0.005). Conclusions VNC generated by DLIR may provide high-quality image compared with the non-contrast images reconstructed with ASIR-V 70% in CTU.The combination of the VNC images generated by DLIR-H from venous phase and delay phase could replace TNC scanning,reducing the radiation dose of CTU scans.

Key words: CT urography, Dual-energy, CT Deep learning image reconstruction, Virtual non-contrast image, Renal calculus

中图分类号:

钱佳乐, 范婧, 朱宏, 王落桐, 孔德艳. 深度学习图像重建在虚拟平扫CT尿路成像中的应用价值[J]. 诊断学理论与实践, 2024, 23(02): 139-145.

QIAN Jiale, FAN Jing, ZHU Hong, WANG Luotong, KONG Deyan. The application of deep learning image reconstruction in dual-energy CT virtual non-contrast CT urography[J]. Journal of Diagnostics Concepts & Practice, 2024, 23(02): 139-145.

图/表 6

表1

表2

图1

图2

图3

图4

参考文献 14

[1]	ZHANG X, CHEN J, YU N, et al. Reducing contrast medium dose with low photon energy images in renal dual-energy spectral CT angiography and adaptive statistical iterative reconstruction (ASIR)[J]. Br J Radiol, 2021, 94(1120):20200974.
[2]	ZENG S E, DU M Y, YU Y, et al. Ultrasound, CT, and MR imaging for evaluation of cystic renal masses[J]. J Ultrasound Med, 2022, 41(4):807-819.
[3]	GEYER L L, SCHOEPF U J, MEINEL F G, et al. State of the art: Iterative CT reconstruction techniques[J]. Radio-logy, 2015, 276(2):339-357.
[4]	ALTHOFF C E, GÜNTHER R W, HAMM B, et al. Intra-arterial ultra low iodine CT angiography of renal transplant arteries[J]. Cardiovasc Intervent Radiol, 2014, 37(4):1062-1067. doi: 10.1007/s00270-014-0838-9 pmid: 24464257
[5]	VIRARKAR M K, VULASALA S S R, GUPTA A V, et al. Virtual Non-contrast Imaging in The Abdomen and The Pelvis: An Overview[J]. Semin Ultrasound CT MR, 2022, 43(4):293-310. doi: 10.1053/j.sult.2022.03.004 pmid: 35738815
[6]	CHENG Y, SUN J, LI J, et al. The added value of virtual unenhanced images obtained from dual-energy CT urography in the detection and measurement of urinary stone[J]. Urology, 2022, 166:118-125.
[7]	MA G, HAN D, DANG S, et al. Replacing true unenhanced imaging in renal carcinoma with virtual unenhanced images in dual-energy spectral CT: a feasibility study[J]. Clin Radiol, 2021, 76(1):81,e21-e27.
[8]	XIAO J M, HIPPE D S, ZECEVIC M, et al. Virtual unenhanced dual-energy ct images obtained with a multimaterial decomposition algorithm: diagnostic value for renal mass and urinary stone evaluation[J]. Radiology, 2021, 298(3):611-619. doi: 10.1148/radiol.2021192448 pmid: 33464180
[9]	MEYER M, NELSON R C, VERNUCCIO F, et al. Virtual unenhanced images at dual-energy CT: influence on renal lesion characterization[J]. Radiology, 2019, 291(2):381-390. doi: 10.1148/radiol.2019181100 pmid: 30860450
[10]	DELABIE A, BOUZERAR R, PICHOIS R, et al. Diagnostic performance and image quality of deep learning image reconstruction (DLIR) on unenhanced low-dose abdominal CT for urolithiasis[J]. Acta Radiol, 2022, 63(9):1283-1292.
[11]	YOO Y J, CHOI I Y, YEOM S K, et al. Evaluation of abdominal ct obtained using a deep learning-based image reconstruction engine compared with CT using adaptive statistical iterative reconstruction[J]. J Belg Soc Radiol, 2022, 106(1):15. doi: 10.5334/jbsr.2638 pmid: 35480337
[12]	PARK J, SHIN J, MIN I K, et al. Image quality and lesion detectability of lower-dose abdominopelvic CT obtained using deep learning image reconstruction[J]. Korean J Radiol, 2022, 23(4):402-412. doi: 10.3348/kjr.2021.0683 pmid: 35289146
[13]	SEO J Y, JOO I, YOON J H, et al. Deep learning-based reconstruction of virtual monoenergetic images of kVp-switching dual energy CT for evaluation of hypervascular liver lesions: Comparison with standard reconstruction technique[J]. Eur J Radiol, 2022, 154:110390.
[14]	ZHONG J, XIA Y, CHEN Y, et al. Deep learning image reconstruction algorithm reduces image noise while alters radiomics features in dual-energy CT in comparison with conventional iterative reconstruction algorithms: a phantom study[J]. Eur Radiol, 2023, 33(2):812-824.

Parameters	TNC（n=90）	VP（n=90）	DP（n=90）
Tube voltage (kV)	120	80/140	80/140
Tube current mode	Smart mA	GSI Assist	GSI Assist
Collimation (mm)	80	80	80
Scanning mode	Helical	Helical	Helical
Matrix	512×512	512×512	512×512
Thickness (mm)	1.25	1.25	1.25
Reconstruction	AR70	VNC+DM/DH	VNC+DM/DH
CTDI_vol（mGy）	9.05±2.63	8.99±2.75	8.99±2.75

Indice	TNC-AR70	VP-VNC-DM	VP-VNC-DH	DP-VNC-DM	DP-VNC-DH	F	P
CT（HU）
Liver	57.09±7.79	58.06±6.33	58.14±6.30	59.12±7.02	59.26±6.96	1.481	0.207
Spleen	50.81±3.17	50.31±3.93	50.56±4.64	51.25±3.07	51.32±3.01	1.300	0.269
Kidney	34.19±4.41^d,e	36.15±7.20	36.33±7.09	38.32±4.62^a	37.93±3.84^a	7.699	<0.001
Fat	-108.95±3.67^d,e	-107.47±5.17	-107.65±4.96^d,e	-105.36±5.09^a,c	-105.36±5.02^a,c	9.496	<0.001
SD（HU）
Liver	11.31±1.58^b,c,d,e	8.83±0.78^a,c,e	6.89±0.65^a,b,d,e	8.45±0.95^a,c,e	6.33±0.68^a,b,c,d	349.022	<0.001
Spleen	10.20±1.65^b,c,d,e	8.98±3.04^a,c,e	7.17±3.18^a,b,d,e	8.08±0.81^a,c,e	6.00±0.58^a,b,c,d	50.902	<0.001
Kidney	9.90±2.50^b,c,d,e	10.10±2.25^a,c,d,e	8.61±2.20^a,b,e	8.10±1.43^a,b,e	6.38±1.33^a,b,c,d	51.360	<0.001
Fat	7.77±1.95^b,d	9.31±4.56^a,c,e	7.04±4.66 ^b,d	8.59±1.53 ^a,c,e	6.69±4.56 ^b,d	7.566	<0.001
SNR
Liver	5.16±1.09^b,c,d,e	6.64±1.03^a,c,d,e	8.53±1.35^a,b,d,e	7.10±1.26^a,b,c,e	9.49±1.64^a,b,c,d	153.403	<0.001
Spleen	5.11±0.89^b,c,d,e	5.82±0.90^a,c,d,e	7.42±1.22^a,b,d,e	6.40±0.67^a,b,c,e	8.62±0.88^a,b,c,d	197.779	<0.001
Kidney	3.70±1.14^c,d,e	3.76±1.13^c,d,e	4.50±1.49^a,b,e	4.88±1.08^a,b,e	6.19±1.40^a,b,c,d	58.707	<0.001
CNR
Liver	23.19±6.53^b,d	18.89±5.36 ^a,c,e	25.88±7.97^b,d	19.17±5.65 ^a,c,e	26.07±7.78^b,d	24.087	<0.001
Spleen	22.28±6.16^b,d	17.93±4.85 ^a,c,e	24.60±7.30^b,d	18.20±5.40^a,c,e	24.79±7.36^b,d	25.349	<0.001
Kidney	19.90±5.45^b,d	16.29±4.53 ^a,c,e	22.32±6.74^b,d	16.63±5.09 ^a,c,e	22.58±6.78^b,d	24.307	<0.001