深度学习重建在低剂量颅脑CT灌注的研究

Study of deep learning reconstruction algorithm in low-dose head CT perfusion

刘昊喆 ¹陈钰 ¹苏童 ¹王彦玲 ¹徐敏 ²王剑 ²金征宇¹

扫码查看

作者信息

1. 100730 北京,中国医学科学院北京协和医学院北京协和医院放射科
2. 100015 北京,佳能医疗系统(中国)有限公司
折叠

摘要

目的评价基于低辐射剂量的深度学习重建(DLR)算法颅脑CT灌注(CTP)的灌注参数及动脉峰值期图像质量,与常规辐射剂量的混合迭代重建(HIR)进行比较.方法回顾性搜集疑似或已有缺血性脑卒中的60例连续病例并分为两组(A组和B组),每组30例.A组采用常规剂量进行颅脑CTP扫描,使用HIR获得三维自适应迭代剂量降低(AIDR 3D)序列,并命名为AIDR-R;B组采用低剂量头灌注序列扫描,使用剂量长度乘积(DLP)重建得到AiCE序列(AiCE-L),同时使用HIR获得AIDR 3D序列(AIDR-L).对3个灌注序列(AIDR-R、AIDR-L、AiCE-L)分别重建灌注参数:脑血流量(CBF)、脑容量(CBV)及平均通过时间(MTT).比较三个序列灌注参数的差异.分别选取3个灌注序列的动脉峰值期图像,测量半卵圆中心、颈动脉虹吸段、基底动脉及脑干的CT值、标准差(SD)值,计算半卵圆中心、颈内动脉虹吸段、基底动脉及脑干的信噪比(SNR),颈内动脉虹吸段及基底动脉的对比噪声比(CNR).同时比较AIDR-R、AIDR-L、AiCE-L序列的额叶、颞叶、枕叶及基底节供血区CBF、CBV及MTT中位数的差异,以及比较AIDR-L、AiCE-L灌注序列在额叶、颞叶、枕叶区、基底节区的动脉峰值图像的CT值、SD值、SNR及CNR值.结果 AIDR-R及AiCE-L在额叶、颞叶、枕叶、基底节区的CBF和CBV值均无统计学差异(P＞0.05).AiCE-L序列在额叶、颞叶、枕叶、基底节区的MTT值均高于AIDR-R 序列(4.03±0.61 比 4.56±0.84、4.18±0.68 比 4.76±0.87、4.21±0.62 比 4.79±0.82、4.05±0.68 比 4.65±0.91,单位 s)(P＜0.001).与AIDR-R动脉峰值期图像比较,AiCE-L序列在半卵圆中心及脑干的CT值低于AIDR-R[脑干(47.18±4.11)HU 比(50.62±5.17)HU、半卵圆中心(40.93±4.64)HU 比(47.96±4.11)HU](P＜0.05).AiCE-L 序列在颈内动脉虹吸段及基底动脉SD值低于AIDR-R[颈内动脉虹吸段(9.62±4.83)HU比(9.85±3.46)HU、基底动脉(9.95±4.96)HU比(11.08± 4.62)H[J](P＜0.05),CT值、SNR及CNR无显著差异(P＞0.05).结论与正常辐射剂量的HIR相比,低辐射剂量组DLR的CBF和CBV参数无统计学差异.DLR可以降低低辐射剂量的颅脑CTP图像动脉峰值期的SD值,提高图像质量,达到与常规剂量HIR相似的结果.

Abstract

Objective To compare the perfusion parameters and image quality of head computed tomography perfusion(CTP)based low radiation dose deep learning reconstruction(DLR)and conventional radiation dose hybrid iterative reconstruction(HIR)techniques.Methods 60 consecutive patients with suspected or known ischemic stroke were retrospectively divided into two groups.A group of 30 patients underwent head CTP scanning with conventional radiation dose and HIR technique to obtain three-dimensional adaptive iterative dose reduction(AIDR 3D)images(AIDR-R).Second group of 30 patients were scanned with low radiation dose and reconstructed using DLR advanced intelligent clear-IQ engine(AiCE)to obtain AiCE-L sequence as well as using HIR to obtain AIDR 3D images(AIDR-L).The cerebral blood flow(CBF),cerebral blood volume(CBV)and mean transit time(MTT)of the three perfusion sequences(AIDR-R,AIDR-L,AiCE-L)were compared.The CT values and standard deviation(SD),signal-to-noise ratio(SNR)and contrast-to-noise ratio(CNR)of the centrum semiovale,carotid siphon,basilar artery and brainstem were determined.Results There was no statistically significant difference in CBF and CBV values between AIDR-R and AiCE-L in the frontal,temporal,occipital,and basal ganglia regions(P＞0.05).The MTT values of the AiCE-L sequence in the frontal,temporal,occipital,and basal ganglia regions are higher than those of the AIDR-R sequence(4.03±0.61 vs 4.56±0.84,4.18±0.68 vs 4.76±0.87,4.21±0.62 vs 4.79±0.82,4.05±0.68 vs 4.65±0.91,unit s)(P＜0.001).Compared with the images of AIDR-R,the CT values of AiCE-L sequence in the center of semiovale and brainstem were lower than those of AIDR-R[brainstem(47.18±4.11)HU vs(50.62±5.17)HU,center of semiovale(40.93±4.64)HU vs(47.96±4.11)HU](P＜0.05).The SD value of AiCE-L was lower than that of AIDR-R[(9.62±4.83)HU vs(9.85±3.46)HU,(9.95±4.96)HU vs(11.08±4.62)HU,(9.85±3.46)HU vs(11.08±4.62)HU](P＜0.05),there was no significant difference in CT value,SNR and CNR(P＞0.05).Conclusion There was no significant difference in CBF and CBV parameters between the low dose group and the normal dose group.DLR can reduce the SD value of peak arterial phase with image quality similar to conventional dose HIR reconstruction in low-dose head CTP scan.

关键词

深度学习/体层摄影术,X线计算机/灌注/辐射剂量

Key words

Deep learning/Tomography,X-ray computed/Perfusion/Radiation dose

引用本文复制引用

基金项目

国家自然科学基金支持(82001814)

中央高水平医院临床科研业务费资助(2022-PUMCH-B-068)

出版年

2024

影像诊断与介入放射学

中山大学

影像诊断与介入放射学

CSTPCD

影响因子：0.51

ISSN：1005-8001

参考文献量20

段落导航