分享:
分享到微信朋友圈
X
特别关注
深度学习重建在改善磁共振神经黑色素图像质量中的价值研究
于阳 赵澄 齐志刚 吴涛 卢洁

Cite this article as: YU Y, ZHAO C, QI Z G, et al. Application of deep learning reconstruction in improving the quality of neuromelanin magnetic resonance image[J]. Chin J Magn Reson Imaging, 2023, 14(5): 11-15.本文引用格式:于阳, 赵澄, 齐志刚, 等. 深度学习重建在改善磁共振神经黑色素图像质量中的价值研究[J]. 磁共振成像, 2023, 14(5): 11-15. DOI:10.12015/issn.1674-8034.2023.05.003.


[摘要] 目的 探讨深度学习重建(deep learning reconstruction, DL Recon)在改善神经黑色素MRI序列图像质量中的价值。材料与方法 前瞻性纳入2022年5月10日至2022年5月31日首都医科大学宣武医院正常志愿者30例,并对每位志愿者进行DL Recon 2D快速自旋回波(fast spin echo, FSE)T1WI序列及临床传统2D FSE T1WI扫描,并保存DL Recon 2D FSE T1WI原始图像(即未施加DL Recon的图像),扫描结束后对3组图像进行主客观评价,主观评价采用“五分法”分别对图像均匀度、锐利度、伪影、图像整体质量进行评分,结果采用四分位间距MP25,P75)进行统计描述;客观评价从中脑黑质(substantia nigra, SN)、蓝斑(locus ceruleus, LC)的信噪比(signal to noise ratio, SNR)以及上述区域与周边组织的对比噪声比(contrast noise ratio, CNR)进行评价,采用方差分析对结果进行统计学分析。结果 DL Recon 2D FSE T1WI图像、原始图像及临床传统2D FSE T1WI图像均匀度的主观评分分别为4(4,5)、4(4,5)、4(4,5)(Z=1.31,P>0.05);锐利度评分为4(4,5)、3(3,4)、3(3,4)(Z=2.57,P<0.001);伪影评分为3(3,4)、4(4,5)、4(4,5)(Z=3.43,P<0.001);图像整体质量评分为4(4,5),3(2,3),3(3,4)(Z=2.77,P<0.001)。在对3组图像的主观评分中,图像均匀度之间的差异无统计学意义,锐利度、伪影和图像整体质量评分差异具有统计学意义(P<0.05);DL Recon 2D FSE T1WI图像、原始图像及临床传统2D FSE T1WI图像客观评价结果为:SNRSN 250.38±9.02、66.19±7.32、110.91±10.10,SNRLC 220.41±12.02、50.26±5.89、90.38±11.70;CNRSN 25.30±3.42、7.87±1.12、8.01±1.38;CNRLC 30.17±2.23、10.54±2.08、11.11±1.89。DL Recon 2D FSE T1W1组在显示SN和LC方面的SNR、CNR值均高于原始图像和传统2D FSE T1WI组,且差异有统计学意义(P<0.001)。结论 DL Recon 2D FSE T1WI序列通过采用原始K空间数据深度学习降噪算法,在保证空间分辨率的情况下,改善原始序列图像SNR及CNR,并且可大幅度缩短扫描时间,有望成为神经黑色素MRI的常规检查手段。
[Abstract] Objective To improve the image quality and shorten the scanning time of neuromelanin magnetic resonance imaging sequence commonly used in clinic by deep learning reconstruction (DL Recon).Materials and Methods A total of 30 volunteers were prospectively enrolled, and each volunteer was scanned with DL Recon 2D fast spin echo (FSE) T1WI sequence and clinical traditional 2D FSE T1WI. The original images of DL Recon 2D FSE T1WI were saved. After scanning, the three groups of images were evaluated subjectively and objectively. The "five point method" was used to score the image uniformity, sharpness, artifact and overall image quality, the results were statistically described by interquartile spacing[M (P25, P75)]. The objective evaluation was carried out from the signal to noise ratio (SNR) of substantia nigra (SC) and locus coeruleus (LC) of the midbrain and the contrast noise ratio (CNR) between the above areas and surrounding tissues. The results were statistically analyzed by ANOVA.Results The score of image evenness of DL Recon 2D FSE T1WI images, original images and clinical traditional 2D FSE T1WI images was 4 (4, 5), 4 (4, 5), 4 (4, 5) (Z=1.31, P>0.05), the score of sharpness was 4 (4, 5), 3 (3, 4), 3 (3, 4) (Z=2.57, P<0.001), the score of artifacts was 3 (3, 4), 4 (4, 5), 4 (4, 5) (Z=3.43, P<0.001), and the score of overall image quality was 4 (4, 5), 3 (2, 3), 3 (3, 4) (Z=2.77, P<0.001). In the subjective scores of the three groups of images, there was no significant difference between the three groups except image uniformity. There were significant differences in sharpness, artifacts and image quality in DL Recon 2D FSE T1WI group (P<0.05). The objective evaluation results of DL Recon 2D FSE T1WI images, original images and clinical traditional 2D FSE T1WI images were as follows: SNRSN 250.38±9.02, 66.19±7.32, 110.91±10.10; SNRLC 220.41±12.02, 50.26±5.89, 90.38±11.70; CNRSN 25.30±3.42, 7.87±1.12, 8.01±1.38; CNRLC 30.17±2.23, 10.54±2.08, 11.11±1.89. The SNR and CNR of DL Recon 2D FSE T1WI group were higher than those of original image and traditional 2D FSE T1WI group, and the differences were statistically significant (P<0.001).Conclusions DL Recon 2D FSE T1WI sequence can improve the signal to noise ratio and contrast to noise ratio of the original sequence image under the condition of ensuring the spatial resolution by using the deep learning noise reduction algorithm of the original K-space data, and can greatly shorten the scanning time. It is expected to become the mainstream means of neuromelanin magnetic resonance imaging.
[关键词] 黑质;蓝斑;信噪比;对比噪声比;深度学习重建;磁共振成像
[Keywords] substantia nigra;locus coeruleus;signal to noise ratio;contrast signal to noise ratio;deep learning reconstruction;magnetic resonance imaging

于阳 1, 2   赵澄 1, 2   齐志刚 1, 2   吴涛 3   卢洁 1, 2*  

1 首都医科大学宣武医院放射与核医学科,北京 100053

2 磁共振成像脑信息学北京市重点实验室,北京 100053

3 通用电气医疗(中国)有限公司临床市场部,北京 100176

通信作者:卢洁,E-mail:imaginglu@hotmail.com

作者贡献声明:卢洁设计本研究的方案,对稿件的重要内容进行了修改;于阳起草和撰写稿件,获取、分析或解释本研究的数据;赵澄、齐志刚、吴涛获取、分析或解释本研究的数据,对稿件的重要内容进行了修改;卢洁获得宣武医院汇智人才工程支持计划领军人才项目资助。全体作者都同意发表最后的修改稿,同意对本研究的所有方面负责,确保本研究的准确性和诚信。


基金项目: 宣武医院汇智人才工程支持计划领军人才项目 HZ2021ZCLJ005
收稿日期:2022-07-11
接受日期:2022-10-13
中图分类号:R445.2  R745.1 
文献标识码:A
DOI: 10.12015/issn.1674-8034.2023.05.003
本文引用格式:于阳, 赵澄, 齐志刚, 等. 深度学习重建在改善磁共振神经黑色素图像质量中的价值研究[J]. 磁共振成像, 2023, 14(5): 11-15. DOI:10.12015/issn.1674-8034.2023.05.003.

0 前言

       神经黑色素(neuromelanin, NM)作为一种顺磁性物质,主要分布于中脑黑质(substantia nigra, SN)、蓝斑(locus ceruleus, LC)以及腹侧被盖区,是脑内类神经递质合成的副产物[1]。由于NM在MRI中表现为顺磁性,有缩短T1弛豫时间的特性,故上述结构在T1WI中表现为高信号[2]。因此衍生出NM-MRI,其作为在体脑干NM变化监测的有效途径,在帕金森病(Parkinson's disease, PD)发生发展的诊断中起到重要的影像指导作用[3, 4, 5, 6, 7]。在PD早期作为一种神经元自身保护机制,NM可以螯合人体内的铁、铜等物质,从而出现组织T1弛豫时间的进一步缩短,但是随着PD的发展,NM上铁过载使多巴胺被氧化成醌类等神经毒性物质,进而引起神经元变性,NM颗粒被释放,此时MRI T1WI上SN、LC处高信号消失,这便是NM-MRI监测PD病情进展的重要影像学征象[8, 9, 10, 11, 12]。在以往NM-MRI的相关研究中,由于采集时间较长、患者配合度低造成图像清晰度较差,并且NM-MRI一般采用高分辨力成像,因此此技术不利于临床的应用推广[13]

       基于深度学习的重建(deep learning reconstruction, DL Recon)是近两年来发展出来的新型MRI重建方式,主要通过采用原始K空间数据层面深度学习降噪算法来达到提高图像信噪比(signal to noise ratio, SNR)的目的,从而缩短扫描时间[14, 15, 16]。尚未有研究将DL Recon应用到NM-MRI上,本研究将采用DL Recon 2D快速自旋回波(fast spin echo, FSE)T1WI序列对比传统2D FSE T1WI验证DL Recon在提高NM-MRI SNR的同时,缩短扫描时间,促进NM-MRI在PD评估中的应用。

1 材料与方法

1.1 一般资料

       招募2022年5月10日至2022年5月31日的首都医科大学宣武医院健康志愿者30例。纳入标准:(1)既往无明显临床症状;(2)无外伤、无手术、无任何中枢神经相关疾病史;(3)无MRI扫描禁忌证。排除标准:(1)存在中枢神经系统相关疾病史;(2)有颅内手术或头颅外伤史;(3)扫描过程出现躁动或图像运动伪影者。试验前志愿者充分了解本次试验目的,并且签署知情同意书,本研究遵守《赫尔辛基宣言》,经首都医科大学宣武医院医学伦理委员会批准,批准文号:临械审(2019)019号。

1.2 设备与参数

       所有检查均在3.0 T超导MRI仪上完成(SIGNA Premier 3.0 T,GE Healthcare,Milwaukee,USA),使用线圈为48通道头颅专用线圈(GE Healthcare,USA)。DL Recon 2D FSE T1WI序列参数:TR 400.0 ms,TE 12.9 ms,FOV 220 mm×220 mm,矩阵512×320,层厚2.0 mm,分辨率0.4 mm×0.7 mm×2.0 mm,激励次数(number of excitations,NEX)3,扫描时间4 min 21 s,原始图像保存功能开启;传统2D FSE T1WI序列参数:TR 400 ms,TE 12.9 ms,FOV 220 mm×220 mm,矩阵 512×320,层厚2.0 mm,分辨率0.4 mm×0.7 mm×2.0 mm,NEX 8,扫描时间12 min 3 s。

1.3 图像分析

       扫描完成后,分别对DL Recon 2D FSE T1WI、原始图像、传统2D FSE T1WI进行主、客观评价,其中主观分析由两名具有10年以上影像诊断经验的副主任医师通过“五分法”进行双盲评判,评判项目包括图像均匀度、锐利度、伪影、图像整体质量;客观分析同样由两名具有10年以上影像诊断经验的副主任医师通过在DL Recon 2D FSE T1WI、原始图像、传统2D FSE T1WI图像勾画感兴趣区(region of interst, ROI)的方式进行SNR及对比噪声比(contrast noise ratio, CNR)的测量,ROI的位置分别为SN、LC、红核(red nucleus, RN)和图像背景,根据公式(1)~(4)分别计算SN、LC的SNR及SN、LC与周边组织的CNR。

       其中,SDnoise代表相应层面图像背景噪声信号强度的标准差[17]

1.4 统计学分析

       采用SPSS 26.0软件进行统计分析,主观评分经Kolmogorov-Smirnov正态性检验,图像质量的综合评分不符合正态分布,以中位数(上下四分位数)表示,同时对两名医生主观评分的一致性做Kappa分析,Kappa值>0.75表示一致性良好;客观评价结果符合正态分布的计量资料,采用单变量方差分析进行统计分析,计算组内相关系数(intra-class correlation coefficient, ICC)评价两名医生客观定量评价结果的一致性,ICC>0.75表示信度良好。

2 结果

       共纳入正常志愿者30例,其中男16例,女14例,年龄28~40(34.4±2.3)岁。

2.1 一致性评价

       DL Recon 2D FSE T1WI、原始图像、传统2D FSE T1WI图像主观评价中的4项评分指标中的观察者间一致性强(表1)。三组图像的定量评价结果一致性结果为:SNRSN:0.927(0.862~0.976)、0.857(0.829~0.901)、0.833(0.792~0.862);SNRLC:0.913(0.872~0.946)、0.862(0.814~0.897)、0.853(0.793~0.882);CNRSN:0.894(0.843~0.921)、0.831(0.789~0.862)、0.847(0.792~0.887);CNRLC:0.872(0.827~0.903)、0.826(0.779~0.852)、0.839(0.794~0.872),差异均有统计学意义(P<0.05),此定量评价指标均有良好的观察者一致性。

表1  两名观察者间的DL Recon 2D FSE T1WI、原始图像、传统2D FSE T1WI图像主观评分的一致性评价
Tab. 1  Consistency evaluation of image quality between two obsever in DL Recon 2D FSE T1WI images, original images and clinical traditional 2D FSE T1WI image

2.2 主观评价结果

       DL Recon 2D FSE T1WI、原始图像、传统2D FSE T1WI图像主观评分结果见表2图1。在图像的主观评分中,三组图像之间的图像均匀度差异无统计学意义,锐利度、伪影和图像整体质量DL Recon 2D FSE T1WI组评分均优于其余两组,并且差异有统计学意义(P<0.05)。

图1  DL Recon 2D FSE T1WI、原始图像、传统2D FSE T1WI图像主观评分示意图。1A、1D、1G分别为DL Recon T1WI图像,1B、1E、1H为原始图像,1C、1F、1I为传统2D FSE T1WI图像,箭所示为SN、LC。DL Recon T1WI图像的锐利度、伪影以及图像的整体质量都优于原始图像和传统T1WI图像,DL Recon T1WI图像的SN和LC的SNR及CNR明显高于原始图像及传统2D FSE T1WI图像。DL Recon为深度学习重建方式;FSE为快速自旋回波;SNR为信噪比;CNR为对比噪声比;SN为脑黑质;LC为蓝斑。
Fig. 1  Deep learning reconstruction (DL Recon) 2D fast spin echo (FSE) T1WI, original image, traditional 2D FSE T1WI image subjective scoring diagram. 1A, 1D and 1G are DL Recon T1WI images; 1B, 1E and 1H are original images; 1C, 1F and 1I are traditional 2D FSE T1WI images. Substantia nigra (SN) and locus coeruleus (LC) are showen by the arrows. The sharpness of the DL Recon T1WI images and the overall quality of the images are better than the original images and traditional T1WI images. The signal to noise ratio (SNR) and contrast noise ratio (CNR) of DL Recon T1WI images are significantly higher than those of original images and traditional 2D FSE T1WI images.
表2  DL Recon 2D FSE T1WI、原始图像、传统2D FSE T1WI图像主观评分比较[MP25,P75)]
Tab. 2  The subjective scores of DL Recon 2D FSE T1WI images, original images and clinical traditional 2D FSE T1WI images [M (P25, P75)]

2.3 客观评价结果

       DL Recon 2D FSE T1WI序列的SNRSN、SNRLC、CNRSN和CNRLC(250.38±9.02,220.41±12.02,25.30±3.42和30.17±2.23)均比原始图像和传统2D FSE T1WI图像的高,且多重组间比较最小显著性差异法显示组间差异均有统计学意义(P<0.001)(表3图1)。而原始图像组与传统2D FSE T1WI组在SN和LC成像中的SNR与CNR差异均无统计学意义(P>0.05)。

表3  DL Recon 2D FSE T1WI、原始图像、传统2D FSE T1WI图像客观评价比较(x¯±s
Tab. 3  The objective evaluation results of DL Recon 2D FSE T1WI images, original images and clinical traditional 2D FSE T1WI images(x¯±s

3 讨论

       本研究通过对SN、LC行3种NM-MRI序列扫描,发现DL Recon 2D FSE T1WI序列在显示SN、LC结构时成像效果优于原始图像与传统2D FSE T1WI序列,可以清晰显示出SN、LC的结构以及与相邻组织之间的边界。本研究为国内首次通过DL Recon方法与2D FSE T1WI序列结合的方式,在缩短采集时间的同时实现高分辨力NM-MRI。

3.1 DL Recon 2D FSE T1WI与传统2D FSE T1WI对于图像质量主观评价的对比性分析

       目前临床常用的NM-MRI序列为基于FSE的2D T1WI,由于其稳定、易行等特点在NM-MRI中可以达到临床诊断需要,但是制约其在临床上大范围应用的原因为扫描时间长[18, 19, 20]。由于NM分布的位置如SN、LC等结构精细,因此扫描分辨力要求较高,层面内分辨力均需达到亚毫米级别,层分辨力也需要小于2 mm,且扫描视野推荐使用小视野(≤20 cm),因此很难在较短时间内保证图像SNR,故现阶段如想得到较为满意的SNR,则需要使用较多的NEX,NEX的增加无疑使扫描时间大幅度延长。临床现在使用的NM-MRI 2D FSE T1WI序列扫描时间均在10 min左右,大大限制了其在临床上的应用范围,尤其是对于PD等不能耐受长时间扫描的患者[13]。而DL Recon技术主要通过卷积神经网络模型(convolutional neural network, CNN)对图像的噪声和截断伪影进行消除[21, 22],其通过深度学习方式从原始K空间数据层面进行截断伪影的去除,在此之前同样是对截断伪影进行学习,而后在不损失图像原始信息的前提下去除截断伪影,与此同时保证了图像的分辨力[23]。并且DL Recon作为新型MRI图像重建技术,不仅可以提高MRI图像质量,还可以缩短扫描时间[24, 25, 26, 27]。本研究采用的2 mm层厚的连续无间隔扫描方式,传统的2D FSE T1WI序列时间需要12 min,而DL Recon联合2D FSE T1WI序列时间缩短至4 min,显著提高了扫描效率。此次研究过程中分别比较了DL Recon 2D FSE T1WI、原始图像及传统2D FSE T1WI序列图像的主客观评价,结果发现在图像均匀度上三者之间并没有明显差异,说明DL Recon不同于信号均匀度校正技术,没有纠正因B0、B1不均匀导致图像不均匀的能力;在锐利度的评价上DL Recon 2D FSE T1WI明显优于原始图像及传统2D FSE T1WI图像,说明DL Recon在改善图像分辨力上有很大作用,其原理则是去除了传统原始数据滤波方式对图像锐利度的影响,这和ZOCHOWSKI等[28]的研究相一致;伪影方面的主观评价显示DL Recon 2D FSE T1WI评分较低,虽然DL Recon可以去除截断伪影,但是并没有集成其他伪影去除算法,相反因为其降噪算法的作用在提升图像SNR的前提下也将伪影的信号提升,因此在DL Recon 2D FSE T1WI图像上会存在部分小血管的搏动伪影,但是这些伪影并不会影响对于SN、LC的显示。因此,本研究结果证明了DL Recon 2D FSE T1WI序列在改善神经NM-MRI序列的图像质量上有较高的价值。

3.2 DL Recon 2D FSE T1WI与传统2D FSE T1WI对于图像质量客观评价的对比性分析

       在客观评价方面,DL Recon 2D FSE T1WI序列在SN、LC结构的SNR明显高于原始图像及传统2D FSE T1WI序列图像,得益于其集成的降噪算法,DL Recon通过深度学习方式从原始K空间数据层面进行SNR提升,在此过程中,对噪声进行学习从而加以剔除,最大程度利用原始信号中的有用成分,避免传统原始数据滤过方式中将K空间周边高频信息全部剔除所导致的图像重要信息丢失,因此DL Recon可以在保证原始数据信息不丢失的前提下大幅度提高图像SNR[29, 30]。研究结果显示DL Recon 2D FSE T1WI图像较传统2D FSE T1WI图像SNR高,因此可以通过施加DL Recon减少传统序列的扫描时间,本试验同时对DL Recon 2D FSE T1WI序列图像及原始图像SNR做了对比,发现施加DL Recon后可以较原始图像SNR提升4倍左右,这和KIDOH等[14]的研究一致。除此以外,本研究还对SN、LC与周边组织进行了CNR的研究,结果显示DL Recon 2D FSE T1WI均高于原始图像及传统2D FSE T1WI图像,从侧面反映了在增加SNR的同时,根据CNR=(SI1-SI2)/SDnoise可得两种组织之间信号强度之差增加为原来的四倍,与此同时,SDnoise随着SNR的提升而降低,因此DL Recon 2D FSE T1WI序列图像的CNR较原始图像及传统2D FSE T1WI明显提高,因此可以减少MRI工作者对基本参数的依赖性。

3.3 局限性

       本研究尚存在些许不足,比如LC结构比较微小,对其进行ROI勾画的过程中难免出现一些误差。其次试验样本量较小,且均属于正常志愿者,在后续研究中会继续扩充样本量并加入PD或者PD相关病种。

4 结论

       综上所述,DL Recon 2D FSE T1WI可以取代传统2D FSE T1WI序列作为NM-MRI的首选方式,在实现高分辨力NM-MRI的同时,不仅可以提高SN、LC的SNR,而且还可以大幅度缩短扫描时间,因此基于DL Recon的NM-MRI技术可在临床诊断中提供强有力的手段。

[1]
ZUCCA F A, BELLEI C, GIANNELLI S, et al. Neuromelanin and iron in human locus coeruleus and substantia nigra during aging: consequences for neuronal vulnerability[J]. J Neural Transm (Vienna), 2006, 113(6): 757-767. DOI: 10.1007/s00702-006-0453-2">10.1007/s00702-006-0453-2">10.1007/s00702-006-0453-2.
[2]
柏福运, 高平. 神经黑色素敏感的MRI在帕金森病中的应用进展[J]. 国际医学放射学杂志, 2021, 44(2): 165-169. DOI: 10.19300/j.2021.Z18366">10.19300/j.2021.Z18366">10.19300/j.2021.Z18366.
BAI F Y, GAO P. The advance of neuromelanin sensitive magnetic resonance imaging in Parkinson's disease[J]. Int J Med Radiol, 2021, 44(2): 165-169. DOI: 10.19300/j.2021.Z18366">10.19300/j.2021.Z18366">10.19300/j.2021.Z18366.
[3]
赵会军, 王新艳, 郑璇, 等. 鼻腔鼻窦原发恶性黑色素瘤MRI表现及信号分型[J]. 中华放射学杂志, 2021, 55(1): 29-33. DOI: 10.3760/cma.j.cn112149-20200107-00019">10.3760/cma.j.cn112149-20200107-00019">10.3760/cma.j.cn112149-20200107-00019.
ZHAO H J, WANG X Y, ZHENG X, et al. MRI features and signal pattern of primary sinonasal malignant melanomas[J]. Chin J Radiol, 2021, 55(1): 29-33. DOI: 10.3760/cma.j.cn112149-20200107-00019">10.3760/cma.j.cn112149-20200107-00019">10.3760/cma.j.cn112149-20200107-00019.
[4]
王娟, 魏敏, 承欧梅. 帕金森病运动亚型磁共振成像研究进展[J]. 磁共振成像, 2018, 9(11): 848-852. DOI: 10.12015/issn.1674-8034.2018.11.010">10.12015/issn.1674-8034.2018.11.010">10.12015/issn.1674-8034.2018.11.010.
WANG J, WEI M, CHENG O M. Advance of magnetic resonance imaging in Parkinson's disease subtypes[J]. Chin J Magn Reson Imaging, 2018, 9(11): 848-852. DOI: 10.12015/issn.1674-8034.2018.11.010">10.12015/issn.1674-8034.2018.11.010">10.12015/issn.1674-8034.2018.11.010.
[5]
崔雨沙, 冯涛. 帕金森病黑质病变的结构影像学研究进展[J]. 中华医学杂志, 2022, 102(21): 1625-1630. DOI: 10.3760/cma.j.cn112137-20211111-02516">10.3760/cma.j.cn112137-20211111-02516">10.3760/cma.j.cn112137-20211111-02516.
CUI Y S, FENG T. Progress in structural imaging of substantia nigra lesions in Parkinson's disease[J]. Natl Med J China, 2022, 102(21): 1625-1630. DOI: 10.3760/cma.j.cn112137-20211111-02516">10.3760/cma.j.cn112137-20211111-02516">10.3760/cma.j.cn112137-20211111-02516.
[6]
杨俊强, 杨晓帆, 仲崇琦, 等. 神经黑色素敏感磁共振成像对诊断帕金森病的应用价值[J]. 临床放射学杂志, 2021, 40(5): 855-859. DOI: 10.13437/j.cnki.jcr.2021.05.005">10.13437/j.cnki.jcr.2021.05.005">10.13437/j.cnki.jcr.2021.05.005.
YANG J Q, YANG X F, ZHONG C Q, et al. The value of neuromelanin-sensitive magnetic resonance imaging in the diagnosis of parkinson's disease[J]. J Clin Radiol, 2021, 40(5): 855-859. DOI: 10.13437/j.cnki.jcr.2021.05.005">10.13437/j.cnki.jcr.2021.05.005">10.13437/j.cnki.jcr.2021.05.005.
[7]
SAEED U, LANG A E, MASELLIS M. Neuroimaging advances in Parkinson's disease and atypical parkinsonian syndromes[J/OL]. Front Neurol, 2020, 11: 572976 [2022-07-10]. https://www.frontiersin.org/articles/10.3389/fneur.2020.572976/full. DOI: 10.3389/fneur.2020.572976">10.3389/fneur.2020.572976">10.3389/fneur.2020.572976.
[8]
VILA M. Neuromelanin, aging, and neuronal vulnerability in Parkinson's disease[J]. Mov Disord, 2019, 34(10): 1440-1451. DOI: 10.1002/mds.27776">10.1002/mds.27776">10.1002/mds.27776.
[9]
YE R, RUA C, O'CALLAGHAN C, et al. An in vivo probabilistic atlas of the human locus coeruleus at ultra-high field[J/OL]. NeuroImage, 2021, 225: 117487 [2022-07-10]. https://www.sciencedirect.com/science/article/pii/S1053811920309721. DOI: 10.1016/j.neuroimage.2020.117487">10.1016/j.neuroimage.2020.117487">10.1016/j.neuroimage.2020.117487.
[10]
ZUCCA F A, SEGURA-AGUILAR J, FERRARI E, et al. Interactions of iron, dopamine and neuromelanin pathways in brain aging and Parkinson's disease[J]. Prog Neurobiol, 2017, 155: 96-119. DOI: 10.1016/j.pneurobio.2015.09.012">10.1016/j.pneurobio.2015.09.012">10.1016/j.pneurobio.2015.09.012.
[11]
SULZER D, CASSIDY C, HORGA G, et al. Neuromelanin detection by magnetic resonance imaging (MRI) and its promise as a biomarker for Parkinson's disease[J/OL]. NPJ Parkinsons Dis, 2018, 4 [2022-07-10]. https://www.nature.com/articles/s41531-018-0047-3. DOI: 10.1038/s41531-018-0047-3.
[12]
WANG J, LI Y, HUANG Z, et al. Neuromelanin-sensitive magnetic resonance imaging features of the substantia nigra and locus coeruleus in de novo Parkinson's disease and its phenotypes[J/OL]. Eur J Neurol, 2018, 25(7): 949-e73 [2022-07-10]. https://onlinelibrary.wiley.com/doi/abs/10.1111/ene.13628. DOI: 10.1111/ene.13628">10.1111/ene.13628">10.1111/ene.13628.
[13]
SASAKI M, SHIBATA E, TOHYAMA K, et al. Neuromelanin magnetic resonance imaging of locus ceruleus and substantia nigra in Parkinson's disease[J]. Neuroreport, 2006, 17(11): 1215-1218. DOI: 10.1097/01.wnr.0000227984.84927.a7">10.1097/01.wnr.0000227984.84927.a7">10.1097/01.wnr.0000227984.84927.a7.
[14]
KIDOH M, SHINODA K, KITAJIMA M, et al. Deep learning based noise reduction for brain MR imaging: tests on phantoms and healthy volunteers[J]. Magn Reson Med Sci, 2020, 19(3): 195-206. DOI: 10.2463/mrms.mp.2019-0018">10.2463/mrms.mp.2019-0018">10.2463/mrms.mp.2019-0018.
[15]
YASAKA K, TANISHIMA T, OHTAKE Y, et al. Deep learning reconstruction for 1.5 T cervical spine MRI: effect on interobserver agreement in the evaluation of degenerative changes[J]. Eur Radiol, 2022, 32(9): 6118-6125. DOI: 10.1007/s00330-022-08729-z">10.1007/s00330-022-08729-z">10.1007/s00330-022-08729-z.
[16]
HAHN S, YI J, LEE H J, et al. Image quality and diagnostic performance of accelerated shoulder MRI with deep learning-based reconstruction[J]. AJR Am J Roentgenol, 2022, 218(3): 506-516. DOI: 10.2214/AJR.21.26577">10.2214/AJR.21.26577">10.2214/AJR.21.26577.
[17]
吴涛, 敖国昆, 许国宇, 等. 3.0T SE序列T1WI反转角对脑灰白质对比度影响的研究[J]. 实用放射学杂志, 2016, 32(1): 20-23. DOI: 10.3969/j.issn.1002-1671.2016.01.006">10.3969/j.issn.1002-1671.2016.01.006">10.3969/j.issn.1002-1671.2016.01.006.
WU T, AO G K, XU G Y, et al. Optimize the flip angle of SE sequence at 3.0T to increase gray-to-white matter contrast and decrease SAR value in brain MR examination[J]. J Pract Radiol, 2016, 32(1): 20-23. DOI: 10.3969/j.issn.1002-1671.2016.01.006">10.3969/j.issn.1002-1671.2016.01.006">10.3969/j.issn.1002-1671.2016.01.006.
[18]
PRASAD S, STEZIN A, LENKA A, et al. Three-dimensional neuromelanin-sensitive magnetic resonance imaging of the substantia nigra in Parkinson's disease[J]. Eur J Neurol, 2018, 25(4): 680-686. DOI: 10.1111/ene.13573">10.1111/ene.13573">10.1111/ene.13573.
[19]
TRUJILLO P, SUMMERS P E, FERRARI E, et al. Contrast mechanisms associated with neuromelanin-MRI[J]. Magn Reson Med, 2017, 78(5): 1790-1800. DOI: 10.1002/mrm.26584">10.1002/mrm.26584">10.1002/mrm.26584.
[20]
LIU Y, LI J C, HE N Y, et al. Optimizing neuromelanin contrast in the substantia nigra and locus coeruleus using a magnetization transfer contrast prepared 3D gradient recalled echo sequence[J/OL]. NeuroImage, 2020, 218: 116935 [2022-07-10]. https://www.sciencedirect.com/science/article/pii/s1053811920304213. DOI: 10.1016/j.neuroimage.2020.116935">10.1016/j.neuroimage.2020.116935">10.1016/j.neuroimage.2020.116935.
[21]
OGAWA R, KIDO T, NAKAMURA M, et al. Reconstruction of cardiovascular black-blood T2-weighted image by deep learning algorithm: a comparison with intensity filter[J/OL]. Acta Radiol Open, 2021, 10(9): 20584601211044779 [2022-07-10]. https://journals.sagepub.com/doi/full/10.1177/20584601211044779. DOI: 10.1177/20584601211044779">10.1177/20584601211044779">10.1177/20584601211044779.
[22]
SREEKUMARI A, SHANBHAG D, YEO D, et al. A deep learning-based approach to reduce rescan and recall rates in clinical MRI examinations[J]. AJNR Am J Neuroradiol, 2019, 40(2): 217-223. DOI: 10.3174/ajnr.A5926">10.3174/ajnr.A5926">10.3174/ajnr.A5926.
[23]
YI D, GRØVIK E, TONG E, et al. MRI pulse sequence integration for deep-learning-based brain metastases segmentation[J]. Med Phys, 2021, 48(10): 6020-6035. DOI: 10.1002/mp.15136">10.1002/mp.15136">10.1002/mp.15136.
[24]
KIM M, KIM H S, KIM H J, et al. Thin-slice pituitary MRI with deep learning-based reconstruction: diagnostic performance in a postoperative setting[J]. Radiology, 2021, 298(1): 114-122. DOI: 10.1148/radiol.2020200723">10.1148/radiol.2020200723">10.1148/radiol.2020200723.
[25]
MUSCOGIURI G, MARTINI C, GATTI M, et al. Feasibility of late gadolinium enhancement (LGE) in ischemic cardiomyopathy using 2D-multisegment LGE combined with artificial intelligence reconstruction deep learning noise reduction algorithm[J]. Int J Cardiol, 2021, 343: 164-170. DOI: 10.1016/j.ijcard.2021.09.012">10.1016/j.ijcard.2021.09.012">10.1016/j.ijcard.2021.09.012.
[26]
KOCH K M, SHERAFATI M, ARPINAR VE, et al. Analysis and evaluation of a deep learning reconstruction approach with denoising for orthopedic MRI[J/OL]. Radiol Artif Intell, 20213(6): e200278 [2022-07-10]. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8637471/. DOI: 10.1148/ryai.2021200278.
[27]
SUN S, TAN E T, MINTZ D N, et al. Evaluation of deep learning reconstructed high-resolution 3D lumbar spine MRI[J]. Eur Radiol, 2022, 32(9): 6167-6177. DOI: 10.1007/s00330-022-08708-4">10.1007/s00330-022-08708-4">10.1007/s00330-022-08708-4.
[28]
ZOCHOWSKI K C, TAN E T, ARGENTIERI E C, et al. Improvement of peripheral nerve visualization using a deep learning-based MR reconstruction algorithm[J]. Magn Reson Imaging, 2022, 85: 186-192. DOI: 10.1016/j.mri.2021.10.038">10.1016/j.mri.2021.10.038">10.1016/j.mri.2021.10.038.
[29]
WANG X Z, MA J F, BHOSALE P, et al. Novel deep learning-based noise reduction technique for prostate magnetic resonance imaging[J]. Abdom Radiol (NY), 2021, 46(7): 3378-3386. DOI: 10.1007/s00261-021-02964-6">10.1007/s00261-021-02964-6">10.1007/s00261-021-02964-6.
[30]
Schmoeller M, Stadter C, Kick MK, et al. A novel approach to the holistic 3D characterization of weld seams—paving the way for deep learning-based process monitoring[J/OL]. Materials (Basel), 2021, 14(22): 6928 [2022-07-10]. https://www.mdpi.com/1996-1944/14/22/6928. DOI: 10.3390/ma14226928">10.3390/ma14226928">10.3390/ma14226928.

上一篇 深度学习MRI重建算法的临床应用和发展前景
下一篇 深度学习在颅脑体素内不相干运动磁共振成像质量中的应用
  
诚聘英才 | 广告合作 | 免责声明 | 版权声明
联系电话:010-67113815
京ICP备19028836号-2