分享:
分享到微信朋友圈
X
特别关注
深度学习重建法提高磁共振高分辨海马冠状位图像质量的比较研究
杨晶 李琼阁 吴涛 齐志刚 赵澄 卢洁

Cite this article as: YANG J, LI Q G, WU T, et al. A comparative study on the enhancement of high resolution coronal image quality by deep learning reconstruction[J]. Chin J Magn Reson Imaging, 2023, 14(5): 21-24, 30.本文引用格式:杨晶, 李琼阁, 吴涛, 等. 深度学习重建法提高磁共振高分辨海马冠状位图像质量的比较研究[J]. 磁共振成像, 2023, 14(5): 21-24, 30. DOI:10.12015/issn.1674-8034.2023.05.005.


[摘要] 目的 探讨深度学习重建(deep learning reconstruction, DLR)在提高高分辨率T2液体衰减反转恢复(fluid-attenuated inversion-recovery, FLAIR)序列海马冠状位MRI图像质量中的作用。材料与方法 前瞻性纳入36例神经系统疾病患者,进行高分辨率海马T2-FLAIR冠状位扫描,并对图像进行DLR,对原始重建(origin reconstruction, OR)图像和DLR图像的噪声、伪影、海马结构辨识度、病灶可识别度和诊断接受度进行主观评价,测量并计算两组图像的信噪比、对比噪声比和双侧海马信号强度差值。结果 DLR的T2-FLAIR海马冠状位图像噪声、海马结构辨识度、病灶可识别度和诊断接受评分均高于OR T2-FLAIR海马冠状位(P<0.001),伪影评分差异无统计学意义(Z=-1.730;P=0.084);DLR T2-FLAIR海马冠状位图像的信噪比和对比噪声比均明显高于OR T2-FLAIR海马冠状位(t=-13.061;P<0.001和t=-16.224;P<0.001);两组图像双侧海马信号差值差异无统计学意义(t=-0.290;P=0.977)。结论 DLR不延长扫描时间就可以明显提高高分辨T2-FLAIR海马冠状位图像的海马结构和小病灶的清晰度,降低噪声,为临床诊断提供高质量图像。
[Abstract] Objective To investigate the role of deep learning reconstruction (DLR) in improving image quality of high resolution T2 fluid-attenuated inversion-recovery sequency (FLAIR) coronal magnetic resonance imaging of hippocampus.Materials and Methods A total of 36 patients were prospectively enrolled in this study. GE Singna Premier 3.0 T MR scanner was used to perform high resolution T2-FLAIR coronal scanning of the hippocampus. Two diagnosticians assessed the image noise, artifacts, hippocampal structure, lesion, and diagnostic acceptance of origin reconstruction (OR) T2-FLAIR and DLR T2-FLAIR coronal images, respectively. Two technologists measured and calculated the signal to noise ratio (SNR), contrast noise ratio (CNR), and the difference in bilateral hippocampal signal intensity between the two groups of images.Results The noise, hippocampal structure identification, lesion identification and diagnostic acceptance score from DLR based coronal T2-FLAIR hippocampal image were higher than OR T2-FLAIR hippocampal coronal image (P<0.001), and there was no significant difference in artifacts evaluation (Z=-1.730; P=0.084). The SNR and CNR values of DLR T2-FLAIR hippocampal coronal images were significantly higher than those of OR T2-FLAIR hippocampal coronal images (t=-13.061, P<0.001; t=16.224; P<0.001). There was no significant difference in the difference of bilateral hippocampal signal between the two groups (t=-0.290; P=0.977).Conclusions Without prolonging the scanning time, Deep learning reconstruction can significantly improve the resolution of hippocampal structures and small lesions in high resolution T2-FLAIR coronal images. It can reduce noise, and provide high quality images for clinical use.
[关键词] 海马;深度学习重建法;癫痫;信噪比;对比噪声比;磁共振成像
[Keywords] hippocampus;deep learning reconstruction;epilepsy;signal to noise ratio;contrast noise ratio;magnetic resonance imaging

杨晶 1, 2   李琼阁 1, 2   吴涛 3   齐志刚 1, 2   赵澄 1, 2   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;杨晶起草和撰写稿件,获取、分析或解释本研究的数据;李琼阁、吴涛、齐志刚、赵澄获取、分析或解释本研究的数据,对稿件重要内容进行了修改;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。


收稿日期:2022-07-02
接受日期:2022-10-12
中图分类号:R445.2 
文献标识码:A
DOI: 10.12015/issn.1674-8034.2023.05.005
本文引用格式:杨晶, 李琼阁, 吴涛, 等. 深度学习重建法提高磁共振高分辨海马冠状位图像质量的比较研究[J]. 磁共振成像, 2023, 14(5): 21-24, 30. DOI:10.12015/issn.1674-8034.2023.05.005.

0 前言

       海马体位于大脑丘脑和内侧颞叶之间,上方是侧脑室颞角,下方是海马旁回,是学习和短期记忆的重要脑功能区[1]。海马体积[2, 3]、信号强度[4, 5]、海马趾和黑带的缺失[6-7]对临床诊断癫痫、认知障碍和精神分裂症有一定的影响。OPPENHEIM等[8]在193例难治性癫痫患者采用1.5 T MR发现海马头部的海马趾缺失作为诊断颞叶内侧硬化的敏感度为92%,特异度为100%。HOWE等[9]和HUESMANN等[10]均发现3 T MR T2图像上黑带的缺失预示着组织的缺失,是诊断颞叶内侧硬化的指标。因此,提高海马体MRI分辨率能够为临床诊断提供有效的诊断依据。

       高分辨率T2液体衰减反转恢复(fluid-attenuated inversion-recovery, FLAIR)序列海马体冠状位成像是观察海马信号强度和内部结构最好的影像学检查手段之一[11],临床常规序列扫描的图像只能看到沿着海马沟的黑带,不能看到海马体内部以及海马旁回的各个层面。应用7 T MR可以在T2加权成像上发现癫痫患者海马信号升高与海马内部异常卷曲、海马细胞外角质层异常增生相关联[12, 13]。同时,应用9.4 T MR将T2-FLAIR海马冠状位的分辨率提高到0.6 mm/pixel及以下时可以清楚地观察到海马趾[6]。但是7 T和9.4 T MR还没有完全应用于临床,而临床常规3 T MR T2-FLAIR海马体冠状位成像分辨率多为1 mm/pixel,如提高分辨率,就会增加扫描时间。尽管压缩感知、并行采集和多层采集等技术可以降低扫描时间,但是增加的噪声会严重降低图像分辨率,影响临床诊断。

       MRI的原始重建(origin reconstruction, OR)法主要通过滤波来降低噪声,但同时会损失部分组织细节,容易导致图像模糊。深度学习重建(deep learning reconstruction, DLR)是一种新的基于人工智能的重建算法[14, 15, 16]。DLR利用有监督的卷积神经网络对近乎完美(层内高分辨率和无噪声)的医学图像和OR图像进行学习,对传统重建图像降噪处理,避免直接生成过度降低噪声的图像,使图像更加自然,去除截断伪影,提高图像锐利度[17, 18]。目前其在脊柱[19]、心脏[20, 21]、前列腺[22]和脑垂体[23, 24]的MRI已有相关应用,但有关采用DLR提高海马体成像质量的相关研究很少。

       本研究目的是探讨DLR是否可以提高高分辨T2-FLAIR海马体成像的图像质量,观察到更多海马体内部结构,为临床诊断海马体疾病提供更多的帮助。

1 材料与方法

1.1 研究对象

       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,全体受试者均签署知情同意书。前瞻性纳入2022年5月20日至6月7日首都医科大学宣武医院神经系统疾病患者的头部MRI资料。纳入标准:(1)初步诊断为神经系统疾病;(2)未接受过手术、穿刺和放化疗。排除标准:(1)MRI检查禁忌证或患者躁动无法完成扫描;(2)图像质量差无法进行评估。

1.2 仪器与方法

       所有受试者均采用3.0 T MR扫描仪(SIGNA Premier, GE Healthcare,Milwaukee,USA)进行检查,高分辨率海马冠状位成像采用T2-FLAIR序列,扫描使用48通道头线圈(原机自带),具体扫描参数为:FOV 18 cm×18 cm,矩阵300×300,TR 8500 ms,TE 96.7 ms,TI 2600 ms,NEX 1,加速因子1.5,层厚2.5 mm,分辨率0.6×0.6×2.5 mm3/pixel。扫描方向垂直于海马体,从海马头扫描到海马尾,共20层,扫描时间2 min 59 s,同时使用深度学习算法进行重建,最终获得原始图像和DLR后的图像。

1.3 图像评价

       MRI图像的调阅、判读和测量在RadiAnt DICOM Viewer [Version:2021.2.2(64-bit)]上进行。应用统一的窗宽和窗中心(窗宽1400,窗中心800)测量评价。

       由两名具有20年中枢神经系统诊断经验的副主任医师采用盲法对OR T2-FLAIR和DLR T2-FLAIR冠状位图像的噪声、伪影、海马结构辨识度、病灶可识别度和诊断接受程度进行评级[22],结果见表1

       由两名有5年临床经验的主管技师分别测量OR T2-FLAIR和DLR T2-FLAIR冠状位图像双侧海马体头部的信号和噪声强度,同时测量同层面右侧皮层下大脑白质信号强度和噪声,以上感兴趣区均为像素200~250 Pix之间的圆形范围。感兴趣区范围排除海马头黑带、海马趾和脑白质区的病灶。计算右侧海马体信噪比(signal to noise ratio, SNR)和对比噪声比(contrast to noise ratio, CNR)。SNR=右侧海马体信号值/右侧海马体噪声值;CNR=(右侧海马体信号值-右侧脑白质信号值)/右侧脑白质噪声值。计算OR T2-FLAIR和DLR T2-FLAIR海马冠状位图像双侧海马头部的信号强度差值。

表1  图像主观评分标准
Tab. 1  Subjective rating criteria for images

1.4 统计学分析

       采用SPSS 20.0进行统计学分析。计量资料若符合正态分布,用(x¯±s)表示,采用配对样本t检验,若不符合正态分布,用MQ1,Q3)表示,采用Wilcoxon秩和检验。两名诊断医师主观评分结果如出现不一致,共同商量决定评分结果,并对两名技师分别测量和计算的SNR、CNR和双侧海马信号强度差值采用组内相关系数(intra-class correlation coefficient, ICC)进行一致性检验。

2 结果

2.1 患者资料

       排除2例海马切除术后和1例躁动无法完成检查的患者后,最终有36例纳入研究。其中男13例,女23例,年龄7~88(51.36±17.22)岁。其中21例为双侧额叶皮层下缺血灶,伴或不伴脑白质变性,7例为正常患者,1例为脑内多发腔梗死灶,1例为双侧额叶癫痫病灶,1例为右侧小脑半球微出血,1例为左侧额叶出血,1例为右侧额叶动静脉畸形,1例为双侧颞角略大,1例为双侧海马体积明显缩小,两侧颞角及脉络膜列明显增宽,1例为两侧侧脑室周围脑白质内斑片状异常信号。

2.2 图像质量主观评价

       如图1表2所示,相较于OR T2-FLAIR成像,DLR T2-FLAIR海马冠状位图像的噪声少,海马轮廓清晰,海马趾结构略可见,>3 mm和≤3 mm病灶均接近于清晰可见。且OR T2-FLAIR海马冠状位图像诊断接受度为平均水平,DLR T2-FLAIR海马冠状位图像诊断接受度特别好。OR T2-FLAIR和DLR T2-FLAIR海马冠状位图像的伪影都接近于没有伪影的较高分数,差异无统计学意义。

图1  OR T2-FLAIR和DLR T2-FLAIR海马冠状位图像对比。1A、1B:女,46岁,癫痫;1A:OR T2-FLAIR海马冠状位图像,噪声略低于平均噪声水平,海马轮廓可见,海马黑带可见,海马趾不可见,右侧海马信号高于左侧,强度差值为113.11,诊断接受度为平均水平;1B:DLR T2-FLAIR海马冠状位图像,只有少量噪声,海马轮廓清晰,海马黑带可见,海马趾欠清晰,右侧海马信号高于左侧,强度差值为94.17,诊断接受度为很好。1C、1D:女,57岁,脑内多发梗死灶;1C:OR T2-FLAIR海马冠状位图像,颅底轻度伪影不影响解剖结构观察(宽箭)右侧皮层下和左侧放射冠区(细箭)>3 mm梗死灶可见,双侧额顶叶皮层下≤3 mm梗死灶(箭头)欠清晰,诊断接受度为平均水平;1D:DLR T2-FLAIR海马冠状位图像,颅底轻度伪影不影响解剖结构观察(宽箭),>3 mm(细箭)和≤3 mm梗死灶(箭头)均清晰可见,诊断接受度为很好。1E、1F:男,33岁,记忆力下降;1E:OR T2-FLAIR海马冠状位图像,噪声明显低于平均水平,运动伪影明显可见,≤3 mm病灶(箭头)略可见。1F:DLR T2-FLAIR
Fig. 1  Contrast of coronal images of the hippocampus with origin reconstruction (OR) T2-fluid-attenuated inversion-recovery (FLAIR)sequency and deep learning reconstruction (DLR) T2-FLAIR. 1A-1B: Female patient, 46 years old, epileptic.1A: OR T2-FLAIR hippocampal coronal image, image noise is slightly lower than average noise level, the outline of hippocampal is visible,hippocampal dark band is visible, hippocampal digitations are not visible, right hippocampal signal higher than left, the intensity difference is 113.11, the diagnostic acceptance is average; 1B: DLR T2-FLAIR coronal image of the hippocampus, with only a small amount of noise, the outline of hippocampal is clear, dark band of the hippocampus is visible clearly, hippocampus digitations is less clear, the signal on the right side of the hippocampus is higher than that on the left side, the intensity difference is 94.17, the diagnostic acceptance is very good. 1C-1D: Female patient, 57 years old, multiple cerebral infarction foci. 1C: OR T2-FLAIR coronal image of the hippocampus, mild artifacts of the skull base do not affect anatomical observation (broad arrow). Infarction focis >3 mm in the right subcortical and left radiographic coronal area (thin arrow) are visible, and infarction focis ≤3 mm in bilateral frontal parietal cortex (arrow head) are not clear,the diagnostic acceptance is average. 1D: DLR T2-FLAIR coronal image of the hippocampus, mild artifacts of the skull base do not affect anatomical observations (broad arrow), infarcts >3 mm (thin arrow) and ≤3 mm (arrow head) are both seen clearly, the diagnostic acceptance is very good.1E-1F: Male patient, 33 years old, memory loss. 1E: OR T2-FLAIR hippocampal coronal image shows significantly lower than average noise, visible motion artifacts, and lesions ≤3 mm (arrow heads) is lightly visible. 1F: DLR T2-FLAIR coronal image of the hippocampus shows significantly less noise, visible motion artifacts, lesions ≤3 mm (arrow head) is clearly visible.
表2  OR T2-FLAIR和DLR T2-FLAIR冠状位图像主观评分
Tab. 2  Subjective scores of OR T2-FLAIR and DLR T2-FLAIR coronal images

2.3 客观评价结果

       组内相关性分析认为ICC值大于0.75证明数据一致性较好。两名技师分别测得图像的SNR和CNR具有很好的一致性,ICC值分别为0.93(95% CI:0.88~0.97)和0.89(95% CI:0.73~0.93),两名技师所测双侧海马信号强度差值具有很好的一致性,ICC值为0.82(95% CI:0.78~0.85)。因此采用两名医师所测图像的SNR、CNR和双侧海马信号强度差值的平均值做统计学分析。如表3所示,DLR T2-FLAIR海马冠状位图像的SNR值和CNR值都明显高于OR T2-FLAIR,两者之间差异均有统计学意义(P<0.05);OR T2-FLAIR和DLR T2-FLAIR海马冠状位图像的双侧海马信号强度差异无统计学意义,说明DLR对图像信号强度没有统计学影响,不影响左右海马信号强度比较的评价。

表3  OR T2-FLAIR和DLR T2-FLAIR冠状位图像质量分析
Tab. 3  Quality analysis of OR T2-FLAIR and DLR T2-FLAIR coronal images

3 讨论

       本研究首次在海马冠状位高分辨MRI中应用深度学习算法进行重建,并与OR进行比较,发现图像分辨率提高到0.6 mm/pixel和不延长扫描时间的前提下,可以明显降低图像噪声,更清楚观察海马头部海马趾和黑带,明显提高海马内部细节和外周病灶可识别度,并且在不影响双侧海马信号强度差值的前提下,明显提高图像SNR和CNR,为临床3 T MRI海马疾病诊断提供更多可能。

3.1 图像伪影

       本研究主观评价伪影评分在OR T2-FLAIR和DLR T2-FLAIR海马冠状位之间评分差异无统计学意义。而HAHN等[17]研究发现DLR可以减少肩关节快速MRI伪影,这与本研究结果不一致。但值得注意的是,van der VELDE等[25]研究发现DLR会使心脏MRI扫描强化后延时期的卷折和伪影更明显。我们推测,造成这种差异的原因在于头部相比肩关节和心脏,呼吸和心脏搏动影响较小,产生运动伪影较少。但也有研究说明DLR没有直接减少运动伪影[26],这一观点还有待进一步验证。

3.2 双侧海马信号强度差值的影响

       临床上双侧海马信号强度差异是诊断单侧海马萎缩和难治性癫痫的重要指标[27, 28, 29, 30, 31, 32]。本研究中OR T2-FLAIR和DLR T2-FLAIR海马冠状位双侧海马信号差值之间差异无统计学意义,说明DLR只提高图像质量,不改变图像信号的强弱,对双侧海马信号强度对比没有影响。更进一步表明DLR算法在临床应用的可靠性。

3.3 图像质量的影响

       本研究中,OR T2-FLAIR和DLR T2-FLAIR海马冠状位SNR值分别为9.03±1.43和20.40±6.15(P<0.001),CNR值分别为4.97±1.05 和13.02±3.35(P<0.001)。有研究[22]将深度学习法应用于109例患者的前列腺T2加权成像,其应用的DLR前、后快速MRI序列SNR值分别为8.8±4.9和14.7±6.8(P<0.001),CNR值分别为3.4±3.6和6.5±6.3(P<0.001)。这与本研究结果基本类似,均能表明DLR对MRI图像质量的提升,有助于临床诊断。

3.4 局限性分析

       本研究仍存在一定局限性。首先,本研究海马冠状位采用扫描层厚是2.5 mm,如能保证图像高分辨不变,进一步降低层厚,有望观察更多的海马结构,但同时扫描时间有可能也会延长。此外,本研究纳入病例较少,以后可纳入更多病例并纳入不同种类的患者进行对比分析,如癫痫、阿尔茨海默病和精神分裂患者等。

4 结论

       综上所述,DLR在不延长扫描时间的前提下,可明显提高高分辨T2-FLAIR海马冠状位的图像质量,为临床应用3 T MRI进行海马疾病的影像学诊断提供更有价值的信息。

[1]
HASSANKHANI A, STEIN J M, HABOOSHEH A G, et al. Anatomical variations, mimics, and pitfalls in imaging of patients with epilepsy[J]. J Neuroimaging, 2021, 31(1): 20-34. DOI: 10.1111/jon.12809">10.1111/jon.12809">10.1111/jon.12809.
[2]
REBSAMEN M, RADOJEWSKI P, MCKINLEY R, et al. A quantitative imaging biomarker supporting radiological assessment of hippocampal sclerosis derived from deep learning-based segmentation of T1W-MRI[J/OL]. Front Neurol, 2022, 13: 812432 [2022-07-01]. https://boris.unibe.ch/166750/. DOI: 10.3389/fneur.2022.812432">10.3389/fneur.2022.812432">10.3389/fneur.2022.812432.
[3]
PAN K C, ZHAO L, GU S, et al. Deep learning-based automatic delineation of the hippocampus by MRI: geometric and dosimetric evaluation[J]. Radiat Oncol, 2021, 16(1): 12. DOI: 10.1186/s13014-020-01724-y">10.1186/s13014-020-01724-y">10.1186/s13014-020-01724-y.
[4]
ITO Y, Fukuda M, Matsuzawa H, et al. Deep learning-based diagnosis of temporal lobe epilepsy associated with hippocampal sclerosis: an MRI study[J/OL]. Epilepsy Res, 2021, 178: 106815 [2022-07-01]. https://www.sciencedirect.com/science/article/abs/pii/S0920121121002709. DOI: 10.1016/j.eplepsyres.2021.106815">10.1016/j.eplepsyres.2021.106815">10.1016/j.eplepsyres.2021.106815.
[5]
LI W, JIANG Y C, QIN Y J, et al. Structural and functional reorganization of contralateral hippocampus after temporal lobe epilepsy surgery[J/OL]. Neuroimage Clin, 2021, 31: 102714 [2022-07-01]. https://www.sciencedirect.com/science/article/pii/S2213158221001583. DOI: 10.1016/j.nicl.2021.102714">10.1016/j.nicl.2021.102714">10.1016/j.nicl.2021.102714.
[6]
CHANG C, HUANG C, ZHOU N Y, et al. The bumps under the hippocampus[J]. Hum Brain Mapp, 2018, 39(1): 472-490. DOI: 10.1002/hbm.23856">10.1002/hbm.23856">10.1002/hbm.23856.
[7]
KATABATHULA S, WANG Q Y, XU R. Predict Alzheimer's disease using hippocampus MRI data: a lightweight 3D deep convolutional network model with visual and global shape representations[J]. Alzheimers Res Ther, 2021, 13(1): 104. DOI: 10.1186/s13195-021-00837-0.
[8]
OPPENHEIM C, DORMONT D, BIONDI A, et al. Loss of digitations of the hippocampal head on high-resolution fast spin-echo MR: a sign of mesial temporal sclerosis[J]. AJNR Am J Neuroradiol, 1998, 19(3): 457-463.
[9]
HOWE K L, DIMITRI D, HEYN C, et al. Histologically confirmed hippocampal structural features revealed by 3T MR imaging: potential to increase diagnostic specificity of mesial temporal sclerosis[J]. AJNR Am J Neuroradiol, 2010, 31(9): 1682-1689. DOI: 10.3174/ajnr.A2154.
[10]
HUESMANN G R, SCHWARB H, SMITH D R, et al. Hippocampal stiffness in mesial temporal lobe epilepsy measured with MR elastography: preliminary comparison with healthy participants[J/OL]. Neuroimage Clin, 2020, 27: 102313 [2022-07-01]. https://www.sciencedirect.com/science/article/pii/s2213158220301509. DOI: 10.1016/j.nicl.2020.102313.
[11]
WELLMER J, QUESADA C M, ROTHE L, et al. Proposal for a magnetic resonance imaging protocol for the detection of epileptogenic lesions at early outpatient stages[J]. Epilepsia, 2013, 54(11): 1977-1987. DOI: 10.1111/epi.12375.
[12]
OPHEIM G, VAN DER KOLK A, MARKENROTH BLOCH K, et al. 7T epilepsy task force consensus recommendations on the use of 7T MRI in clinical practice[J]. Neurology, 2021, 96(7): 327-341. DOI: 10.1212/WNL.0000000000011413.
[13]
GUYOT A, FOUQUIER A B G, GERARDIN E, et al. A diffeomorphic vector field approach to analyze the thickness of the Hippocampus from 7 T MRI[J]. IEEE Trans Biomed Eng, 2021, 68(2): 393-403. DOI: 10.1109/TBME.2020.2999941.
[14]
ZENG G S, GUO Y, ZHAN J Y, et al. A review on deep learning MRI reconstruction without fully sampled k-space[J]. BMC Med Imaging, 2021, 21(1): 195. DOI: 10.1186/s12880-021-00727-9.
[15]
LIU H, LIU J, LI J, et al. DL-MRI: a unified framework of deep learning-based MRI super resolution[J/OL]. J Healthc Eng, 2021, 2021: 5594649 [2022-07-01]. https://www.hindawi.com/journals/jhe/2021/5594649/. DOI: 10.1155/2021/5594649.
[16]
EDUPUGANTI V, MARDANI M, VASANAWALA S, et al. Uncertainty quantification in deep MRI reconstruction[J]. IEEE Trans Med Imaging, 2021, 40(1): 239-250. DOI: 10.1109/TMI.2020.3025065.
[17]
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.
[18]
UEDA T, OHNO Y, YAMAMOTO K, et al. Deep learning reconstruction of diffusion-weighted MRI improves image quality for prostatic imaging[J]. Radiology, 2022, 303(2): 373-381. DOI: 10.1148/radiol.204097.
[19]
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.
[20]
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 [2022-07-01]. https://journals.sagepub.com/doi/pdf/10.1177/20584601211044779. DOI: 10.1177/20584601211044779.
[21]
SANDINO C M, LAI P, VASANAWALA S S, et al. Accelerating cardiac cine MRI using a deep learning-based ESPIRiT reconstruction[J]. Magn Reson Med, 2021, 85(1): 152-167. DOI: 10.1002/mrm.28420.
[22]
PARK J C, PARK K J, PARK M Y, et al. Fast T2-weighted imaging with deep learning-based reconstruction: evaluation of image quality and diagnostic performance in patients undergoing radical prostatectomy[J]. J Magn Reson Imaging, 2022, 55(6): 1735-1744. DOI: 10.1002/jmri.27992.
[23]
Lee DH, Park JE, Nam YK, et al. Deep learning-based thin-section MRI reconstruction improves tumour detection and delineation in pre- and post-treatment pituitary adenoma[J/OL]. Sci Rep, 2021, 11(1): 21302 [2022-07-01]. https://www.nature.com/articles/s41598-021-00558-2. DOI: 10.1038/s41598-021-00558-2.
[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.
[25]
Van der VELDE N, HASSING H C, BAKKER B J, et al. Improvement of late gadolinium enhancement image quality using a deep learning-based reconstruction algorithm and its influence on myocardial scar quantification[J]. Eur Radiol, 2021, 31(6): 3846-3855. DOI: 10.1007/s00330-020-07461-w.
[26]
LEBEL R M. Performance characterization of a novel deep learning-based MR image reconstruction pipeline[EB/OL]. 2020: arXiv: 2008.06559 [2022-07-01]. https://arxiv.org/abs/2008.06559
[27]
LIAO C Y, WANG K, CAO X Z, et al. Detection of lesions in mesial temporal lobe epilepsy by using MR fingerprinting[J]. Radiology, 2018, 288(3): 804-812. DOI: 10.1148/radiol.2018172131.
[28]
CHOI J Y, KRISHNAN B, HU S Y, et al. Using magnetic resonance fingerprinting to characterize periventricular nodular heterotopias in pharmacoresistant epilepsy[J]. Epilepsia, 2022, 63(5): 1225-1237. DOI: 10.1111/epi.17191.
[29]
JORDAN S P, HU S Y, ROZADA I, et al. Automated design of pulse sequences for magnetic resonance fingerprinting using physics-inspired optimization[J/OL]. Proc Natl Acad Sci USA, 2021, 118(40): e2020516118 [2022-07-01]. https://www.pnas.org/doi/abs/10.1073/pnas.2020516118. DOI: 10.1073/pnas.2020516118.
[30]
CARDOSO G T M, GOMES-LEAL W, FRANCO E C S, et al. Compensatory hippocampal neurogenesis in the absence of cognitive impairment following experimental hippocampectomy in adult rats[J/OL]. Front Cell Neurosci, 2021, 15: 709291 [2022-07-01]. https://www.frontiersin.org/articles/10.3389/fncel.2021.709291/full. DOI: 10.3389/fncel.2021.709291.
[31]
TIPPAREDDY C, ZHAO W, SUNSHINE J L, et al. Magnetic resonance fingerprinting: an overview[J]. Eur J Nucl Med Mol Imaging, 2021, 48(13): 4189-4200. DOI: 10.1007/s00259-021-05384-2.
[32]
MOSTARDEIRO T R, PANDA A, WITTE R J, et al. Whole-brain 3D MR fingerprinting brain imaging: clinical validation and feasibility to patients with meningioma[J]. Magn Reson Mater Phy, 2021, 34(5): 697-706. DOI: 10.1007/s10334-021-00924-1.

上一篇 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
下一篇 基于多中心MRI的3D-ResNet101深度学习模型预测脑胶质瘤术前分级的研究
  
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