分享:
分享到微信朋友圈
X
特别关注
基于磁共振深度学习图像重建算法的心肌延迟强化在未识别心肌梗死中的应用
陆雪芳 闫玉辰 龚威 权光南 刘薇音 查云飞

Cite this article as: LU X F, YAN Y C, GONG W, et al. Feasibility study of deep learning-based MRI image reconstruction algorithms for myocardial delayed enhancement in unrecognized myocardial infarction[J]. Chin J Magn Reson Imaging, 2024, 15(10): 8-14, 49.本文引用格式:陆雪芳, 闫玉辰, 龚威, 等. 基于磁共振深度学习图像重建算法的心肌延迟强化在未识别心肌梗死中的应用[J]. 磁共振成像, 2024, 15(10): 8-14, 49. DOI:10.12015/issn.1674-8034.2024.10.003.


[摘要] 目的 探讨基于磁共振深度学习重建(deep learning reconstruction, DLR)算法的心肌延迟强化(late gadolinium enhancement, LGE)提高临床未识别心肌梗死(unrecognized myocardial infarction, UMI)患者识别率的诊断价值。材料与方法 前瞻性纳入2022年4月至2023年8月于我院就诊的可疑UMI并完成心脏磁共振(cardiac magnetic resonance, CMR)检查的患者98例,分析常规重建的原始LGE(original LGE, LGEO)及通过DLR算法获得的LGE(DLR LGE, LGEDL)短轴位图像。测量两组图像心肌的信噪比(signal-to-noise ratio, SNR)、对比噪声比(contrast-to-noise ratio, CNR)、标准差(standard deviation, SD)。采用2~5倍标准差(2SD~5SD)法以及全宽半高(full width at half maximum, FWHM)法进行强化面积百分比(percentage of enhanced area, Parea)分析。评估心肌整体各测量值的组内及组间一致性。以临床诊断UMI为金标准计算LGEDL和LGEO的诊断效能。结果 SNRDL和CNRDL均较SNRO和CNRO提高超过2倍(P<0.001)。LGEDL序列心肌、强化灶及背景噪声SDDL较SDO均显著减低(P均<0.001)。不同阈值法分析,Parea-DL较Parea-O序列升高,以2SD法为著(P<0.001),但FWHM法中差异无统计学意义(P>0.05)。整体心肌SNR、CNR、不同阈值法Parea测量值组内及组间一致性好[组内相关系数(intra-class correlation coefficient, ICC)>0.600,P均<0.001]。受试者工作特征(receiver operating characteristic, ROC)曲线分析显示,每种SD方法在检测UMI方面都表现出良好的诊断效果,其中以5SD Parea-DL法最优(P<0.001)。结论 DLR能显著提高LGE的图像质量。当 LGEDL 和 LGEO 的信号阈值与参考平均值(signal threshold versus reference mean, STRM)分别≥4SD和≥3SD时,对UMI的诊断效果较好。
[Abstract] Objective To investigate the diagnostic value of deep learning reconstruction (DLR)-based late gadolinium enhancement (LGEDL) in improving the recognition rate of unrecognized myocardial infarction (UMI).Materials and Methods This prospective study included 98 patients with suspected UMI who underwent cardiac magnetic resonance imaging (CMR) at our hospital from April 2022 to August 2023. Original LGE of conventional reconstruction (LGEO) and LGEDL images were obtained using conventional and commercially available inline DLR algorithms. The myocardial signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and percentage of enhanced area (Parea) were analysed using the standard deviation (SD) threshold approach (2SD-5SD) and full width at half maximum (FWHM) method. The diagnostic efficacies based on LGEDL and LGEO images were calculated.Results The SNRDL and CNRDL were two times better than the SNRO and CNRO, respectively (P<0.001). Parea-DL was higher than Parea-O, especially in the 2SD method (P<0.001). However, there was no intergroup difference based on the FWHM method (P>0.05). Overall myocardial SNR, CNR, and Parea measurements with different threshold methods had good intra- and interobserver agreement [intra-class correlation coefficient (ICC)>0.600, all P<0.001]. The receiver operating characteristic curve analysis revealed that each SD method exhibited good diagnostic efficacy for detecting UMI, with the Parea-DL having the best diagnostic efficacy based on the 5SD method (P<0.001). Overall, the LGEDL images had better image quality. Strong diagnostic efficacy for UMI identification was achieved when the signal threshold versus reference mean (STRM) was ≥4SD and ≥3SD for the LGEDL and LGEO, respectively.Conclusions STRM selection for LGEDL magnetic resonance images helps improve clinical decision-making in patients with UMI.
[关键词] 未识别心肌梗死;诊断效能;深度学习重建;钆延迟强化;磁共振成像
[Keywords] unrecognized myocardial infarction;diagnostic efficacy;deep learning reconstruction;late gadolinium enhancement;magnetic resonance imaging

陆雪芳 1   闫玉辰 1   龚威 1   权光南 2   刘薇音 2   查云飞 1*  

1 武汉大学人民医院放射科,武汉 430030

2 通用电气医疗(中国)有限公司,北京 100176

通信作者:查云飞,E-mail: zhayunfei999@126.com

作者贡献声明:查云飞设计本研究的方案,对稿件重要内容进行了修改;陆雪芳起草和撰写稿件,获取、分析和解释本研究的数据;闫玉辰、龚威、权光南、刘薇音获取、分析和解释本研究的数据,对稿件重要内容进行了修改;查云飞获得了国家自然科学基金项目资助;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: 国家自然科学基金项目 82171895
收稿日期:2024-02-03
接受日期:2024-07-05
中图分类号:R445.2  R542.2 
文献标识码:A
DOI: 10.12015/issn.1674-8034.2024.10.003
本文引用格式:陆雪芳, 闫玉辰, 龚威, 等. 基于磁共振深度学习图像重建算法的心肌延迟强化在未识别心肌梗死中的应用[J]. 磁共振成像, 2024, 15(10): 8-14, 49. DOI:10.12015/issn.1674-8034.2024.10.003.

0 引言

       目前心肌梗死(myocardial infarction, MI)的诊断主要依赖于检测急性心梗时因心肌损伤导致的心脏生物标志物异常[1]。然而,无典型心脏症状的MI占所有MI的一半[2]。研究报道,在没有冠状动脉疾病临床病史的心源性猝死患者中,42.4%的患者是未识别心肌梗死(unrecognized myocardial infarction, UMI)[3]。近年来,UMI的发病率逐年升高,且常由于症状不典型导致未能及时诊疗,从而预后不佳[4, 5]。因此,更准确的判断是否为UMI患者并定量梗死范围有助于改善预后及远期生存率[6, 7]

       在心脏磁共振(cardiac magnetic resonance, CMR)成像中,延迟钆增强(late gadolinium enhancement, LGE)可观察和量化心肌坏死以及微血管闭塞范围,在揭示UMI的组织学和功能学方面具有广泛的应用前景[8]。LGE强化面积定量可通过后处理软件半自动实现,但也依赖于图像的分辨率及阅片者的经验[9]。一项研究报告中,872名受检者中仅有23人(2.64%)的LGE提示存在UMI[10],该研究中UMI被定义为没有MI的症状、体征或病史,但在1.5 T CMR LGE上表现为MI区域(在LGE图像中显示为高信号病变,且具有特定冠状动脉的分布特征,涉及心肌的内膜下或透壁区域)。因此,准确及时地提示UMI及其强化灶范围,对患者进行危险分层、临床管理和治疗是亟待解决的临床问题[6, 7, 11]。在临床实践中,标准差(standard deviation, SD)和全宽半高(full width at half maximum, FWHM)法均被使用,但梗死面积的量化目前尚无统一结论[12, 13, 14]

       深度学习重建(deep learning reconstruction, DLR)技术可提高图像质量[15, 16, 17],降低噪声[18, 19, 20],还可能改变MI区域的诊断阈值,提供更准确的诊疗策略[21, 22, 23]。然而,目前尚未有基于DLR的LGE 研究对疑似UMI患者进行评估。因此,本研究旨在分析和比较常规重建的原始LGE和DLR的LGE序列(分别为 LGEO 和 LGEDL)对UMI的检出率,分析其在检测UMI方面的效果,并且使用不同的SD和FWHM方法评估两种图像的增强区域百分比(percentage of enhanced area, Parea),以确定最佳的SD法或FWHM法,这对于临床上对UMI患者的准确诊断具有重要的意义。

1 材料与方法

1.1 研究对象

       本前瞻性研究遵守《赫尔辛基宣言》,经武汉大学人民医院伦理委员会批准,批准文号:2022K-K083,全体受试者均签署了知情同意书。纳入标准:根据欧洲心脏病学会、美国心脏病学会、美国心脏协会和世界心脏联盟等共同起草的第四版“心肌梗死全球定义”的标准及既往文献报道[1, 10, 24],将无典型的心绞痛症状,有血清心脏肌钙蛋白(cardiac troponin,cTn)升高和/或回落,且至少1次高于正常值上限(参考值上限值的99百分位值),同时心电图提示有急性心肌缺血者诊断为UMI受试者。排除标准:(1)既往有肿瘤病史,既往心血管疾病手术史;(2)不愿意进行CMR检查,无法耐受或不能配合CMR检查、临床状况不稳定、失代偿性心力衰竭、有CMR检查禁忌证、预估肾小球滤过率≤30 mL/min和钆对比剂禁忌证;(3)存在左心室肥厚或左束支传导阻滞;(4)LGE图像不能用于临床的诊断和客观评估、CMR图像均无UMI表现。根据纳入与排除标准,纳入2022年4月至2023年8月于武汉大学人民医院就诊的可疑UMI患者98例,男68例,女30例,年龄(55.8±8.1)岁,因不愿行CMR检查排除5例,因不能耐受/不能配合CMR检查排除2例,因左心室肥厚或左束支传导阻滞排除14例,因CMR图像均无UMI表现排除16例,最终本研究纳入受试者61例。

1.2 CMR检查

       CMR检查采用3.0 T磁共振扫描仪(SIGNA Architect, GE Healthcare, Milwaukee, WI, USA)与30通道AIR线圈。患者仰卧位,头先进,加以心电门控及呼吸门控,范围覆盖左心室心尖至心底。本研究采用新型商用内置DLR技术(品牌名称:AIRTM Recon DL, DV29.1_R04, GE Healthcare, USA)通过识别以前的各类图像的440万特征,构建新网络,不使用偏置项,并采用线性整流单元激活函数,对影像数据即时处理,以降低噪声和吉布斯伪影,并进一步消除观察者之间和观察者内的差异[13]。LGE序列的参数如下:回波时间2.7 ms;重复时间5.6 ms;翻转角25°;视野34 cm×34 cm;矩阵260×174;层厚8 mm;层间距2 mm;接收带宽83.33 kHz;每段K空间线24;激励次数1;理论采集时间8 s×9个心跳。LGEO和LGEDL图像通过传统内置重建和AIR™ Recon DL算法同时生成。

       每名受试者于扫描前15 min采用高压造影注射器(Spectris Solaris EP,Bayer Medical Care Inc.)依每kg体质量0.1 mmol/kg(0.2 mL/kg)的钆贝葡胺注射液(莫迪司,上海博莱科信谊药业有限责任公司)注射,流率为3.5 mL/s,并以等流率注射等量生理盐水。

1.3 CMR影像评估

       首先由两名具有5年以上CMR诊断经验的放射科主治医师分别采用盲法手动勾画心肌内膜和外膜(图1),红圈表示心内膜边界,绿圈表示心外膜边界。随后,在心内膜和外膜之间的区域中,定义感兴趣区(region of interest, ROI),用于半定量分析梗死心肌所占左心室心肌的百分比。ROI的选择应尽可能覆盖病变区域,并避开邻近的血管和腔室,以确保测量的准确性和可靠性。在勾画出现不一致时,由另1位有25年心血管影像诊断经验的放射科主任医师独立审阅并进行最终评定。为避免误差,各参数连续测量3次并取其平均值。其中一位评估者在1个月后重新进行测量与计算,以验证结果的稳定性。

1.3.1 图像质量客观评价

       图像质量的客观定量指标包括两组序列图像中正常心肌信号强度(signal intensity, SI)(SIMyo-O和 SIMyo-DL)、心肌延迟强化区域(SIMDEA-O和SIMDEA-DL)以及背景处噪声标准差(standard deviation, SD)[25](SDBG-O,SDBG-DL)、心肌信号噪声比(signal-to-noise ratio, SNR)(SNRO,SNRDL)和对比噪声比(contrast-to-noise ratio, CNR)(CNRO,CNRDL)。计算公式见式(1)~(2[11]

1.3.2 心肌强化面积百分比评估

       于CVI42,v.5.17.1(3504)工作站(Circle Cardiovascular Imaging Inc., Calgary, AB, Canada)进行Parea半定量评估。分别使用高于参考心肌的平均SI阈值的2SD、3SD、4SD、5SD以及FWHM法对延迟强化灶进行量化,必要时由具备5年以上CMR诊断经验的放射科主治医师进行手动修正(图1)。记录每种技术所得S1~S16节段对应的Parea,并取其平均值为整体心肌Parea值。

1.3.3 诊断效能评估

       以临床诊断UMI为金标准计算LGEDL和LGEO的诊断效能(图2, 3, 4, 5)。

图2  男,54岁,UMI志愿者,后经DSA证实该患者左前降支近中段管腔重度狭窄,右冠状动脉近段管腔中度狭窄,其供血范围与图中所示梗死区域相符,最终诊断为UMI。LGEDL显示了更清晰、更少噪音、更均匀的正常心肌信号,并且增强区域与正常心肌之间的对比更好(2A)。使用2SD(2B)、3SD(2C)、4SD(2D)、5SD(2E)和FWHM方法(2F)定量Parea的示意图。UMI:未识别心肌梗死;DSA:数字减影血管造影;LGEDL:基于深度学习重建心肌延迟强化序列;SD:标准差;FWHM:全宽半高技术;Parea:心肌强化面积百分比。
Fig. 2  Male, 54 years old, UMI volunteer. DSA later confirmed severe stenosis in the mid-distal segment of the left anterior descending artery and moderate stenosis in the proximal segment of the right coronary artery, with the infarct area shown in the figure corresponding to the affected blood supply region. LGEDL shows clearer, less noisy, and more uniform normal myocardial signals with better contrast between the enhanced area and normal myocardium (2A). Schematic diagram of quantification of Parea using 2SD (2B), 3SD (2C), 4SD (2D), 5SD (2E), and FWHM methods (2F). UMI: unrecognized myocardial infarction; DSA: digital subtraction angiography; LGEDL: deep learning-based reconstruction late gadolinium enhancement; SD: standard deviation; FWHM: full width at half maximum; Parea: percentage of myocardial enhancement area.
图3  男,54岁,UMI志愿者,后经DSA证实该患者左前降支近中段管腔重度狭窄,右冠状动脉近段管腔中度狭窄,其供血范围与图中所示梗死区域相符,最终诊断为UMI。LGEO显示了心内膜下的增强区域(3A)。使用2SD(3B)、3SD(3C)、4SD(3D)、5SD(3E)和FWHM方法(3F)定量Parea的示意图。UMI:未识别心肌梗死;DSA:数字减影血管造影;LGEO:原始心肌延迟强化序列;SD:标准差;FWHM:全宽半高技术;Parea:心肌强化面积百分比。
Fig. 3  Male, 54 years old, UMI volunteer. DSA later confirmed severe stenosis in the mid-distal segment of the left anterior descending artery and moderate stenosis in the proximal segment of the right coronary artery, with the infarct area shown in the figure corresponding to the affected blood supply region. LGEO shows the enhancement area in the subendocardium (3A). Schematic diagram of quantification of Parea using 2SD (3B), 3SD (3C), 4SD (3D), 5SD (3E), and FWHM methods (3F). UMI: unrecognized myocardial infarction; DSA: digital subtraction angiography; LGEO: original late gadolinium enhancement; SD: standard deviation; FWHM: full width at half maximum; Parea: percentage of myocardial enhancement area.
图4  男,45岁,非UMI志愿者,最终诊断为肥厚型心肌病。LGEDL显示了更清晰、更少噪音、更均匀的正常心肌信号,并且增强区域与正常心肌之间的对比更好4A。使用2SD(4B)、3SD(4C)、4SD(4D)、5SD(4E)和FWHM方法(4F)定量Parea的示意图。UMI:未识别心肌梗死;LGEDL:基于深度学习重建心肌延迟强化序列;SD:标准差;FWHM:全宽半高技术;Parea:心肌强化面积百分比。
Fig. 4  Male, 45 years old, non-UMI volunteer, ultimately diagnosed with hypertrophic cardiomyopathy. LGEDL shows clearer, less noisy, and more uniform normal myocardial signals with better contrast between the enhanced area and normal myocardium (4A). Schematic diagram of quantification of Parea using 2SD (4B), 3SD (4C), 4SD (4D), 5SD (4E), and FWHM methods (4F). UMI: unrecognized myocardial infarction; LGEDL: deep learning-based reconstruction late gadolinium enhancement; SD: standard deviation; FWHM: full width at half maximum; Parea: percentage of myocardial enhancement area.
图5  男,45岁,非UMI志愿者,最终诊断为肥厚型心肌病。LGEO显示了室间隔插入部的增强区域(5A)。使用2SD(5B)、3SD(5C)、4SD(5D)、5SD(5E)和FWHM方法(5F)定量Parea的示意图。UMI:未识别心肌梗死;LGEO:原始心肌延迟强化序列;SD:标准差;FWHM:全宽半高技术;Parea:心肌强化面积百分比。
Fig. 5  Male, 45 years old, non-UMI volunteer, ultimately diagnosed with hypertrophic cardiomyopathy. LGEO shows enhancement in the insertion regions of the interventricular septum (5A). Schematic diagram of quantification of Parea using 2SD (5B), 3SD (5C), 4SD (5D), 5SD (5E), and FWHM methods (5F). UMI: unrecognized myocardial infarction; LGEO: original late gadolinium enhancement; SD: standard deviation; FWHM: full width at half maximum; Parea: percentage of myocardial enhancement area.
图1  左心室心肌强化面积百分比评估勾画的示意图。1A:标准差法;1B:全宽半高法。依次勾画左心室心内膜(红圈)和心外膜轮廓(绿圈),以及感兴趣区域(粉圈)。
Fig. 1  Schematic illustration of the delineation for evaluating the percentage of myocardial enhancement in the left ventricle. 1A: Standard deviation method; 1B: Full width at half maximum method. The red circle indicates the endocardial border, the green circle indicates the epicardial border, and the pink circle represents the region of interest.

1.4 统计学方法

       所有数据均使用R语言(version 4.0.4,http://www.r-project.org)进行统计分析。定量数据应用x¯±sMP25,P75)表示。所有定量数据符合正态分布和方差齐性的使用配对t检验,不符合则使用配对Wilcoxon检验。使用组内相关系数(intra-class correlation coefficient, ICC)评估客观定量指标观察者内和观察者间的一致性,ICC值在0.200~0.400之间说明一致性程度一般;>0.400~0.600之间说明一致性程度中等;>0.600~0.800之间说明一致性程度较强;>0.800~1.000之间说明一致性程度很强。纳入各阈值法评估Parea-DL和Parea-O绘制受试者工作特征(receiver operator characteristic, ROC)曲线,通过曲线下面积(area under the curve, AUC)比较Parea-DL和Parea-O对UMI诊断效能的差异。P<0.05为差异具有统计学意义。

2 结果

2.1 临床特征

       本研究最终纳入61例无典型的心绞痛症状,血清cTn及心电图同时提示为UMI的患者,男43例,女18例,年龄(55.9±8.7)岁。

2.2 图像质量客观评价

       LGEDL序列16个节段SDDL、整体心肌SDMyo-DL、心肌强化灶SDMDEA-DL、背景噪声SDBG-DL均较SDO、SDMyo-O、SDEA-O、SDBG-O显著减低(P<0.001)。LGEDL上的SNR值在16个节段上均较LGEO升高,差异有统计学意义(P<0.001),以SNRS16段为著(92.44±78.39 vs. 27.39±24.56,P<0.001)(表1)。

       LGEDL序列心肌平均SNRDL较SNRO升高(99.93±81.42 vs. 33.29±30.89,P<0.001);CNRDL平均值(123.72±45.00)较CNRO平均值(60.15±15.52)明显升高,差异有统计学意义(P<0.001)。除S7~S9和S11段外,其他节段的SIDL均高于SIO(P<0.001)。S1~S6、S10、S12、S16和整个心肌节段的SIMyo-DL均高于相应节段的SIMyo-O(P<0.001)。S7~S9和S11节段的SIMyo-DL略高于SIMyo-O,但差异无统计学意义(P>0.05)(图6A)。

图6  心肌信号强度及心肌强化面积百分比评估。LGEDL和LGEO序列左心室心肌信号强度评估(6A);运用2SD(6B)、3SD(6C)、4SD(6D)、5SD(6E)和FWHM法(6F)评估LGEDL和LGEO序列左心室心肌强化面积百分比。*代表P<0.05,**代表P<0.001,ns代表P>0.05。LGEDL:基于深度学习重建心肌延迟强化序列;LGEO:原始心肌延迟强化序列;SI:信号强度;Parea:心肌强化面积百分比;MDEA:心肌延迟强化区域; SD:标准差;FWHM:全宽半高技术。
Fig. 6  Assessment of myocardial signal intensity and percentage of myocardial enhancement area. Evaluation of left ventricular myocardial signal intensity using LGEDL and LGEO sequences(6A); Assessment of the percentage of left ventricular myocardial enhancement area using 2SD (6B), 3SD (6C), 4SD (6D), 5SD (6E), and FWHM methods (6F) for LGEDL and LGEO sequences. * represents P<0.05, ** represents P<0.001, ns represents P>0.05. LGEDL: deep learning-based reconstruction late gadolinium enhancement; LGEO: original late gadolinium enhancement; SI: signal intensity; Parea: percentage of myocardial enhancement area; MDEA: myocardial delayed enhancement area; SD: standard deviation; FWHM: full width at half maximum technique.
表1  LGEDL 和LGEO 图像质量客观评价
Tab. 1  Objective assessment of image quality for LGEDL and LGEO

2.3 心肌强化面积百分比评估

       使用标准差阈值法及半高宽度法对心肌强化面积进行半定量分析。结果表明,2SD、3SD、5SD法,S1~S16节段及其平均Parea-DL均高于Parea-O,差异有统计学意义(P均<0.05);4SD法,S1~S12节段及整体心肌平均Parea-DL均高于Parea-O,差异有统计学意义(P均<0.05);4SD法,S13~S16节段心肌平均Parea-DL略高于Parea-O,差异无统计学意义(P>0.05);FWHM法示S1~S16节段及其平均Parea-DL略高于Parea-O,差异无统计学意义(P>0.05)(图6B~6F)。

2.4 定量指标一致性评估

       LGEDL序列、LGEO 序列图像质量评价各平均值(SDMyo、SDMDEA、SDBG、SNR、CNR、SIMyo)、2SD~5SD及FWHM法Parea定量观察者内和观察者间有较好的一致性(ICC均>0.600,P均<0.001),且LGEDL序列观察者内和观察者间一致性均优于LGEO 序列(表2)。

表2  LGEDL和LGEO序列各测量值组内和组间一致性评估
Tab. 2  Intra- and inter-group consistency assessment of measurements for LGEDL and LGEO

2.5 诊断效能分析

       通过绘制ROC曲线图,比较AUC,提示:(1)各SD法展现出对UMI良好的诊断效能,AUC值至少大于0.781,LGEDL序列最优诊断效能为5SD Parea-DL法(敏感度=68.8%,特异度=100.0%);LGEO序列最优诊断效能为3SD Parea-O法,AUC值为0.840;(2)基于深度学习的LGEDL法对UMI的诊断效能优于LGEO法;而FWHMDL法与FWHMO法对UMI的诊断效能相似(图7)。

图7  对UMI的诊断效能。ROC:受试者工作特征;Parea:心肌强化面积百分比;SD:标准差;FWHM:全宽半高技术;DL:深度学习;O:原始序列;Parea:心肌强化百分比;AUC:曲线下面积。
Fig. 7  Diagnostic accuracy for UMI. ROC: receiver operating characteristic; Parea: percentage of myocardial enhancement area; SD: standard deviation; FWHM: full width at half maximum; DL: deep learning; O:original; Parea: percentage of myocardial enhancement area; AUC: area under the curve.

3 讨论

       本研究使用不同的SD阈值法和FWHM法分析传统与DLR的LGE序列。除FWHM法,使用阈值法 的Parea差异均有统计学意义,这表明Parea-O的信号阈值参考平均值(signal threshold versus reference mean, STRM)≥3,而Parea-DL则≥4,用5SD阈值的Parea-DL对UMI患者的诊断效能最优。此外,本研究首次应用DLR LGE显示UMI患者的延迟增强病灶,在所有UMI患者中LGEDL显示出明显优于LGEO的图像质量(如噪声更少、SD更低、SNR和CNR更高),从而提高了诊断可信度,且无损诊断效能。

3.1 图像质量

       LGE作为一种无创、无辐射且图像分辨率高的检查方法,不仅可以准确地诊断典型的MI灶,还可以发现临床症状不典型的MI灶[26],本研究入组的研究对象即为无典型的心绞痛症状的UMI受试者。研究表明,UMI不仅是冠状动脉疾病进展中的一个严重且易被忽略的事件[27, 28],还与未来不良心脏事件的高风险密切相关。梗死灶的存在是预测这些不良事件的主要特征[29, 30, 31],因此早期识别和管理UMI对于改善患者预后至关重要。心肌保护治疗的目标是减少不可逆损伤,而Parea被视为最直接反映心肌保护效果的预后指标,与病理水平上的不可逆损伤密切相关[32, 33, 34]。然而,图像质量可能影响LGEDL图像上远端正常心肌的准确勾画[35]。常规LGE通过多次图像平均和运动校正来提高图像质量,减少呼吸伪影,从而提升了LGE在识别MI中的敏感性和特异性[36]。本研究中,LGEDL图像在所有UMI患者中显示出更好的图像质量,包括较低的背景噪声、较低的SD以及更高的SNR和CNR,与既往研究类似[25]。这些改进不仅提高了图像的清晰度和对比度,还改善了Parea测量的观察者间和观察者内一致性。这表明,LGEDL图像能更精确地描绘心内膜、心外膜和病灶边界,从而提高了诊断的可靠性。

3.2 心肌延迟强化

       基于SD阈值的量化主要取决于远端正常心肌的SI和SD,而图像质量优化的重建技术可能影响LGEDL延迟强化灶的可视化[13]。Parea的增加在不同心脏分段之间存在差异,例如,相较于LGEO,LGEDL上S12段的改善最为显著,主要是该节段为中间段侧壁,本身伪影干扰少,病灶分界清楚。常规LGE中,阈值的选择在特定情况下表现出最佳的一致性和诊断性能[36]。通常,STRM≥3SD被认为是最佳参考阈值,使用2SD阈值可能会高估增强区域,在本研究中,4SD Parea-DL和3SD Parea-O的观察者间和观察者内一致性方面表现最佳。因此,在选择参考阈值时,需要谨慎考虑。尽管FWHM方法相对于视觉评估可能低估LGE质量,但其对噪声水平最不敏感,与先前文献的结果一致[12, 37],由于LGEDL序列只去除了噪声而没有改变病灶信息,故FWHM法的Parea-DL与Parea-O在统计学上没有差异[29, 38, 39]。这些发现与以往研究的结果相符,进一步证实了LGEDL序列在提高图像质量和诊断准确性方面的优势。

3.3 诊断效能

       本研究评估和直接比较LGEDL和LGEO图像中UMI患者延迟强化灶的结果显示,阈值方法中,Parea-DL的诊断性能高于Parea-O,尤其是基于5SD阈值的Parea-DL表现出最佳的AUC(0.891)。对于LGEO图像,基于3SD阈值的Parea-O展示了最佳的AUC(0.840),这一结果与以往研究建议的使用STRM≥3SD评估梗死面积的数据显示一致。本研究招募了无明显心源性胸痛且强化灶范围相对较小的UMI患者,结果证实3SD阈值对于常规LGE图像是足够的。相比之下,对于DLR LGE图像应使用≥4SD的阈值,以优化观察者间和观察者内的一致性及诊断效能。尽管5SD阈值在LGEDL成像中的诊断效能更好,但使用4SD阈值诊断UMI相关病例的梗死范围可能是LGEO和LGEDL图像中更可靠的参数。此外,UMI的检测率为67%;尽管LGEDL成像的图像质量更好且病理特征评估更可靠,但LGEO和LGEDL图像的检测率相似。

3.4 局限性

       本研究尚存在一些局限性:首先,样本量较少,且志愿者均来自单中心,需多中心研究进一步验证结果的稳定性;其次,由于UMI患者自身缺乏典型症状,部分受试者未能/不愿意进行进一步的检查,可能导致研究结果存在一定的偏倚;再次,以往研究证实LGE在MI的诊断具有很高的准确性,但MI的判定仍依赖于阅片者的主观经验,缺乏病理学上的金标准;最后,由于本研究的随访时间较短且样本量有限,我们计划在未来的研究中进行更长期的随访,以全面评估UMI患者的长期预后和相关不良心脏事件的发生率。通过扩展的随访研究,我们希望能够更充分地验证LGEDL在UMI诊断和风险评估中的临床价值,从而为临床实践提供更可靠的依据。

4 结论

       总之,基于DLR技术心肌延迟强化序列能显著优化图像质量,提高UMI诊断效能,为UMI诊治提供可靠的影像学证据,能够在未来临床诊疗中获得较大的收益。不同节段图像质量优化及不同阈值法半定量值差异显著,FWHM技术独立于DL图像质量优化技术;SD法对异常心肌的敏感性不同,在定量分析时应予以关注。

[1]
THYGESEN K, ALPERT J S, JAFFE A S, et al. Fourth universal definition of myocardial infarction (2018)[J/OL]. Circulation, 2018, 138(20): e618-e651 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/30571511/. DOI: 10.1161/CIR.0000000000000617.
[2]
CHENG Y J, JIA Y H, YAO F J, et al. Association between silent myocardial infarction and long-term risk of sudden cardiac death[J/OL]. J Am Heart Assoc, 2021, 10(1): e017044 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/33372536/. DOI: 10.1161/JAHA.120.017044.
[3]
VÄHÄTALO J H, HUIKURI H V, HOLMSTRÖM L T A, et al. Association of silent myocardial infarction and sudden cardiac death[J]. JAMA Cardiol, 2019, 4(8): 796-802. DOI: 10.1001/jamacardio.2019.2210.
[4]
WEIR-MCCALL J R, FITZGERALD K, PAPAGIORCOPULO C J, et al. Prevalence of unrecognized myocardial infarction in a low-intermediate risk asymptomatic cohort and its relation to systemic atherosclerosis[J]. Eur Heart J Cardiovasc Imaging, 2017, 18(6): 657-662. DOI: 10.1093/ehjci/jew155.
[5]
SUGIYAMA T, KANAJI Y, HOSHINO M, et al. Relationship between unrecognized myocardial infarction and underlying coronary plaque characteristics on optical coherence tomography[J]. JACC Cardiovasc Imaging, 2022, 15(10): 1830-1832. DOI: 10.1016/j.jcmg.2022.05.021.
[6]
DASTIDAR A G, BARITUSSIO A, GARATE E D, et al. Prognostic role of CMR andConventional risk factors inMyocardial infarction with nonobstructed coronary arteries[J]. JACC Cardiovasc Imaging, 2019, 12(10): 1973-1982. DOI: 10.1016/j.jcmg.2018.12.023.
[7]
PESAPANE F, CODARI M, SARDANELLI F. Artificial intelligence in medical imaging: threat or opportunity? Radiologists again at the forefront of innovation in medicine[J/OL]. Eur Radiol Exp, 2018, 2(1): 35 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/30353365/. DOI: 10.1186/s41747-018-0061-6.
[8]
REINDL M, EITEL I, REINSTADLER S J. Role of cardiac magnetic resonance to improve risk prediction following acute ST-elevation myocardial infarction[J/OL]. J Clin Med, 2020, 9(4): 1041 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/32272692/. DOI: 10.3390/jcm9041041.
[9]
KIM C, PARK C H, KIM D Y, et al. Semi-quantitative scoring of late gadolinium enhancement of the left ventricle in patients with ischemic cardiomyopathy: improving interobserver reliability and agreement using consensus guidance from the Asian society of cardiovascular imaging-practical tutorial (ASCI-PT) 2020[J]. Korean J Radiol, 2022, 23(3): 298-307. DOI: 10.3348/kjr.2021.0387.
[10]
CHA M J, KIM S M, KIM Y, et al. Unrecognized myocardial infarction detected on cardiac magnetic resonance imaging: association with coronary artery calcium score and cardiovascular risk prediction scores in asymptomatic Asian cohort[J/OL]. PLoS One, 2018, 13(9): e0204040 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/30216389/. DOI: 10.1371/journal.pone.0204040.
[11]
DEMIRKIRAN A, EVERAARS H, AMIER R P, et al. Cardiovascular magnetic resonance techniques for tissue characterization after acute myocardial injury[J]. Eur Heart J Cardiovasc Imaging, 2019, 20(7): 723-734. DOI: 10.1093/ehjci/jez094.
[12]
PRADELLA S, MAZZONI L N, LETTERIELLO M, et al. FLORA software: semi-automatic LGE-CMR analysis tool for cardiac lesions identification and characterization[J]. Radiol Med, 2022, 127(6): 589-601. DOI: 10.1007/s11547-022-01491-8.
[13]
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.
[14]
FLETT A S, HASLETON J, COOK C, et al. Evaluation of techniques for the quantification of myocardial scar of differing etiology using cardiac magnetic resonance[J]. JACC Cardiovasc Imaging, 2011, 4(2): 150-156. DOI: 10.1016/j.jcmg.2010.11.015.
[15]
ZUCKER E J, SANDINO C M, KINO A, et al. Free-breathing accelerated cardiac MRI using deep learning: validation in children and young adults[J]. Radiology, 2021, 300(3): 539-548. DOI: 10.1148/radiol.2021202624.
[16]
BETANCUR J, COMMANDEUR F, MOTLAGH M, et al. Deep learning for prediction of obstructive disease from fast myocardial perfusion SPECT: a multicenter study[J]. JACC Cardiovasc Imaging, 2018, 11(11): 1654-1663. DOI: 10.1016/j.jcmg.2018.01.020.
[17]
OSCANOA J A, MIDDIONE M J, ALKAN C, et al. Deep learning-based reconstruction for cardiac MRI: a review[J/OL]. Bioengineering, 2023, 10(3): 334 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/36978725/. DOI: 10.3390/bioengineering10030334.
[18]
XUE H, REHMAN A, DAVIES R H, et al. CNNT DB-LGE: free-breathing dark blood late enhancement imaging using the convolutional neural network transformer speeds acquisition by 50%[J/OL]. Eur Heart J Cardiovasc Imaging, 2022, 23(Supplement_2): jeac141.006 [2024-02-02]. https://academic.oup.com/ehjcimaging/article/23/Supplement_2/jeac141.006/6673964. DOI: 10.1093/ehjci/jeac141.006.
[19]
EL-REWAIDY H, NEISIUS U, MANCIO J, et al. Deep complex convolutional network for fast reconstruction of 3D late gadolinium enhancement cardiac MRI[J/OL]. NMR Biomed, 2020, 33(7): e4312 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/32352197/. DOI: 10.1002/nbm.4312.
[20]
FAHMY A S, ROWIN E J, CHAN R H, et al. Improved quantification of myocardium scar in late gadolinium enhancement images: deep learning based image fusion approach[J]. J Magn Reson Imaging, 2021, 54(1): 303-312. DOI: 10.1002/jmri.27555.
[21]
LIU T, TIAN Y, ZHAO S F, et al. Residual convolutional neural network for cardiac image segmentation and heart disease diagnosis[J/OL]. IEEE Access, 2020, 8: 82153-82161 [2024-02-02]. https://ieeexplore.ieee.org/document/9082648. DOI: 10.1109/ACCESS.2020.2991424.
[22]
GHANBARI F, JOYCE T, LORENZONI V, et al. AI cardiac MRI scar analysis aids prediction of major arrhythmic events in the multicenter DERIVATE registry[J/OL]. Radiology, 2023, 307(3): e222239 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/36943075/. DOI: 10.1148/radiol.222239.
[23]
YOON S, NAKAMORI S, AMYAR A, et al. Accelerated cardiac MRI cine with use of resolution enhancement generative adversarial inline neural network[J/OL]. Radiology, 2023, 307(5): e222878 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/37249435/. DOI: 10.1148/radiol.222878.
[24]
BARBIER C E, THEMUDO R, BJERNER T, et al. Long-term prognosis of unrecognized myocardial infarction detected with cardiovascular magnetic resonance in an elderly population[J/OL]. J Cardiovasc Magn Reson, 2016, 18(1): 43 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/27430315/. DOI: 10.1186/s12968-016-0264-z.
[25]
CHEN Q, FANG S, YUCHEN Y, et al. Clinical feasibility of deep learning reconstruction in liver diffusion-weighted imaging: improvement of image quality and impact on apparent diffusion coefficient value[J/OL]. Eur J Radiol, 2023, 168: 111149 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/37862927/. DOI: 10.1016/j.ejrad.2023.111149.
[26]
刘茜, 杨志刚, 李媛. 心肌梗死的心脏磁共振成像技术临床应用及其研究现状[J]. 磁共振成像, 2021, 12(8): 98-100, 107. DOI: 10.12015/issn.1674-8034.2021.08.022.
LIU Q, YANG Z G, LI Y. Clinical application and current research status of cardiac magnetic resonance imaging of myocardial infarction[J]. Chin J Magn Reson Imag, 2021, 12(8): 98-100, 107. DOI: 10.12015/issn.1674-8034.2021.08.022.
[27]
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/OL]. Int J Cardiol, 2021, 343: 164-170 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/34517017/. DOI: 10.1016/j.ijcard.2021.09.012.
[28]
DAVIS T M E, FORTUN P, MULDER J, et al. Silent myocardial infarction and its prognosis in a community-based cohort of Type 2 diabetic patients: the Fremantle Diabetes Study[J]. Diabetologia, 2004, 47(3): 395-399. DOI: 10.1007/s00125-004-1344-4.
[29]
GANESAN A N, GUNTON J, NUCIFORA G, et al. Impact of Late Gadolinium Enhancement on mortality, sudden death and major adverse cardiovascular events in ischemic and nonischemic cardiomyopathy: a systematic review and meta-analysis[J/OL]. Int J Cardiol, 2018, 254: 230-237 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/29407096/. DOI: 10.1016/j.ijcard.2017.10.094.
[30]
HALLIDAY B P, BAKSI A J, GULATI A, et al. Outcome in dilated cardiomyopathy related to the extent, location, andPattern of late gadolinium enhancement[J]. JACC Cardiovasc Imaging, 2019, 12(8Pt 2): 1645-1655. DOI: 10.1016/j.jcmg.2018.07.015.
[31]
WENG Z, YAO J L, CHAN R H, et al. Prognostic value of LGE-CMR in HCM: a meta-analysis[J]. JACC Cardiovasc Imaging, 2016, 9(12): 1392-1402. DOI: 10.1016/j.jcmg.2016.02.031.
[32]
STONE G W, SELKER H P, THIELE H, et al. Relationship between infarct size and outcomes following primary PCI: patient-level analysis from 10 randomized trials[J]. J Am Coll Cardiol, 2016, 67(14): 1674-1683. DOI: 10.1016/j.jacc.2016.01.069.
[33]
TAYLOR A M. The role of artificial intelligence in paediatric cardiovascular magnetic resonance imaging[J]. Pediatr Radiol, 2022, 52(11): 2131-2138. DOI: 10.1007/s00247-021-05218-1.
[34]
VAN DER ENDE M Y, JUAREZ-OROZCO L E, WAARDENBURG I, et al. Sex-based differences in unrecognized myocardial infarction[J/OL]. J Am Heart Assoc, 2020, 9(13): e015519 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/32573316/. DOI: 10.1161/JAHA.119.015519.
[35]
BUSTIN A, JANICH M A, BRAU A C, et al. Joint denoising and motion correction: initial application in single-shot cardiac MRI[J/OL]. J Cardiovasc Magn Reson, 2015, 17: Q29 [2024-02-02]. https://www.sciencedirect.com/science/article/pii/S1097664723020409?via%3Dihub. DOI: 10.1186/1532-429x-17-s1-q29.
[36]
VAN DEN HOOGEN I J, WANG X, BUTCHER S C, et al. Incorporating coronary artery calcium scoring in the prediction of obstructive coronary artery disease with myocardial ischemia: a study with sequential use of coronary computed tomography angiography and positron emission tomography imaging[J]. J Nucl Cardiol, 2023, 30(1): 178-188. DOI: 10.1007/s12350-022-03132-z.
[37]
JENISTA E R, WENDELL D C, AZEVEDO C F, et al. Revisiting how we perform late gadolinium enhancement CMR: insights gleaned over 25 years of clinical practice[J/OL]. J Cardiovasc Magn Reson, 2023, 25(1): 18 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/36922844/. DOI: 10.1186/s12968-023-00925-0.
[38]
MIKAMI Y, KOLMAN L, JONCAS S X, et al. Accuracy and reproducibility of semi-automated late gadolinium enhancement quantification techniques in patients with hypertrophic cardiomyopathy[J/OL]. J Cardiovasc Magn Reson, 2014, 16(1): 85 [2024-02-02]. https://pubmed.ncbi.nlm.nih.gov/25315701/. DOI: 10.1186/s12968-014-0085-x.
[39]
ANTIOCHOS P, GE Y, STEEL K, et al. Imaging of clinically unrecognized myocardial fibrosis in patients with suspected coronary artery disease[J]. J Am Coll Cardiol, 2020, 76(8): 945-957. DOI: 10.1016/j.jacc.2020.06.063.

上一篇 基于深度学习的图像重建在提高颞下颌关节MRI图像质量中的初步应用研究
下一篇 基于深度学习重建的扩散加权成像在肺部病变良恶性鉴别中的应用价值
  
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