by HPC_ZY 期刊文献《The 39 steps: evading error and deciphering the secrets for accurate dynamic susceptibility contrast MRI》阅读笔记 INTRODUCTION 这39个步骤遵循采集、分析的工作流程以及DSC-MRI数据的解释 如下所示。 The 39 steps follow the workflow for the acquisition, analysisand interpretation of DSC-MRI data, as follows.• The contrast agent (steps 1–4). # 造影剂• The acquisition of DSC-MRI data (steps 5–12). # 获取DSC-MRI数据• Data pre-processing (steps 13–15). # 数据预处理• The contrast concentration–time course (steps 16–19). # 对比剂浓度-时间过程• The arterial input function (steps 20–22). # 动脉输入功能• Data deconvolution (steps 23–27). # 数据反褶积• Common perfusion parameters (steps 28–31). # 常见灌注参数• Post-processing possibilities (steps 32 and 33). # 后处理可能性• Considerations for patient studies (steps 34–36). # 患者病例的注意事项• Absolute versus relative quantification (steps 37 and 38). # 绝对定量与相对定量• Automated analysis (step 39). # 自动分析The steps explain the associated principles, practicalities, caveats, pitfalls and possibilities. 接下来逐一说明 THE CONTRAST AGENT Quantification of perfusion using DSC-MRI is based on the principles of tracer kinetic modelling (1), whereby a contrast agent is injected into the blood and monitored as it passes through the microvasculature. 使用DSC-MRI定量灌注是基于示踪剂动力学模型的原理 即将造影剂注入血液并在其通过微血管时进行监控。 1. The contrast material 我们关注的是数据处理技术 跳过临床操作 2. The contrast dose 我们关注的是数据处理技术 跳过临床操作 3. The contrast injection 我们关注的是数据处理技术 跳过临床操作 4. The contrast risks 我们关注的是数据处理技术 跳过临床操作 DATA PRE-PROCESSING Before proceeding with the analysis of DSC-MRI data, it is essential to check the quality of the datasets. For example, sometimes, in a clinical setting, the data will be completely corrupted by patient motion or a failed contrast injection, thereby rendering any subsequent analysis meaningless. For DSC-MRI datasets that pass this initial quality control, the data may be enhanced by the careful application of pre-processing techniques. 在进行DSC-MRI数据分析之前 必须检查数据集的质量。例如 在临床环境中 数据有时会因为患者的运动或失败的对比剂注射完全破坏 从而使任何后续分析变得毫无意义。对于通过初始质量控制的DSC-MRI数据集 可通过仔细应用预处理技术来增强数据。 13. Subject motion When scanning seriously ill patients or very uncooperative subjects, subject motion during the passage of the contrast agent through the brain can result in significant distortion in the MR signal profile. Furthermore, as DSC-MRI is typically a two-dimensional multi-slice acquisition (with inter-slice gaps), through-slice motion cannot be dealt with. 当扫描重病患者或非常不合作的受试者时 对比剂通过大脑时受试者的运动会导致MR信号的严重失真。此外 由于DSC-MRI通常是二维多层面采集 具有层间间隙 因此无法处理层间运动。 14. Acquisition noise The amount of noise in the DSC-MRI data affects the accuracy to which the perfusion parameters can be determined (see step 25). In particular, accurate perfusion estimates in hypoperfused areas (where contrast is low) are difficult. Pre-processing methods to improve SNR can be helpful. One means to improve SNR is to implement spatial smoothing, in which each time frame is filtered independently of the other time frames DSC-MRI数据中的噪声量影响可以确定灌注参数的精度 参见步骤25 。尤其是在低灌注区域 对比度低 的准确灌注估计是困难的。预处理方法可以提高信噪比。提高信噪比的一种方法是实现空间平滑 spatial independent component analysis (30,31)空间独立分量分析wavelet denoising (32)小波去噪partial differential equations (33)偏微分方程four-dimensional nonlinear filtering (34)四维非线性滤波Bayesian random effects models (37)贝叶斯随机效应模型15. Macrovessel artefacts In GE images, where large vessels are pronounced, PVEs may result in an overestimation of perfusion in nearby tissue voxels. The use of independent component analysis to minimise macrovessel signals was first suggested by Carroll et al. (38), and has been combined recently with principal component analysis (39), and made fully automated (40). This automated methodology was found to minimise macrovessel artefacts in cortical grey matter perfusion, whilst leaving white matter perfusion parameters largely unaffected. Minimisation of the macrovessel signal in GE images may assist in the identification of perfusion abnormalities, as well as in helping to quantify true tissue perfusion. 在大血管显影的GE图像中 pve可能导致对邻近组织体素灌注的高估。使用独立成分分析结合主成分分析来最小化大血管信号(38-40)。这种自动化的方法可以最大限度地减少大脑皮质灰质灌注中的大血管伪影 同时使白质灌注参数基本不受影响。最小化GE图像中的大血管信号有助于识别灌注异常 也有助于量化真实的组织灌注。 THE CONCENTRATION–TIME COURSE The first stage in the analysis of DSC-MRI data is to convert the MR signal–time course (in each voxel) into a curve that is proportional to the concentration of contrast agent in the tissue – the concentration–time course (CTC). 分析DSC-MRI数据的第一步是将MR信号-时间过程 在每个体素中 转换成与组织中对比剂浓度成比例的曲线-浓度-时间过程 CTC16. measurement of the baseline MR signal; In order to convert the MR signal into contrast concentration passing through the capillary bed (see steps 17 and 18), it is usually necessary to obtain an accurate estimate of the baseline MR signal intensity S(0) prior to the arrival of contrast agent. To maximise SNR, the pre-contrast baseline signal should be calculated as the average signal over several time points after the establishment of a steady state. 为了将MR信号转换为通过毛细管床的对比剂浓度 参见步骤17和18 通常需要在造影剂到达之前获得基线MR信号强度S 0 的准确估计。为了使信噪比最大化 对比前基线信号应计算为稳态建立后几个时间点的平均信号。 These time points must also avoid periods of through-slice movement, which will disturb the steady state. 这句原理没有理解到 17. quantification of the susceptibility (T2*) contrast; 这里描述了MR信号与对比剂浓度的关系 由于我使用的CTP数据 所以不需要考虑这一点。 18. computation of the tissue CTC; 直接跳过这里 19. computation of the arterial CTC 直接跳过这里 THE ARTERIAL INPUT FUNCTION As the shape of the tissue CTC depends on both the tissue properties and how the contrast enters the tissue, a proper quantification of perfusion must account for the arterial input. 由于组织CTC的形状取决于组织的性质和对比剂如何进入组织 因此灌注的量化必须考虑到动脉的输入。 20. Estimation of a global arterial input function (AIF) Measurement of the arterial CTC, commonly termed the ‘arterial input function’ (AIF). a global AIF methodology is used, whereby a single AIF is used to analyse the whole imaging slice or brain. 测量动脉CTC 通常称为\"动脉输入功能” AIF 。使用了一种全局AIF方法 其中单个AIF用于分析整个成像切片或大脑。 It is important to recognise that between the global AIF measurement site (e.g. a major branch of the middle cerebral artery, MCA) and the actual tissue input, the contrast bolus may become dispersed in time. Consequently, a global AIF is likely to be narrower than the true tissue input, resulting in an underestimation of perfusion (25,51). Local and regional AIF methodologies (see step 35) have been suggested to minimise these errors, but are currently not in widespread use. 重要的是要认识到 在整体AIF测量部位 例如大脑中动脉的一个主要分支 MCA 和实际组织输入之间 造影剂团可能会在时间上分散。因此 整体AIF可能比真正的组织输入窄 导致对灌注的低估。 有人建议采用本地和区域AIF方法 以尽量减少这些误差 但目前尚未广泛使用。 There are many difficulties in achieving an accurate characterisation of AIF, including errors from dispersion, nonlinearity, T1 effects, PVEs, saturation and geometric distortions; artefacts such as these can lead to significant perfusion errors (22,46,51–55). 高精度AIF的获取有很大难度 包括色散、非线性、T1效应、pve、饱和和几何畸变的误差 伪影可能导致严重的灌注误差。 21. Estimation of AIF using the MR signal magnitude Commonly, AIF measurements are made in the contralateral MCA or even more distal locations in the arterial tree (56), which are closer to the tissue and therefore limit the influence of delay and dispersion errors (see steps 34 and 35). However, as the spatial resolution of DSC-MRI is generally larger than the diameter of MCA, AIF voxels are also likely to contain signal contribution from the tissue. These PVEs can cause severe disruption to the AIF profile and lead to a large underestimation or large overestimation of perfusion (52,55). 通常 AIF测量是在对侧MCA或动脉树远端的位置进行的 这些位置更接近组织 能有效减少延迟和离散误差的影响。然而 由于DSC-MRI的空间分辨率通常大于MCA的直径 AIF体素也可能包含来自组织的信号。这些PVE会严重破坏AIF曲线 导致对灌注的严重低估或高估。 不过我使用的CTP数据 分辨率为0.5mm 大于动脉直径 后面讲的都是关于核磁各种序列对于AIF选取的问题 不再深入去看。因为我使用的CT。 22. Estimation of AIF using the MR signal phase 除了基于幅度的AIF选取 还有基于相位的 它们都针对AIF的形状而非大小。 DATA DECONVOLUTION In DSC-MRI, perfusion and other related haemodynamic parameters are estimated using the mathematical technique of deconvolution. 在DSC-MRI中 灌注和其他相关的血流动力学参数是用去卷积的数学技术来估计的。 23. The convolution equation 卷积方程的确定 在很多论文都讲过了。 基本一致 24. Model-independent deconvolution 反卷积求解 在很多论文都讲过了。 基本一致 25. Regularisation techniques Since removing the high-frequency components of CBFR(t) inevitably results in an underestimation of CBF, the most accurate CBF estimates are obtained when the input data have high SNR and therefore only a small amount of regularisation is needed to stabilise the solution. Conversely, CBF will be underestimated most severely for noisy data, where stronger regularisation is necessary to stabilise the solution. 由于去除CBFR t 的高频分量不可避免地会导致CBF的低估 因此当输入数据具有高信噪比时 可以获得最准确的CBF估计 因此只需少量正则化即可稳定解。相反地 对于噪声数据 CBF将被严重低估 因为需要更强大的正则化来稳定解。 The optimum amount of regularisation depends on the SNR and perfusion parameters of the input data (72,79). If there is too little regularisation, CBFR(t) will be unstable and nonphysiological; if there is too much regularisation, CBFR(t) will be oversmoothed and CBF will be severely underestimated. 最佳正则化量取决于输入数据的信噪比和灌注参数 72,79 。如果正则化太少 CBFR t 将是不稳定和非生理的 如果正则化太多 CBFR t 将过度平滑 CBF将被严重低估。 The best regularisation approaches are therefore adaptive to the noise level of individual CTCs. 因此 最好的正则化方法可以适应单个CTC的噪声水平。 From a Fourier domain perspective, it is clear that the sharp truncation of singular values in SVD can introduce additional oscillations in CBFR(t). 从Fourier域的角度来看 奇异值的急剧截断会导致CBFR t 的附加振荡。 Several studies have demonstrated the strong dependence of the estimated CBF on the choice of deconvolution, regularisation method and degree of regularisation, e.g. refs. (29,80,81). 一些研究表明 估计的CBF与反褶积、正则化方法和正则化程度的选择有很强的依赖性 如参考文献。 26. Discretisation methods As a result of the relatively coarse sampling rate (typically TR 1.5–2.5 s) compared with the width of the first passage of the bolus (~10 s), discrete deconvolution of DSC-MRI data (even in the absence of noise) results in a distorted CBF R(t). 由于相对较粗的采样率 通常TR 1.5–2.5 s 与第一次通过的bolus ~10 s 相比 离散的DSC-MRI数据反褶积 即使在没有噪声的情况下 会导致结果失真Discretisation errors can be minimised (although not eliminated) by formulating the AIF matrix to assume values of AIF(t) and R(t) between measurement points, but without actual subsampling. 离散化误差可以通过建立AIF矩阵来假设测量点之间的AIF t 和R t 的值而最小化 尽管没有消除 但不需要实际的子采样。 Assuming a linear variation over time, the elements of the AIF matrix become (69): Alternative matrix approximations have been proposed (29,80),Volterra method for discretisation . (82) 假设随时间呈线性变化 AIF矩阵的元素变为 替代矩阵近似 Volterra离散化方法。 27. Bolus recirculation effects Although, in principle, the recirculation signal in the tissue and arterial compartments does not affect the analysis, the convolution integrals are only strictly valid for infinite sampling durations. Numerical simulations have shown that finite sampling durations progressively underestimate CBV and overestimate CBF (24), especially for late bolus arrival times and prolonged MTT. One way to deal with the truncation of recirculation is to use modelbased extrapolation (83). 尽管原则上 组织和动脉室中的再循环信号不影响分析 但卷积积分仅对无限采样持续时间严格有效。数值模拟表明 有限采样持续时间逐渐低估CBV并高估CBF 特别是对于延迟到达时间和延长MTT。处理再循环截断的一种方法是使用基于模型的外推法。 More promising approaches preserve the characteristics of the first pass by separating it from the recirculation using, for example, iterative deconvolution algorithms that intrinsically model the recirculation (85), spatial-independent component analysis to identify the recirculation components (86) or data-driven approaches based on a knowledge of the injection timing and vascular transport (83). 更具前景的方法是通过使用迭代反褶积算法 固有的再循环模型 、空间独立分量分析来识别再循环成分或基于知识的数据驱动方法将其与再循环分离 从而保留了第一次通过的特征注射时机和血管运输。 COMMON PERFUSION PARAMETERS A number of useful haemodynamic parameters can be estimated from the deconvolved flow-scaled residue function. 从解卷积的流量标度残差函数可以估计出许多有用的血流动力学参数。 28. Cerebral blood flow (CBF) CBF max[R(t)] 29. Cerebral blood volume (CBV) CBV is a measure of the capillary density or microvascular blood volume, which is proportional to the total amount of intravascular contrast agent in the tissue (1). The integral of the tissue CTC gives the entire amount of contrast agent passing through the voxel. However, this integral must be normalised by the AIF integral to account for the amount of injected contrast agent. CBV是一种测量毛细血管密度或微血管血容量的方法 它与组织中血管内造影剂的总量成比例。组织CTC的积分给出了通过体素的全部造影剂量。然而 这个积分必须由AIF积分归一化 以说明注入造影剂的量。 30. Mean transit time (MTT) MTT is the characteristic time the blood (and therefore the contrast agent) spends in the capillary bed. As the tissue CTC is not the distribution of transit times in the capillary bed, the first-moment definition of MTT does not provide accurate estimates of the true MTT (89). In fact, the first moment of the CTC is more strongly dependent on arterial delay and dispersion (90), and therefore its use as a measure of tissue perfusion status may be potentially misleading. MTT是血液 以及造影剂 在毛细血管床上停留的特征时间。 由于组织CTC不是毛细血管床中传输时间的分布 MTT的第一时刻定义不能准确估计真实MTT 89 。事实上 CTC的第一个时刻更强烈地依赖于动脉延迟和弥散 90 因此 将其作为组织灌注状态的测量可能会产生潜在的误导。 MTT CBFCBV​ MTT is the characteristic time the blood (and therefore the contrast agent) spends in the capillary bed. As the tissue CTC is not the distribution of transit times in the capillary bed, the first-moment definition of MTT does not provide accurate estimates of the true MTT (89). In fact, the first moment of the CTC is more strongly dependent on arterial delay and dispersion (90), and therefore its use as a measure of tissue perfusion status may be potentially misleading. MTT是血液 以及造影剂 在毛细血管床上停留的特征时间。 由于组织CTC不是毛细血管床中传输时间的分布 MTT的第一时刻定义不能准确估计真实MTT 89 。事实上 CTC的第一个时刻更强烈地依赖于动脉延迟和弥散 90 因此 将其作为组织灌注状态的测量可能会产生潜在的误导。 31. Time to maximum (Tmax) However, its interpretation is complex. Tmax is estimated from the time of the maximum of the deconvolved R(t). In theory, Tmax is the arrival delay between AIF and the tissue CTC (which may be significant if there is collateral flow). Simulations have shown that Tmax is also influenced by the bolus dispersion between AIF and the tissue CTC (which may be significant if there is arterial stenosis/ occlusion) and, to a much lesser extent, by the tissue MTT (91). Tmax是中风研究中常用的参数。然而 它的解释是复杂的。Tmax是根据去卷积R t 的最大值的时间估计的。理论上 Tmax是AIF和组织CTC之间的到达延迟 如果有侧支循环 这可能很重要 。模拟显示 Tmax还受AIF和组织CTC之间的团注分散度的影响 如果存在动脉狭窄/闭塞 这可能是显著的 并且在较小程度上受组织MTT的影响。 Tmax is also affected by the slice acquisition order, and the discrete sampling of DSC-MRI data causes rounding errors. Calamante et al. (91) have recommended temporal interpolation of the data (see step 32) prior to Tmax calculation and correction of slice timing differences. Tmax还受切片采集顺序的影响 DSC-MRI数据的离散采样会导致舍入误差。卡拉曼特等人。在计算Tmax和校正切片定时差之前 建议对数据进行时间内插。 POST-PROCESSING POSSIBILITIES 32. Recovery of high-frequency information Improved accuracy of perfusion parameters can be achieved by frequency domain interpolation and extrapolation. Several studies have utilised zero filling of the high-frequency components in the Fourier domain in order to improve the temporal resolution of the deconvolved function R(t) (92), enabling more accurate estimates of temporal parameters 通过频域内插和外推可以提高灌注参数的精度。一些研究利用傅里叶域中高频分量的零填充来提高去卷积函数R(t)的时间分辨率 从而能够更准确地估计时间参数33. calculation of confidence intervals Calamante and Connelly (94) demonstrated how the precision of DSC-MRI perfusion parameters can be estimated using a wild-bootstrap method, which allows the estimation of confidence intervals from just one dataset. Such analysis is feasible in a clinical setting and could aid the interpretation of clinical findings. 通过频域内插和外推可以提高灌注参数的精度。一些研究利用傅里叶域中高频分量的零填充来提高去卷积函数R(t)的时间分辨率 从而能够更准确地估计时间参数CONSIDERATIONS FOR PATIENT STUDIES In patient studies, accurate perfusion quantification is often made more difficult as a result of issues of arterial bnormalities or BBB leakage. 在病人研究中 由于动脉异常或BBB渗漏的问题 准确的灌注定量往往变得更加困难。 34. Bolus delay Perfusion values obtained using the widely implemented standard SVD (sSVD) deconvolution algorithm (69) are highly dependent on the delay between the measured global AIF and the tissue CTC data (51), commonly termed the arterial–tissue delay (ATD) (72). 使用广泛实施的标准SVD sSVD 反褶积算法获得的灌注值高度依赖于被测的全局AIF与组织CTC数据之间的延迟 通常称为动脉-组织延迟 ATD 。 A more elegant solution uses the block-circulant AIF matrix (72), such that any ATD simply (circularly) shifts the deconvolved CBFR(t), yielding ATD-invariant perfusion estimates (Fig. 6). 一个不错的解决方案使用块循环AIF矩阵 使得任何ATD简单地 循环地 移动解卷积的CBFR t 产生ATD不变的灌注估计值 图6 。 When circular SVD is regularised using an adaptive oscillation index, it is known as oSVD (72). 当使用自适应振荡指数正则化循环SVD时 它被称为oSVD Despite the robustness and delay insensitivity of the FT and oSVD techniques, it is important to recognise that both methods will still underestimate CBF for short MTT tissue due to the unavoidable regularisation and discretisation errors. In stroke patients, the effect will be decreased image contrast between normal and ischaemic tissue. 尽管FT和oSVD技术具有鲁棒性和延迟不敏感性 但必须认识到 由于不可避免的正则化和离散化误差 这两种方法仍然会低估MTT组织的CBF。在中风患者中 正常组织和缺血组织之间的图像对比度会降低。 35. Bolus dispersion In stroke patients, delay and temporal dispersion of the contrast bolus may be considerable in the ipsilateral hemisphere as a result of stenosis, occlusions or collateral supply (99). 在脑卒中患者中 由于狭窄、闭塞或侧支供血 同侧半球造影剂推注延迟和时间分散可能相当大。 Although deconvolution algorithms can be made insensitive to delay (see step 34), dispersion cannot be easily accounted for in the deconvolution. As a result, CBF estimates obtained using a global AIF deconvolution analysis may be severely underestimated (and therefore MTT will be severely overestimated). 虽然反卷积算法可以对延迟不敏感 但在反褶积中 色散不能很容易地被考虑。因此 使用全局AIF反褶积分析获得的CBF估计值可能被严重低估 因此MTT将被严重高估 。 When interpreting perfusion maps calculated using a global AIF, it is important to be aware of the potential for errors, especially in stroke patients in whom arterial abnormalities may cause significant bolus dispersion (100). 在解释使用整体AIF计算的灌注图时 必须意识到可能出现的错误 特别是在动脉异常可能导致明显的丸粒弥散的中风患者中。 In theory, dispersion error may be circumvented using multiple regional or local AIFs, located downstream of any arterial abnormality, to deconvolve small branch arterial territories, e.g. refs. (30,31,90,101–104). In a validation study, Willats et al. (90) determined that local AIF methodologies reduced dispersionrelated perfusion error, and could therefore better characterise the severity and extent of a perfusion deficit. 理论上 通过使用位于异常动脉下游的多个区域性或局部性aif来解卷小分支动脉区域 如参考文献 可以避免离散误差。在一项验证研究中 Willats等人确定局部AIF方法减少了与弥散相关的灌注错误 因此可以更好地描述灌注不足的严重程度和程度。 Despite their benefits for accurate CBF quantification, the benefit of using local AIF methods for infarct prediction is still uncertain (104–106). An evaluation of the signal formation in local AIFs suggested that they were broader than the ground truth as a result of PVEs with the surrounding tissue (107,108). 尽管局部AIF方法有利于准确的CBF定量 但使用局部AIF方法预测梗死的益处仍不确定。对局部aif信号形成的评估表明 由于pve与周围组织的关系 它们比实际情况更宽。 When the measured AIF is broader than the true tissue input, perfusion will be overestimated, potentially masking any areas of hypoperfusion. 当测得的AIF比真实的组织输入更宽时 灌注将被高估 可能掩盖所有低灌注区域。 36. BBB leakage. 血脑屏障(blood-brain barrier,BBB)是指血管壁与神经胶质细胞形成的血浆与脑细胞外液间,以及由脉络膜丛形成的血浆与脑脊液间的屏障 One of the primary assumptions for perfusion quantification using DSC-MRI is that the BBB remains intact, so that the contrast agent remains intravascular and can be treated as a nondiffusible tracer (1). Where there is leakage, this model must be modified (110). 使用DSC-MRI进行灌注定量的一个主要假设是BBB保持完整 因此造影剂仍然是血管内的 可以作为不扩散示踪剂处理。如果存在泄漏 则必须修改此模型。 后面在讲一些MR后处理方法 这个不是我灌注的内容。 ABSOLUTE VERSUS RELATIVE QUANTIFICATION The question of whether absolute perfusion quantification is possible using DSC-MRI is an active area of research and a topic of controversy. In the previous steps, we have described many of the issues that impede accurate and absolute perfusion measurement. 37. Elative quantification As a result of the numerous factors that may introduce bias and error into perfusion estimates (see previous steps), clinical applications typically report relative perfusion measures. Relative values are referenced to normal-appearing tissue, such as contralateral white matter or grey matter. 由于许多因素可能会导致灌注估计的偏差和误差 参见前面的步骤 临床应用通常报告相对灌注测量。相对值是指参照正常组织 如对侧白质或灰质。 38. Absolute quantification Absolute quantification is difficult, not only because of the many potential imaging and data processing artefacts (see previous steps), but also because the proportionality constants in the various relationships are not accurately known. 绝对量化很困难 这不仅是因为许多潜在的成像和数据处理伪影 而且还因为各种关系中的比例常数值不明了。 AUTOMATED ANALYSIS Software that automates and standardises DSC-MRI perfusion analysis is important in a clinical context. However, because of the numerous quantification issues described in the previous steps, it is important not to treat perfusion quantification as a black-box analysis. DSC-MRI灌注分析自动化和标准化的软件在临床应用中非常重要。然而 由于在前面的步骤中描述了大量的量化问题 所以不要将灌注量化视为黑盒分析。 39. The use of DSC-MRI software There are several deconvolution-based perfusion software packages that are either freely available for research 有几个基于反褶积的灌注软件包可以免费用于研究 PerfToolhttp://www.cybertrial.ca/Penguinhttp://www.cfin.au.dk/software/pgui/NordicICEhttp://www.nordicneurolab.com/commercially availableStrokeToolhttp://www.digitalimagesolutions.de/commercially available文章提供的几个网址 第一个要邮箱 第二个不让下了 后两个是收费的 As expected, no software is perfect, and the advantages of such automated approaches need to be weighed against their limitations on an application-by-application basis. 正如预期的那样 没有一个软件是完美的 这种自动化方法的优势需要与它们在应用程序基础上的局限性进行权衡。 [1] Willats L, Calamante F. The 39 steps: evading error and deciphering the secrets for accurate dynamic susceptibility contrast MRI[J]. NMR in Biomedicine, 2013, 26(8): 913-931. [31] Willats L, Connelly A, Calamante F. Minimising the effects of bolus dispersion in bolus‐tracking MRI[J]. NMR in Biomedicine: An International Journal Devoted to the Development and Application of Magnetic Resonance In vivo, 2008, 21(10): 1126-1137. [37] King M D, Calamante F, Clark C A, et al. Markov chain Monte Carlo random effects modeling in magnetic resonance image processing using the BRugs interface to WinBUGS[J]. Journal of Statistical Software, 2011, 44(1): 1-23. 【MATLAB】椭圆检测（Ellipse Detection）算法（含代码）15835 【MATLAB图像分割结果可视化】利用伪彩色更好的展示你的分割结果（含代码）14620 抵扣说明：1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。 2.余额无法直接购买下载，可以购买VIP、C币套餐、付费专栏及课程。