CADx-4/research-wiki/scratch/2026-10-17-TIC拟合约束文献核查/C-guidelines.md

C 组：DCE-US 定量方法学指南 / 共识 / 方法综述核查

结论先说：这些指南和方法学文献都没有给出 TIC 拟合参数的边界、峰位置约束或拟合优度的数值门槛（R²、NRMSE、拟合质量百分比）。它们明确要求、而我们做不到的是：在线性化数据上拟合；用参考组织 ROI（同一深度的正常组织）做归一化。胰腺章节（EFSUMB 2017、SIUMB 2020）只讲定性增强模式，没有 TIC 定量的条目。

文本抽取在 /tmp/claude-1000/-home-iomgaa-Projects-CADx-4/d2239309-a620-4197-8b4b-d4e97e3a3537/scratchpad/lit/txt/，文件名与 PDF 相同。

1. 文献清单

下表中带 * 的文件是用无头浏览器把开放获取的 HTML 全文页（PMC 或出版社页面）打印成的 PDF，不是出版社排版的 PDF，所以没有原刊页码，下文按章节标题定位。

#	文件名（reference-pdf/）	标题	年份 / 期刊	DOI	类型	全文？
G1	2018-sidhu-efsumb-ceus-non-hepatic.pdf（本地原有）	The EFSUMB Guidelines and Recommendations for the Clinical Practice of CEUS in Non-Hepatic Applications: Update 2017 (Short Version)	2018, Ultraschall Med 39:154–180	10.1055/s-0044-101254	指南	是，读了胰腺、CE-EUS、Tumor Response Assessment（DCE-US）三节
G1L	2018-sidhu-efsumb-ceus-non-hepatic-long.pdf（新下载，TU/e 机构库）	同上，Long Version	2018, Ultraschall Med 39:e2–e44	10.1055/a-0586-1107	指南	是。胰腺和 DCE-US 两节正文与短版相同，只多了参考文献编号，下文引短版页码
G2	2024-dietrich-efsumb-dceus-update.pdf *	EFSUMB Technical Review – Update 2023: Dynamic Contrast-Enhanced Ultrasound (DCE-CEUS) for the Quantification of Tumor Perfusion	2024, Ultraschall Med 45:36–46	10.1055/a-2157-2587	指南级技术综述（EFSUMB 立场文件，取代 2012 版 introduction）	是（出版社 OA 的 HTML 全文）
G3	（未取得）	An EFSUMB introduction into Dynamic Contrast-Enhanced Ultrasound (DCE-US) for quantification of tumour perfusion（Dietrich 等）	2012, Ultraschall Med 33:344–351	10.1055/s-0032-1313026	指南级技术文件	否，付费，只读了摘要
G4	2018-dietrich-how-to-perform-ceus.pdf *	How to perform Contrast-Enhanced Ultrasound (CEUS)（Dietrich, Averkiou, Nielsen 等）	2018, Ultrasound Int Open 4:E2–E15	10.1055/s-0043-123931	技术共识（被 G2 引作「consensus paper」，即 G2 的参考文献 [93]）	是（PMC 全文）
G5	2020-donofrio-siumb-focal-pancreatic-lesions.pdf *	SIUMB recommendations for focal pancreatic lesions	2020, J Ultrasound 23:599–606	10.1007/s40477-020-00522-2	国家学会建议（胰腺专门）	是（PMC 全文）
R1	2011-tang-quantitative-ceus-variability.pdf *	Quantitative contrast-enhanced ultrasound imaging: a review of sources of variability	2011, Interface Focus 1:520–539	10.1098/rsfs.2011.0026	综述	是（PMC 全文）
R2	2020-turco-ceus-quantification-kinetic-ml.pdf（TU/e 机构库）	Contrast-Enhanced Ultrasound Quantification: From Kinetic Modeling to Machine Learning	2020, Ultrasound Med Biol 46:518–543	10.1016/j.ultrasmedbio.2019.11.008	方法综述	是
M1	（未取得）	Indicator dilution models for the quantification of microvascular blood flow with bolus administration of ultrasound contrast agents（Strouthos 等）	2010, IEEE TUFFC 57:1296–1310	10.1109/TUFFC.2010.1550	方法论文	否，只读了摘要
M2	（未取得）	Contrast ultrasonography: necessity of linear data processing for the quantification of tumor vascularization（Peronneau 等）	2010, Ultraschall Med 31:370–378	10.1055/s-0029-1245450	方法论文	否，只读了摘要
M3	（未取得）	Evaluation of the Reproducibility of Bolus Transit Quantification With CEUS Across Multiple Scanners and Analysis Software Packages — A QIBA Study（Averkiou 等）	2020, Invest Radiol	10.1097/RLI.0000000000000702	方法论文（体模）	否，只读了摘要
M4	（未取得）	Standardization of DCE-US for the Evaluation of Antiangiogenic Therapies: the French Multicenter STIC Study（Lassau 等）	2012, Invest Radiol 47:711–716	10.1097/RLI.0b013e31826dc255	方法 / 标准化（多器官肿瘤）	否，只读了摘要
M5	2011-gauthier-dceus-intraoperator-variability.pdf *	Estimation of intra-operator variability in perfusion parameter measurements using DCE-US	2011, World J Radiol 3:70–81	10.4329/wjr.v3.i3.70	方法论文（体模与小鼠）	是，但它用的是 IGR 模型（Gustave Roussy 研究所的专利拟合方程），不是 Lognormal，与我们的约束关系很弱，下文只引一处

查过但没有找到的：

• 2020 年以后 WFUMB 或 EFSUMB 专门针对胰腺 CEUS 或 CEUS 定量的新指南。Europe PMC 标题检索加 WebSearch，只查到 EFSUMB 2023 DCE-US 技术更新（G2）和 2023–2024 年「Comments and illustrations of the EFSUMB guidelines」系列。后者讲罕见胰腺肿瘤的病例图示，不涉及定量。
• Arditi/Rognin 关于 VueBox 线性化的论文：Rognin 2008 是 IEEE Ultrasonics Symposium 会议论文，没找到开放全文；Arditi 2006 讲的是破坏-再灌注（infusion），不是团注。两篇都没有下载。

2. 逐条约束

约束 1：团注 TIC 用 Lognormal 模型是否被推荐；显示域（对数压缩）数据能否直接拟合

1a. Lognormal 模型

• G4，章节 "“Wash-in” and “wash-out”"，Fig. 3 图注。这是指南级文件里唯一出现团注 Lognormal 的地方，但它只是体模示例，不是推荐条款：

  "Time-intensity curve for a bolus injection in a tissue mimicking flow phantom. A lognormal curve (solid line) is fitted to the data and it is used to calculate the important quantification parameters (44) (PI, RT, MTT, and AUC)."

• G2，章节 "Clinical aspects of a DCE-US protocol → How to perform DCE-US…"。只要求拟合，没有指定模型：

  "Next, a curve is fitted to the TIC data and the important perfusion parameters (rise time RT, mean transit time MTT, peak intensity PI, and area under the curve AUC) are extracted."

• G2，章节 "CEUS time intensity curve parameters"：

  "Different from the other parameters, MTT can be calculated only in combination with a fitted mathematical model, while the other parameters are curve-descriptive parameters and thus can be derived also without a dedicated model."

• G2 里 "Lognormal perfusion model" 只出现在破坏-再灌注（infusion）的 Arditi–Hudson 模型描述中（同一章节），不是团注模型：

  "This model has 3 components that were not present in Wei’s model: accounts for tissue perfusion through realistic microvascular geometry (Lognormal perfusion model), …"

• G1 p.178–179（Tumor Response Assessment → Study procedure）：

  "Analyses of the TIC, including wash-in and washout times, can be performed with curve fitting to determine functional indices."

• R2 p.524（TEMPORAL ANALYSIS）。综述，不是指南：

  "Among the non-compartmental models, the log-normal distribution is one of the models most commonly used to fit TICs measured by DCE-US (Wise 1966; Band et al. 1997; Krix et al. 2004; Strouthos et al. 2010; …)." "Fitting the measured TIC to a suitable mathematical model permits filtering out the noise and extracting the first pass of the contrast bolus."

• R1 §1.2：

  "Given the noisy nature of the data, models are commonly used to fit the data to remove the noise and extract relevant indices [8,10–12]."

• M1，仅摘要（数据来自肝转移瘤和绵羊黄体）：

  "The models that best fit the experimental data are the lognormal function and the diffusion with drift."

小结：没有任何指南把 Lognormal 写成团注 TIC 的推荐模型。G2 只说「拟合一条曲线」；G4 拿 Lognormal 作示例图；R2 说它「最常用之一」。

1b. 显示域（对数压缩）数据能否直接拟合：反对的证据明确，且来自指南级文件。

• G2，"CEUS time intensity curve parameters"：

  "All time and intensity values should be calculated from a curve fitted to the linearized echo intensity values and not from image data."

• G2，Statement 11：

  "It is necessary to perform calculations on linearized data to maintain the linear relationship between microbubble concentration and signal intensity."

• G2，"General considerations"：

  "The average CEUS signal intensity within a region of interest (ROI) is calculated in linear units and is displayed as a function of time, i.e., a time-intensity curve (TIC)…"

• G2，"How to perform DCE-US…"：

  "Save the DICOM video loop in a format that allows data linearization."

• R1 §1.3 "Linear versus logarithmic data"：

  "Early studies of quantification relied on video intensity data, to which logarithmic compression had been applied. It has been gradually recognized that linear echo data (sometimes called ‘raw data’) acquired before compression should be used instead [21,22]." "…a direct acquisition before log compression is preferred since the compression process may cause information loss that cannot be fully recovered by reversing the compression…"

• R1 §2.3 "Dynamic range and gain"。这一条对「时间参数」有利，但它比较的是「线性化后的压缩数据」与「原始线性数据」，不是我们这种未经线性化的显示灰度：

  "…the errors in the peak intensity and AUC derived from linearized log-compressed data, with linear data as references, were found to be 45 ± 18% and 39 ± 16%, respectively, under a typical clinical dynamic range setting of 30 dB and a gain of 55 dB. … Time-related parameters on the other hand were found to be much less affected."

• R2 p.523：

  "Thus, processing log-compressed video data without appropriate linearization induces quantification results, which are dependent on user and system settings." "…perfusion quantification based on linearized log-compressed video data can be equivalent to that based on raw data only under two conditions: (i) the gain settings should be adjusted in such a way to prevent data truncation or saturation in images, and (ii) a sufficiently high dynamic range of log compression (approx >45 dB) should be used." "…in many cases ultrasound systems apply non-linear and sometimes non-reversible signal processing or image processing … In such cases, the linearization endeavor becomes futile, and results are bound to be inaccurate."

• M2，仅摘要（小鼠黑色素瘤）：

  "TPI was found to be similar with either data set (r = 0.98, p < 0.05, factor of 1.09)." "In conclusion, the use of linear data is the only mathematically valid methodology for determining a tumor's time-intensity curve…"

• M3，仅摘要（体模）：

  "Currently, it is not possible to directly compare amplitude values from different scanners and analysis software packages owing to the arbitrary linearization algorithm used among manufacturers; however, it is possible for time parameters (RT and MTT)."

• G2，"Further technical and methodological aspects"：

  "Time-related parameters (rise time) are robust since they do not depend on the acoustic signal level – if the bolus arrival time is subtracted to avoid circulation time dependencies."

• G4，"Dynamic range"：

  "For perfusion quantification studies, a wide dynamic range should be used to avoid signal saturation."

约束 2：参数边界（峰高、峰时刻在平滑曲线峰 ±5 s 内、σ 0.08–2.5、t0、基线）；参数贴边判失败

未找到依据。 读过的全文里，没有一份给出 Lognormal（或任何团注模型）参数的取值范围、初值策略，或「参数贴边即判失败」的规则。最接近的两句话只要求记录拟合质量：

• G2，"Further technical and methodological aspects"：

  "However, the quality of the fit must be recorded to avoid misinterpretation."

• R2 p.526 提到最小二乘之外的方法，同样没有边界：

  "Next to standard least-squares approaches, several authors have introduced maximum-likelihood approaches that account for the statistics of arrival times (measured TIC) as well as noise, which has non-stationary characteristics depending on the UCA concentration levels…"

约束 3：模型峰不能落在拟合区间头尾

未找到直接依据。 相关但不等同的证据：

• G4，"CEUS phases"。要求录像包含峰：

  "A useful sequence is to scan continuously and record a cine loop from the earliest arrival of the microbubbles to include the peak of arterial enhancement, and up to 60 s."

• M4，仅摘要。「是否完整采到洗入期」是一项质量判据：

  "Five criteria were found to have a major impact on quality: lesion size, motion, loss of target, borders, and total acquisition of wash-in."

• R1 §3.2.3。尾部受再循环影响，有人只用洗入段：

  "The effect of recirculation on the TIC would manifest itself as a small increase in contrast intensity approximately 60 s following the start of agent administration." "To address the problems associated with recirculation, some researchers have considered for bolus injection only the wash-in phase of the curve [33,34], or limited the time window post-injection during which scanning takes place [66]."

约束 4：拟合优度门槛（R²、NRMSE、拟合质量百分比）；不合格怎么处理

没有数值门槛，未找到依据。 指南只要求记录拟合质量。

• G2，"Further technical and methodological aspects"：

  "However, the quality of the fit must be recorded to avoid misinterpretation."

• R2 p.526，Fig. 6 图注。这是参数图（逐像素）的做法，例子是前列腺，不是规则：

  "Gaps in the color map correspond to time–intensity curve fits with poor determination coefficients."

• G1 p.179。这里的质量评分针对采集质量，不针对拟合：

  "A quality score was proposed in a standardized acquisition, with AUC being the best parameter."

• M4，仅摘要。0–5 分的采集质量分，0 分视为不可解读：

  "The final quality score, ranging from 0 to 5, was defined according to the value of coefficient of variability for each criterion." "Only 3% of examinations carried out at these centers were considered not interpretable."

约束 5：ROI 怎么画（整个病灶 vs 增强部分），能否并用

有依据支持「整病灶 ROI 为主，可加增强区 ROI」。 但同一出处写明尚无共识，而且指南还要求一个参考组织 ROI。

• G2，"How to perform DCE-US…"：

  "One ROI should cover the whole tumor, and the placement of optional additional ROI(s) should follow representative areas of the “whole tumor” guided by highly vascularized parts of the tumor. For early relapse prediction, focusing on highly vascularized ROIs may be useful. In partially necrotic tumors, this guidance can make an important difference. For some of the mentioned recommendations, no consensus has been reached so far."

• G2，"General considerations"：

  "Additional ROIs can be placed in a reference tissue for comparison purposes or in different areas of the lesion."

• G2，"Perspectives"：

  "Placing more than one ROI in a tumor, e.g., in the periphery and the center, is the simplest approach to add a spatial analysis." "For instance, the size of the AUC above a certain threshold/size of the whole tumor can reflect the vascularization – similar to “% of vascularized tumor”."

• 参考 ROI 与归一化（我们目前没有）：
  • G2，Statement 6：

    "Relative quantification of perfusion using a reference area at the same depth should be preferred to absolute evaluation of contrast enhancement."

  • G2，"Further technical…"：

    "The second ROI should be placed at the same depth." "Due to signal linearization, large vessels should not be used as the standard ROI for comparison."

• R1 §3.3：

  "For example, a large vessel full of bubbles moving in and out of the ROI could cause intensity variations of hundreds of per cents."

• 胰腺专门条目：
  • G1 p.162（Pancreas → Pancreatic Masses），定性增强，要求视野同时包含病灶和邻近胰腺：

    "The enhancement pattern of focal pancreatic lesions is compared with the adjacent pancreatic tissue. The field of view should include both. This is mandatory with an isovascular mass but not essential with a hypovascular (hypoenhanced with few internal microbubbles) or hypervascular (hyperenhanced) mass."

  • G1 p.162（Mucin-producing cystic tumors）。「有灌注区 vs 无灌注区」的区分，与 roi_perf 的思路相近，但这一段是定性的：

    "CEUS is helpful for differentiating between perfused (nodules) and non-perfused (mucin plugs) areas."

约束 6：洗入期截断、运动伪影、采集时长、录制起点

采集时长

• G2，Statement 5：

  "The bolus technique quantifies the entire course of contrast kinetics, from wash-in to wash-out. The analysis is carried out along one single plane for each injection and a cineloop of at least one minute in duration is recommended."

• G2，"How to perform DCE-US…"：

  "Scan continuously for 2 up to 5 minutes (depending on the clinical application) avoiding bubble destruction."

• G2，"CEUS time intensity curve parameters"：

  "Most studies focusing on AUC and wash-out recorded 3-minute loops [68]."

• 帧率：
  • G2，"General considerations"：

    "…single-plane imaging at a low mechanical index (MI) is usually performed at about 10 frames per second for the duration of the enhancement. Frame rates that are too high should be avoided to prevent bubble destruction."

  • G1 p.178：

    "Single plane imaging is usually performed at 10 – 20 frames per second for the duration of the enhancement."

• 胰腺时相，出自 G1 p.162。G5 的 "Approval > 95%" 一段有基本相同的文字：

  "Enhancement begins immediately after aortic enhancement, with an arterial phase (10 to 30 s), a venous phase (30 to approximately 120 s)."

录制起点 / 洗入期截断

• 没找到「录制必须在注射前开始」的原句。最接近的有三条：
  • G4，"CEUS phases"：

    "…record a cine loop from the earliest arrival of the microbubbles to include the peak of arterial enhancement…"

  • G4，"Contrast timer"。这里讲的是计时器，不是录像：

    "This timer should be started at the time of the beginning of the UCA injection for SonoVue™/Lumason™."

  • M4，仅摘要："total acquisition of wash-in" 是影响质量的主要判据之一（原句见约束 3）。
• 对「截断后怎么补救」（例如外推 t0、丢弃该例），未找到依据。

运动

• G2，Statement 10：

  "Suitable planning and choice of a representative imaging plane is crucial to avoid respiratory motion of the ROI which is a major source of error in the quantification of DCE-US. Especially out-of-plane motion cannot be corrected, and out-of-plane acquisitions must be excluded from the DCE-US analysis, which is a time-consuming and demanding process."

• G2，"Clinical Applications"：

  "Techniques such as respiratory gating [59][66][67] and motion correction have been shown to improve the reproducibility of DCE-US measurements."

• G2，"How to perform DCE-US…"：

  "In order to keep the probe stationary, be aware of and compensate for any motion."

• R1 §3.3：

  "Various motion correction algorithms are available that can deal with translation, rotation or non-rigid deformation of tissues. While most quantification packages have already incorporated such algorithms, they are only able to deal with in-plane motion. Out–of-plane motion means different tissues are imaged at different times, and this type of motion is impossible to correct for with scans on only a single plane. One way to deal with this is to track the motion and only use a subset of the image sequence that corresponds to the same anatomical cross section."

• R2 p.537：

  "As DCE-US recording can take up to several minutes, several sources of intrinsic (e.g., breathing) and extrinsic (e.g., probe tilting) motion unavoidably corrupt DCE-US data. As a result, motion-compensation strategies should be employed prior to quantitative analysis."

• G4，"Background signal (noise)" 节（双幅显示段落）：

  "For quantitative studies, it is critical to maintain the transducer at the same place and avoid motion."

• G4，"Pre-contrast examination"：

  "…the optimal scan plane along the axis of the respiratory movements (usually longitudinal) to minimize out-of-plane motion from respiration."

胰腺指南对定量的立场

• G5，Methods。SIUMB 在纳入文献时明确排除了定量研究：

  "studies providing a qualitative evaluation of the contrast agent (we excluded studies providing a quantitative evaluation based on wash in/wash out times or image interpretation using software such as Photoshop etc.);"

• G1 p.169。唯一提到胰腺 TIC 的地方，针对的是超声内镜造影（CEH-EUS）下的实性病灶，不是经腹 CEUS，也不是囊性病灶：

  "With this technique, microvasculature and perfusion comparative qualitative or quantitative (time intensity curves, TIC) analysis may be performed."

3. 我没把握的地方

• EFSUMB 2012 introduction（G3）没读到全文，它是付费文献，只读了摘要。我用 2023 更新版（G2）代替。G2 说参数定义 "are explained in detail [3]"，也就是 2012 版里可能有 T0、TP 等参数的定义细节，甚至拟合注意事项，我没核到。
• Strouthos 2010、Peronneau 2010、Averkiou 2020、Lassau 2012 只读了摘要。其中 Lassau 2012 的质量评分各项判据怎么定义（例如 "total acquisition of wash-in" 怎么判），只有全文里有。
• 带 * 的 PDF 是 HTML 打印件，没有原刊页码，引用按章节标题定位。G1 用的是出版社 PDF 页码（短版 154–180）。R2 用的是 TU/e 机构库副本上印的原刊页码。
• R1 §2.3 和 M2、M3 都说「时间参数受线性化影响小」，但它们比较的都是「（线性化后的）视频数据 vs 原始线性数据」，或「不同厂商的线性化」。没有一篇直接检验「完全不线性化的显示灰度上拟合 Lognormal，得到的时刻参数是否可靠」。这一点能否外推到我们的数据，我判断不了。
• G2 和 R1 讲的运动、ROI、时长要求，主要来自肿瘤疗效评估和肝脏场景。它们是一般方法学要求，没有专门针对胰腺囊性病灶的版本。
• 2020 年后若有 WFUMB 胰腺 CEUS 指南，我的检索（Europe PMC 标题检索加两次 WebSearch）没找到，但不能排除检索遗漏。
• 本地下载目录 scratchpad/lit/dl/ 里有其他并行任务的文件，我用的是 scratchpad/lit/dlC/。reference-pdf/INDEX.md 没有改。