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

# A：胰腺囊性病变 CEUS / CE-EUS 的 TIC 定量文献核查

总体结论：胰腺囊性病变里真正做了 TIC 定量、而且我读到全文方法部分的原始研究只有 4 篇（Vasile 2012、Yamamoto 2016、Saito 2017、Tang 2026）。这 4 篇都没有用 Lognormal 或其他参数模型拟合，没有报告 R²、NRMSE 之类的拟合优度，也没有给参数设边界。约束 2、3、4（门槛部分）都**未找到依据**。约束 4（怎么处理不合格病例）、5、6 有零散依据：分析不了的病例一律直接剔除；ROI 多放在实性成分、壁结节或囊壁上；有一篇剔除了「首帧已有造影剂」的病例。

## 1. 文献清单

### 1a. 读了全文方法部分，且确实做了 TIC 定量（构成依据）

| # | 文件 / 全文来源 | 标题 | 年份 / 期刊 | DOI / URL | 人群与例数 | 设备、软件、模型 | 全文 |
|---|---|---|---|---|---|---|---|
| V | `reference-pdf/2012-vasile-ceus-tic-cystic-pancreatic-masses.pdf` | Added value of intravenous contrast-enhanced ultrasound for characterization of cystic pancreatic masses: a prospective study on 37 patients | 2012, Medical Ultrasonography 14(2):108-114 | PMID 22675710；https://medultrason.ro/medultrason/index.php/medultrason/article/view/296 | 罗马尼亚 Cluj-Napoca。经腹 CEUS，SonoVue 2.4 ml。纳入 41 例，剔除 4 例，分析 37 例：假性囊肿 12、SCN 10、MCN 5、囊腺癌 6、SPN 2、IPMN 2 | GE Logiq 7；在 Origin 8 里画 TIC；直接从曲线读参数（PI、TTP、GRAD、TTG、AUC），未拟合模型；单位为 dB、dB×s | 是（PDF） |
| Y | `reference-pdf/2016-yamamoto-ch-eus-tic-ipmn.pdf`（冈山大学机构库里的作者稿，即学位论文全文） | Contrast-enhanced harmonic endoscopic ultrasonography with time-intensity curve analysis for intraductal papillary mucinous neoplasms of the pancreas | 2016, Endoscopy 48:26-34 | 10.1055/s-0034-1393563 | 日本冈山。CH-EUS，Sonazoid。30 例手术切除的有壁结节 IPMN（LGD/IGD 14 例，HGD/浸润癌 16 例） | Aloka ProSound α-10 机载软件；帧率 10–11 Hz；从曲线读参数，未拟合模型；强度用 “Level” 单位，乘 0.173 换算为 dB | 是（作者稿 PDF，26 页） |
| S | Europe PMC 全文 XML，PMC5494864（出版社 PDF 被反爬拦下，未入 `reference-pdf/`） | Differential diagnosis between intraductal papillary mucinous neoplasm with an associated invasive carcinoma and pancreatic ductal adenocarcinoma on ultrasonography: the utility of echo intensity and contrast enhancement | 2017, Ultrasonography 36(3):260- | 10.14366/usg.16039 | 日本札幌。经腹 CEUS，Sonazoid。IPMN 合并浸润癌对比 PDAC，增强定量最终纳入 41 例 | Toshiba Aplio；Image Lab；只取 ΔTI（dB）和 slope（dB/s），未拟合模型 | 是（XML 全文） |
| T | Europe PMC 全文 XML，PMC12896552（MDPI PDF 被拦下） | A Preliminary Study on Contrast Enhanced Ultrasound Characteristics of Solid Pseudopapillary Neoplasms and Pancreatoblastoma in Children | 2026, Diagnostics 16(3):474 | 10.3390/diagnostics16030474 | 中国重庆儿童医院。经腹 CEUS，SonoVue，左右分屏。66 例中纳入 22 例（SPN 19、胰母细胞瘤 3），均为儿童 | Siemens Sequoia 机载 SonoLiver（TomTec Image-Arena 4.3）；14 个相对参数；文中未写拟合模型 | 是（XML 全文） |

### 1b. 读了全文，但只做定性 CEUS、没有 TIC（作为「这批文献不做定量」的反面证据）

| 文件 | 标题 / 年份 / 期刊 | DOI | 说明 |
|---|---|---|---|
| `2015-fan-ceus-cystic-pancreatic-classification.pdf` | Application of CEUS in Cystic Pancreatic Lesions Using a Simplified Classification Diagnostic Criterion, 2015, BioMed Res Int | 10.1155/2015/974621 | 64 例，按形态分 I–IV 型，无 TIC。讨论部分引用 Vasile 2012 作为定量研究的先例 |
| `2019-wang-ceus-classification-pancreatic-cystic.pdf` | The Value of CEUS Classification in Diagnosis of Pancreatic Cystic Lesions, 2019, BioMed Res Int | 10.1155/2019/5698140 | 84 例，无 TIC |
| `2020-sun-ceus-ct-mri-pancreatic-cystic.pdf` | Comparative Diagnostic Evaluation with CEUS, CT and MRI in PCNs, 2020, Cancer Manag Res | — | 90 例，无 TIC |
| `2024-yuan-ceus-pcn-differential.pdf` | The differential diagnosis of PCNs with conventional US and CEUS, 2024, QIMS 14(7):4304 | 10.21037/qims-24-154 | 100 例，四类 PCN，只做定性特征的 logistic 回归，无 TIC |
| PMC5338980（XML） | Sun Y et al. Discrimination of serous cystadenoma from mucinous cystadenoma ... prospective study in 61 patients, 2017, OncoTargets Ther | 10.2147/OTT.S125497 | 定性特征，无 TIC |
| PMC9497750（XML） | Yamashita Y et al. Usefulness of CH-EUS for Diagnosis of Malignancy in IPMN, 2022, Diagnostics | 10.3390/diagnostics12092141 | 定性，只在讨论里引用 Yamamoto 2016 |
| WJG 2015;21(20):6252（WebFetch 读的全文页，内容经小模型转述，可信度低于自己读） | Harima H et al. Differential diagnosis of benign and malignant BD-IPMN using CE-EUS | — | 只看有无血流，无 TIC |

### 1c. 只读到摘要（不作为依据）

- Miyamoto K et al. *Efficacy of diagnosing IPMN with mural nodules by CE-EUS using TIC analysis with a new support program: a multicenter retrospective study*. Pancreatology 2025. DOI 10.1016/j.pan.2025.09.035。32 例，IPMN 壁结节的 TIC。付费墙。
- Itoh T et al. *Usefulness of contrast-enhanced transabdominal ultrasonography in the diagnosis of IPMT*. Am J Gastroenterol 2005;100:144-152。21 例，Levovist 造影剂，用 HDI Lab 测强度变化。超出 15 年范围，未读全文。
- Clin Hemorheol Microcirc ch231932（SPN 对比 pNET 的 CEUS）。页面抓不下来，连是否做了 TIC 都无法确认。
- 排除：Omoto 2017 近畿大学学位论文（CH-EUS TIC）研究的是实性胰腺肿瘤，方法里写明 “Pancreatic cystic tumors with tumor solids content less than 25% were excluded”，不属于本组范围。

## 2. 逐条约束

### 约束 1：用 Lognormal 拟合胰腺病灶 TIC 是否有先例；用显示域灰度还是线性化数据

**Lognormal 拟合：未找到依据。** 4 篇做了 TIC 的研究都直接从原始曲线读参数，没有拟合任何参数模型：

- V（`2012-vasile...pdf`，第 109 页，方法）："The quantitative analysis consisted of automatic plotting of time / intensity curves (TIC) in selected 5 mm areas situated in PCM, in the lesion wall and a proximity artery. The curves were drawn using the Origin 8 software." 第 109 页 Table I 给的参数定义是 "Peak Intensity (PI) Max(I(t))"、"Time to Peak (TTP) tm for which I(t)=Max"、"Area Under the Curve (AUC) ∫I(t)dt"，都直接在原始曲线上计算。
- Y（`2016-yamamoto...pdf`，作者稿第 7 页，"Time–intensity curve analysis"）："The echo intensity in the ROI was quantified, and the time–intensity curve was calculated using the software program built into the ultrasonography imaging system." 图 2 图注："Ipeak − Ibase, echo intensity change from baseline to peak ... (Ipeak − I120)/Ipeak, echo intensity reduction rate"。
- S（PMC5494864，Methods）："The chronological changes in the signal strength in the ROI set at the center of the solid part of the tumor were measured using Image Lab ver. 3.01 (Toshiba Medical Systems Corp., Tokyo, Japan)... The parameters of the increase in echo intensity (ΔTI) and echo intensity per unit of time (slope) were used to evaluate contrast enhancement."
- T（PMC12896552，2.5 Quantitative Analysis）："Regions of interest (ROIs) were manually delineated on dynamic images using Sonoliver software to obtain time–intensity curves (TICs) and quantitative parameters." 全文没有提到拟合模型的名字。

**显示域还是线性化：** 没有一篇写明是否做过线性化。只能看到强度单位用的是 dB，也就是对数域：

- V，第 112 页 Table III 表头："nPI (dB) ... nAUC (dBxs)"。
- Y，作者稿图 3 图注："(“Level” value × 0.173 = value in decibels [dB].)"
- S，Results："IPMN-IC exhibited a significantly greater ΔTI than PDAC (IPMN-IC, 14.9±4.4 dB; PDAC, 8.8±6.0 dB; P=0.012)."
- T：未说明强度的域。

### 约束 2：参数范围与约束（峰高、峰时刻 ±5 s、σ 0.08–2.5、t0、基线），参数贴边界判失败

**未找到依据。** 4 篇都没有拟合模型，所以也就没有参数边界，没有「贴边界即失败」的规则。

### 约束 3：模型峰不能落在拟合区间头尾

**未找到依据。** 唯一沾边的是 Y 规定了固定的观察窗：作者稿第 6–7 页 "The nodule was then observed for 120 s continuously"，参数之一定义为 "echo intensity reduction rate from peak to 120 s after injection"。这只是定义了取值窗口，不是对峰位置的检查。

### 约束 4：拟合优度门槛（R²、NRMSE），不合格病例怎么处理

**R²、NRMSE 门槛：未找到依据。** 4 篇都没有报告拟合优度。

**不合格病例怎么处理：4 篇都在纳入阶段直接剔除，没有人工复核后再纳入的做法。**
- V，第 109 页："The exclusion criteria were: allergies to the i.v. contrast agent, inadequate ultrasound quality, lack of precision of quantitative measurements due to out-of-plane movements... four patients were finally excluded from the study (2 with inadequate ultrasound quality and 2 with lack of precision of quantitative measurements due to out-of-plane movements)."
- V，第 113 页（局限性）："By choosing smaller samples, the number of subjects rejected from our study may have increased due to out-of-plane movements, increasing the errors of the method we proposed."
- S，Methods："For the quantitative analysis of contrast enhancement, three cases in which the contrast agent had already been injected in the initial frame or in which quantitative analysis was difficult due to poor respiratory control were excluded. Ultimately, 41 cases were included in the quantitative evaluation of contrast enhancement."
- T，2.1 排除标准第 (5) 条："Insufficient image quality after software correction to complete TIC analysis and quantitative parameter assessment."
- Y，作者稿第 13 页（局限性）："it was sometimes difficult to obtain target images continuously with EUS during this procedure, which in turn made it difficult to measure the time–intensity curve." 纳入条件本身就要求 "IPMNs with mural nodules for which CH-EUS and time–intensity curve analysis had been performed"（作者稿 "Patients and study design" 节）。

### 约束 5：两个 ROI（整体 vs 只取增强部分）都要拟合成功；论文里 ROI 怎么画

**「两个 ROI 都成功才算」：未找到依据。** 下面是各篇 ROI 的实际画法：

- 只画实性成分或壁结节、避开囊液（与我们的 roi_perf 思路一致）：
  - T，2.5："Place the ROI within the most intensely enhanced solid region of the lesion, avoiding areas with grossly visible cystic changes, necrosis, and calcifications. Place an ROI of identical size within normal tissue at the same depth to minimize variability."
  - S，Methods："The ROI for IPMN-IC was placed in the center of the mural nodule in a way that enabled the exclusion of cystic lesions and branched pancreatic ducts from the ROI."
  - Y，作者稿第 7 页："Two circular regions of interest (ROI) were placed, on the mural nodule and normal pancreatic parenchyma, respectively. In most cases, we put the ROI on the base of the nodule. If there were multiple nodules in IPMNs, we performed time–intensity curve analysis for the largest nodule."
- 囊壁、囊腔各放一个固定大小的小 ROI，再用比值归一化（与「整体 + 增强部分」两 ROI 并用最接近，但不是同一回事）：
  - V，第 109 页："in selected 5 mm areas situated in PCM, in the lesion wall and a proximity artery... Normalized Param = (Param. wall – Param. cavity)x(Param. artery/Param. wall)."
  - V，第 113 页："we were challenged for the quantitative analysis of PCM to find a single, common and relevant denominator which has been considered the contrast-enhancement within the cystic wall... the need of a standardization has imposed the use of a constant size of samples, which we have arbitrarily chosen to be 5 mm, size which may not be appropriate in all individual situations (e.g. thin cystic walls...)"
- 描整个病灶轮廓：只出现在 S 的**非增强**回声强度评估里（"For the quantitative analysis, the entire shape of the tumor was traced as the region of interest (ROI)"），S 的增强 TIC 用的是放在实性部分中央的 ROI。在增强 TIC 上用整个病灶做 ROI 的做法：**未找到**。
- 与约束 5 相关的反面信号：Y 第 13 页指出，单个 ROI 不一定代表整个病灶："nodules may have components of both LGD/IGD and HGD/invasive carcinoma... it is difficult to correctly identify the area of HGD/invasive carcinoma when we decide the location of the ROI"。同页还提到深度效应："the echo intensity is dependent on depth, and to take this into account, both the ROI of the tumor and the pancreatic parenchyma must be as near to the same depth as possible."
- 呼吸运动：Y 第 7 页 "The position of the ROI was manually corrected for the patient’s respiratory movements."；T 2.5 "Respiratory artifact effects were minimized using compensation techniques."

### 约束 6：洗入期被截断（开始录制晚于造影剂到达）论文怎么处理

- **直接相关的只有一处，做法是剔除：** S，Methods："three cases in which the contrast agent had already been injected in the initial frame or in which quantitative analysis was difficult due to poor respiratory control were excluded."
- S 对增强起点的定义："the point in time when the contrast agent flowed into the abdominal aorta or the superior mesenteric artery in the vicinity of the tumor was set as the start of contrast enhancement."
- 其他研究都在注射时就开始录，所以不存在截断：Y 第 7 页 "The digital CH-EUS data generated for 120 s continuously after the injection of contrast agents were stored"；T 2.3 "Timing commenced immediately after injection... CEUS data were recorded for 3–6 min and digitally stored on hard drives."；Fan 2015（`2015-fan...pdf` 第 2 页）"Real-time observation of the lesion should be no less than 120 s. Dynamic images were preserved for later analysis."
- 允许 t0 早于视频开头、再外推补回截断部分的做法：**未找到依据**。

## 3. 我没把握的地方

- **样本量很小，可能有漏检。** 我主要用 Europe PMC 的检索接口和几轮网页检索，共找到 4 篇有全文的胰腺囊性 TIC 原始研究。中文期刊（中华超声影像学杂志等）可能有用 SonoLiver 做囊性病变 TIC 的研究，但检索只能拿到摘要或什么都拿不到，没有覆盖到。
- **Yamamoto 2016 读的是机构库作者稿，不是出版社正式版**，页码是作者稿 PDF 的页码。
- **S、T 两篇读的是 Europe PMC 的 XML 全文**，出版社 PDF 被反爬拦下，所以没有放进 `reference-pdf/`，引文只能标到章节，标不到页码。Harima 2015 是 WebFetch 小模型转述的结果，我没亲眼读原文。
- **SonoLiver 等商业软件内部是否做了拟合或线性化，文章里都没写。** T 用 SonoLiver 得到 14 个参数，软件内部可能用了某种参数模型（厂商资料是否写明我没核），但论文没写，不能当作「论文里用 Lognormal」的依据。约束 1 的 Lognormal 先例应交给负责通用 CEUS 定量或肝脏、软件文献的检索组去核。
- **dB 单位只说明是对数域强度**，看不出是机器内部的原始回波数据还是显示灰度。4 篇都没有写「线性化」，不能据此推断它们用的是显示灰度。
- 关于约束 4 的门槛数值（R² > 0.75、0.8 之类），本组文献完全没有，只能从其他器官或方法学文献里找。