第一部分 超声造影在胰腺癌与胰腺炎鉴别诊断中的应用研究

背景及目的：

胰腺导管腺癌（pancreatic ductal adenocarcinoma，PDAC）与肿块型胰腺炎（mass-forming pancreatitis，MFP）在临床及影像表现上高度重叠，鉴别诊断存在挑战性。尽管超声造影（contrast-enhanced ultrasound，CEUS）在胰腺疾病诊断中展现出潜力，但现有研究仍存在以下问题：首先，多数研究MFP样本量相对较小；其次，既往认为PDAC在CEUS动脉期呈低增强，但我们在临床中发现部分PDAC病例亦可表现为等增强，此部分病例与MFP（多表现为等增强）的鉴别更加困难，目前极少有研究关注此类病例；此外，肿块型I型自身免疫性胰腺炎（autoimmune pancreatitis，AIP）是MFP的一种特殊亚型，以血清IgG4升高为特征，但约10%的PDAC患者也存在血清IgG4升高，为临床鉴别难点，CEUS是否具有鉴别价值尚有待明确。

本研究拟探讨CEUS在鉴别PDAC与MFP中的价值，具体目标包括：

（1）纳入MFP及PDAC病例，建立基于超声特征的鉴别诊断列线图模型，并评估其效能；

（2）开发基于CEUS时间-强度曲线（time-intensity curve，TIC）定量参数的机器学习模型，评估其在鉴别等增强PDAC与MFP中的价值；

（3）探讨CEUS对肿块型I型AIP与血清IgG4升高的PDAC的鉴别价值。

方法：

回顾性纳入2017年9月至2024年4月于北京协和医院接受胰腺CEUS检查并最终诊断为PDAC及MFP的患者。

（1）列线图模型构建：将PDAC与MFP病例按7:3随机分为训练集和验证集，通过Logistic回归分析筛选鉴别特征，在此基础上构建列线图模型并进行验证；

（2）机器学习模型开发：基于等增强型PDAC与MFP病例，按照检查时间分为两个数据集，第一数据集用于建立模型，第二数据集用于测试模型。采用t检验筛选TIC定量特征，在此基础上构建并测试机器学习模型；

（3）AIP与血清IgG4升高的PDAC鉴别诊断：比较肿块型I型AIP与血清IgG4升高的PDAC患者的超声特征差异，并通过构建受试者工作特征曲线（receiver operator characteristic curve，ROC）分析CEUS的鉴别诊断效能       。

结果：

（1）列线图模型纳入281例患者（188例PDAC，93例MFP）。训练集分析显示纵横比（P=0.002，OR=0.12）、钙化（P=0.003，OR=13.76）及廓清模式（P=0.002，OR=0.13）为独立预测因子。基于上述特征建立的列线图模型在训练集和验证集的曲线下面积（area under the curve，AUC）分别为0.930（95%CI：0.895-0.965）和0.914（95%CI：0.853-0.976）。校准曲线图及Hosmer-Lemeshow检验（P>0.05）证实模型校准度良好，决策曲线分析图显示模型临床获益显著；

（2）机器学习模型纳入152例患者（85例等增强PDAC，67例MFP），第一数据集使用五个特征组合的分类准确率、精确度、F1分数、灵敏度和特异度分别为0.827、0.828、0.834、0.841和0.814。第二数据集的分类准确率为0.833；

（3）171例患者纳入AIP与PDAC鉴别分析（29例AIP，142例PDAC，其中17例IgG4升高），CEUS联合CA19-9诊断血清IgG4升高的PDAC的准确率高达95.7%，灵敏度为100%。

结论：

CEUS征象对PDAC与MFP的鉴别准确性较高，具有重要临床应用价值：

（1）基于纵横比特征、钙化及廓清模式的列线图模型可作为术前无创鉴别PDAC与MFP的工具；

（2）基于CEUS-TIC定量特征构建的机器学习模型可以有效鉴别等增强型PDAC与MFP；

（3）CEUS特征联合CA19-9可有效区分肿块型I型AIP与血清IgG4升高的PDAC。

第二部分 VEGFR-2靶向超声造影在胰腺癌分子诊疗中的应用探索

背景及目的：

胰腺癌的高度异质性可能是其化疗及靶向治疗疗效有限的重要原因，通过对胰腺癌患者进一步分型筛选出对靶向药物敏感的患者群将有望提升疗效。因此，本研究旨在利用血管内皮生长因子受体2（vascular endothelial growth factor receptor 2，VEGFR-2）靶向CEUS评估小鼠胰腺癌模型的血供异质性，并探讨不同CEUS表现的胰腺癌对抗血管生成联合化疗敏感性的差异。

方法：

（1）构建小鼠胰腺癌模型，在模型中对比VEGFR-2靶向CEUS与非靶向CEUS的成像特征，分析CEUS-TIC参数与微血管密度（microvessel density，MVD）及VEGFR-2表达的相关性；

（2）根据VEGFR-2靶向CEUS成像特征将肿瘤分为富血供型及乏血供型，进一步再分为治疗组及对照组。具体分组如下：A1组（富血供型＋抗血管生成联合化疗）；A2组（富血供型＋空白治疗）；B1组（乏血供型＋抗血管生成联合化疗）；B2组（乏血供型＋空白治疗）。比较各组肿瘤的生长状况、治疗前后CEUS特征的差异及治疗后MVD、VEGFR-2表达的差异。

结果：

（1）VEGFR-2靶向CEUS的峰值强度（peak intensity，PI）、TIC曲线下面积（area under the time-intensity curve，AUTC）及平均渡越时间（mean transit time，MTT）显著优于非靶向CEUS。两种CEUS的PI及AUTC均与MVD及VEGFR-2表达呈正相关，其中靶向CEUS的AUTC与VEGFR-2表达强相关；

（2）治疗后，A1组PI值显著下降，A2组PI及AUTC值显著下降；A1组肿瘤生长最缓慢（体积及重量最小），B2组生长最快（体积及重量最大）；A1组MVD值高于A2组，VEGFR-2表达低于A2组，B1组VEGFR-2表达低于B2组。

结论：

（1）VEGFR-2靶向CEUS可精准评估胰腺癌的血供异质性，并可无创评估肿瘤内MVD及VEGFR-2表达水平；

（2）CEUS表现为富血供型的胰腺癌对抗血管生成联合化疗的敏感性显著高于乏血供型，本研究成果有望成为临床个体化治疗的影像学依据。

Part I Application of contrast-enhanced ultrasound in the differential diagnosis of pancreatic cancer and pancreatitis

Background and Objectives:

Pancreatic ductal adenocarcinoma (PDAC) and mass-forming pancreatitis (MFP) exhibit significant overlap in clinical and imaging presentations, posing challenges for differential diagnosis. Although contrast-enhanced ultrasound (CEUS) has shown promise in diagnosing pancreatic diseases, existing studies face several limitations: Firstly, most studies have relatively small sample sizes for MFP; Secondly, while PDAC is traditionally believed to display hypo-enhancement in the arterial phase on CEUS, we have clinically observed that some PDAC cases can also exhibit iso-enhancement, making differentiation from MFP (which often shows iso-enhancement) more difficult, yet few studies have focused on these cases; Additionally, type I autoimmune pancreatitis (AIP), a special type of MFP characterized by elevated serum IgG4 levels, presents a clinical challenge since approximately 10% of PDAC patients also have elevated IgG4 levels, and the diagnostic value of CEUS is unclear.

This study aims to investigate the value of CEUS in differentiating PDAC from MFP, with specific objectives including:

(1) Establishing a Diagnostic Model: Enrolling MFP and PDAC cases to develop a diagnostic nomogram based on ultrasound features and evaluating its efficacy;

(2) Machine Learning Model Development: Creating a machine learning model using quantitative time-intensity curve (TIC) parameters from CEUS to assess its value in distinguishing iso-enhancing PDAC from MFP;

(3) Differentiating AIP and PDAC with Elevated IgG4: Exploring the diagnostic value of CEUS in distinguishing type I AIP from PDAC with elevated serum IgG4 levels.

Methods:

This study retrospectively enrolled patients diagnosed with PDAC and MFP who underwent CEUS at Peking Union Medical College Hospital between September 2017 and April 2024.

(1) Nomogram model construction: PDAC and MFP cases were randomly divided into a training set and a validation set in a 7:3 ratio. Logistic regression analysis was used to select discriminative features, which were then used to construct and validate a nomogram model;

(2) Machine learning model development: For cases of iso-enhancing PDAC and MFP, the dataset was divided into two subsets based on examination time. The first subset was used for model development, and the second subset was used for testing. T-tests were employed to select TIC features, which were then used to build and test a machine learning model;

(3) Differential diagnosis of AIP and PDAC with elevated serum IgG4: Difference in ultrasound characteristics between patients with mass type I AIP and PDAC with elevated serum IgG4 was compared, and the differential diagnostic efficacy of CEUS was analyzed by constructing a receiver operator characteristic curve (ROC).

Results:

(1) 281 patients (188 PDAC, 93 MFP) were included in the nomogram model. Training set analysis showed aspect ratio (P=0.002, OR=0.12), calcification (P=0.003, OR=13.76) and washout pattern (P=0.002, OR=0.13) as independent predictors. The area under the curve (AUC) of the nomogram model based on the above features was 0.930 (95% CI: 0.895-0.965) and 0.914 (95% CI: 0.853-0.976) in the training and validation sets, respectively. Calibration curve plots and the Hosmer-Lemeshow test (p>0.05) confirmed that the model has good calibration, and decision curve analysis plots showed significant clinical benefit of the model;

(2) 152 patients (85 iso-enhanced PDAC, 67 MFP) were included in the machine-learning model, and the classification accuracy, precision, F1 score, sensitivity, and specificity for the first dataset using a combination of the five features were 0.827, 0.828, 0.834, 0.841, and 0.814, respectively. The accuracy for the second dataset was 0.833;

(3) 171 patients were included in the differential analysis of AIP and PDAC (29 AIP, 142 PDAC, including 17 with elevated IgG4). CEUS combined with CA19-9 diagnosis of PDAC with elevated serum IgG4 has an accuracy rate of up to 95.7% and a sensitivity of 100%.

Conclusion:

CEUS features demonstrate high accuracy in differentiating PDAC from MFP, offering significant clinical utility:

(1) The nomogram model based on aspect ratio, calcification and washout patterns can serve as a non-invasive tool for preoperative differentiation between PDAC and MFP;

(2) The machine learning model based on the quantitative features of CEUS-TIC can effectively differentiate iso-enhanced PDAC from MFP;

(3) The combination of CEUS features with CA19-9 allows for effective differentiation between mass type I AIP from PDAC with elevated serum IgG4.

Part II Exploration of VEGFR-2-targeted contrast-enhanced ultrasound in the molecular diagnosis and treatment of pancreatic cancer

Background and Objectives:

The high heterogeneity of pancreatic cancer is a significant factor contributing to the limited efficacy of chemotherapy and targeted therapy. Further stratification of pancreatic cancer patients to identify those sensitive to targeted drugs could potentially enhance treatment outcomes. This study aims to utilize vascular endothelial growth factor receptor 2 (VEGFR-2) targeted CEUS to assess vascular heterogeneity in a mouse model of pancreatic cancer. Additionally, it explores the differences in sensitivity to anti-angiogenic therapy combined with chemotherapy among pancreatic cancers with varying CEUS presentations.

Methods:

(1) A mouse model of pancreatic cancer was established to compare the imaging characteristics of VEGFR-2 targeted CEUS with non-targeted CEUS. The study analyzed the correlation between CEUS-TIC parameters and microvessel density (MVD) as well as VEGFR-2 expression;

(2) Tumors were classified into hypervascular and hypovascular types based on VEGFR-2 targeted CEUS imaging features. These were further divided into treatment and control groups: A1 Group: Hypervascular tumors treated with anti-angiogenic therapy combined with chemotherapy; A2 Group: Hypervascular tumors receiving blank treatment; B1 Group: Hypovascular tumors treated with anti-angiogenic therapy combined with chemotherapy; B2 Group: Hypovascular tumors receiving blank treatment. The study compared tumor growth, differences in CEUS features before and after treatment, and post-treatment differences in MVD and VEGFR-2 expression among these groups.

Results:

(1) The peak intensity (PI), area under the time-intensity curve (AUTC) and mean transit time (MTT) of VEGFR-2-targeted CEUS were significantly superior to those of non-targeted CEUS. Both PI and AUTC from both CEUS methods showed positive correlations with MVD and VEGFR-2 expression. Notably, the AUTC of targeted CEUS was strongly correlated with VEGFR-2 expression;

(2) Following treatment, the PI value significantly decreased in the A1 group, while both PI and AUTC values decreased in the A2 group. Tumor growth was slowest in the A1 group, with the smallest volume and weight, whereas the B2 group exhibited the fastest growth, with the largest volume and weight. The A1 group had higher MVD values than the A2 group but lower VEGFR-2 expression. Additionally, VEGFR-2 expression was lower in the B1 group compared to the B2 group.

Conclusion:

(1) VEGFR-2-targeted CEUS can accurately assess the vascular heterogeneity of pancreatic cancer and non-invasively evaluate MVD and VEGFR-2 expression levels within tumors;

(2) Pancreatic cancers with hypervascular CEUS features exhibit significantly higher sensitivity to anti-angiogenic therapy combined with chemotherapy compared to those with hypovascular features. These findings have the potential to serve as an imaging basis for personalized clinical treatment strategies.