分享:
分享到微信朋友圈
X
临床研究
多参数MRI影像组学在术前预测胰腺导管腺癌淋巴结转移中的价值
曾飘娥 曲超 崔景景 修典荣 刘剑羽 袁慧书

Cite this article as: ZENG P E, QU C, CUI J J, et al. The value of multi-parametric MRI radiomics model in predicting lymph node metastasis of pancreatic ductal adenocarcinoma[J]. Chin J Magn Reson Imaging, 2024, 15(3): 114-121.本文引用格式曾飘娥, 曲超, 崔景景, 等. 多参数MRI影像组学在术前预测胰腺导管腺癌淋巴结转移中的价值[J]. 磁共振成像, 2024, 15(3): 114-121. DOI:10.12015/issn.1674-8034.2024.03.019.


[摘要] 目的 联合常规MRI及表观扩散系数(apparent diffusion coefficient, ADC)图的影像组学特征构建多参数MRI影像组学模型术前预测胰腺导管腺癌(pancreatic ductal adenocarcinoma, PDAC)淋巴结转移(lymph node metastasis, LNM),并与建立的常规MRI影像组学模型和临床模型比较预测效能,探索基于ADC图影像组学的附加价值。材料与方法 218例PDAC按照7∶3的比例随机分为训练集和验证集。纳入临床及常规影像特征构建临床影像学模型。提取常规MRI图像(T1WI、T2WI、动脉期图像及门静脉期图像)及ADC图的影像组学特征。在训练集中采用最小绝对收缩和选择算子筛选出与LNM最相关的特征用于模型构建。构建基于常规MRI影像组学模型(影像组学模型1)和联合常规MRI和ADC图的影像组学模型(影像组学模型2)。使用受试者工作特征曲线下面积(area under the curve, AUC)评估模型预测效能。采用DeLong检验比较模型间的AUC值的差异是否有统计学意义。校准曲线评估模型的准确性。决策曲线分析评估模型的临床价值。结果 临床影像学模型、影像组学模型1、影像组学模型2术前预测LNM的AUC值在训练集和验证集中分别是0.741和0.674、0.818和0.702、0.854和0.839。影像组学模型2术前预测LNM的AUC值高于临床影像学模型(训练集P=0.009,验证集P=0.023)及影像组学模型1(训练集P=0.044,验证集P=0.041),差异均具有统计学意义。影像组学模型1的预测效能与临床影像学模型相比,差异不具有统计学意义(训练集P=0.095,验证集P=0.759)。三个模型的校准曲线均显示预测值与实际值具有较好的一致性。决策曲线显示影像组学模型2比影像组学模型1和临床影像学模型具有更高的净效益。结论 联合常规MRI及ADC图构建的多参数MRI影像组学模型具有术前预测PDAC患者LNM的潜能,且其效能优于常规MRI影像组学模型及临床影像学模型。
[Abstract] Objective A multi-parametric MRI radiomics model was constructed by combining the radiomics features of conventional MRI and apparent diffusion coefficient (ADC) map to predict lymph node metastasis (LNM) in pancreatic ductal adenocarcinoma (PDAC), and the prediction performance was compared with the established conventional MRI radiomics model and clinical model, to explore the added value of ADC map radiomics.Methods and Materials A total of 218 patients with PDAC were randomly divided into a training cohort and a validation cohort at a ratio of 7∶3. Clinical and conventional imaging features were used to construct the clinical imaging model. Then the radiomics features were extracted based on conventional MRI images (T1WI, T2WI, arterial phase images and portal venous phase images) and ADC maps. Least absolute shrinkage and selection operator was used to select the most relevant features of LNM in the training cohort for model construction. Radiomics models based on conventional MRI images (represented as radiomics model 1) and radiomics models combined with conventional MRI images and ADC maps (represented as radiomics model 2) were constructed. The area under the curve (AUC) of receiver operator characteristic was used to evaluate the prediction performance of the models. DeLong validation was used to compare the difference of AUC values between models. The calibration curve was used to evaluate the accuracy of the model. The clinical value of the model was evaluated by decision curve analysis.Results The AUC values of clinical and radiographic model, radiomics model 1 and radiomics model 2 for preoperative prediction of LNM in the training and validation cohorts were 0.741 and 0.674, 0.818 and 0.702, 0.854 and 0.839, respectively. The AUC value of radiomics model 2 for preoperative prediction of LNM was significantly higher than that of the clinical and radiographic model (P=0.009 in the training cohort; P=0.023 in the validation cohort) and radiomics model 1 (P=0.044 in the training cohort; P=0.041 in the validation cohort). The prediction performance of radiomics model 1 was not significantly different from that of the clinical radiographic model (P=0.095 in the training cohort; P=0.759 in the validation cohort). The calibration curves of the three models showed good agreement between the predicted values and the actual values. Decision curve analysis curve showed that radiomics model 2 had a higher net benefit than radiomics model 1 and clinical imaging model.Conclusions The multi-parametric MRI radiomics model by combining conventional MRI and ADC map radiomics can improve the diagnostic efficiency of predicting LNM of PDAC, which is significantly better than the conventional MRI radiomics model and clinical-radiographic model.
[关键词] 胰腺导管腺癌;淋巴结转移;影像组学;磁共振成像;表观扩散系数
[Keywords] pancreatic ductal adenocarcinoma;lymph node metastasis;radiomics;magnetic resonance imaging;apparent diffusion coefficient

曾飘娥 1   曲超 2   崔景景 3   修典荣 2   刘剑羽 1   袁慧书 1*  

1 北京大学第三医院放射科,北京 100191

2 北京大学第三医院普外科,北京 100191

3 联影智能(北京)有限公司研发部,北京 100094

通信作者:袁慧书,E-mail:huishuy@bjmu.edu.cn

作者贡献声明:袁慧书设计并实施本研究的方案,参与论文重要内容的修改,获得北京市自然科学基金项目资助;曾飘娥酝酿设计并实施研究,起草和撰写稿件,参与论文重要内容的修改,获取、分析和解释本研究的数据;曲超、崔晶晶、修典荣、刘剑羽获取、分析或解释本研究的数据,对论文重要内容进行了修改;全体作者都同意发表最后的修改稿,同意对本研究工作各方面的诚信问题负责。


基金项目: 北京市自然科学基金项目 7212126
收稿日期:2023-10-27
接受日期:2024-02-23
中图分类号:R445.2  R735.9 
文献标识码:A
DOI: 10.12015/issn.1674-8034.2024.03.019
本文引用格式曾飘娥, 曲超, 崔景景, 等. 多参数MRI影像组学在术前预测胰腺导管腺癌淋巴结转移中的价值[J]. 磁共振成像, 2024, 15(3): 114-121. DOI:10.12015/issn.1674-8034.2024.03.019.

0 引言

       胰腺导管腺癌(pancreatic ductal adenocarcinoma, PDAC)居我国总体恶性肿瘤发病率和死亡率的第9位和第6位[1]。PDAC预后极差,淋巴结转移(lymph node metastasis, LNM)是PDAC预后较差的重要影响因素[2]。AJCC更新的第八版TNM系统更加强调了LNM在预后评估中的意义。同时NCCN指南和中国胰腺癌诊疗指南(2022年版)建议对伴有区域LNM等高危因素的PDAC可以考虑新辅助治疗[3, 4]。因此,术前准确识别PDAC的淋巴结状态对患者的预后评估和治疗方案制订具有重要意义。目前基于术前影像形态学对PDAC淋巴结状态评估的能力较差。文献报道淋巴结大小不能作为术前LNM的可靠性指标[5, 6]。表观扩散系数(apparent diffusion coefficient, ADC)可以定量评估组织内水分子扩散程度,从而间接反映肿瘤的组织学成分和侵袭性。影像组学可以提取肉眼无法识别的定量的、高通量的影像特征,反映病变的生物学行为[7]。先前的研究表明,基于CT图像的影像组学在淋巴结状态评估及预后预测中有一定的价值。既往有研究探索MRI影像组学对PDAC淋巴结状态评估价值,但基于ADC图影像组学的附加价值尚不清楚,且无与临床及影像形态学预测模型的对比研究。本研究拟联合常规MRI及ADC图的影像组学特征构建多参数MRI影像组学模型术前评估PDAC淋巴结状态,并与建立的常规MRI影像组学模型和临床模型比较,探索基于ADC图影像组学的附加价值。

1 材料与方法

1.1 研究对象

       本研究遵守《赫尔辛基宣言》,已通过北京大学第三医院伦理委员会批准,免除受试者知情同意,批准文号:IRB00006761-LM2021120。

       本研究为单中心、回顾性临床研究,连续性收集2012年1月至2021年6月在我院进行手术并病理确诊为PDAC患者的临床及影像资料。入组标准:(1)接受了手术切除和区域淋巴结清扫术;(2)术前2周内的MRI图像完整。排除标准:(1)因局部晚期胰腺癌行姑息性手术或R2切除;(2)术前接受放疗或化疗;(3)图像伪影严重。按照LNM以病理报告将PDAC分为伴有LNM和不伴LNM。按照7∶3的比例随机分为训练集与验证集。

1.2 MRI扫描

       MRI影像数据包含常规MRI及扩散加权成像(diffusion-weighted imaging, DWI)。MRI扫描采用3 T MRI扫描仪(Discovery MR750, GE Healthcare, USA),8通道体部相控阵线圈。扫描序列及参数:(1)轴位脂肪抑制T2WI序列,重复时间(repetition time, TR)/回波时间(echo time, TE)15 000 ms/81 ms,层厚6.0 mm,层间距1 mm,视野32 cm×32 cm,矩阵320×320;(2)轴位肝脏快速容积成像T1WI序列,TR/TE 4 ms/2.4 ms,层厚5.0 mm,视野36 cm×36 cm,矩阵264×256;(3)轴位T1WI动态增强MRI,高压注射器将钆对比剂(Beilu Pharmaceutical, Beijing, China)以2 mL/s的速率及0.2 mmol/kg的剂量团注到前臂静脉,注射对比剂后33~38 s扫描动脉期,70~75 s扫描门脉期图像;(4)用轴位、自由呼吸、单次发射回波平面成像进行DWI序列扫描,TR/TE 7 500 ms/62 ms;层厚6.0 mm;层间距1.0 mm;带宽250 kHz;视野35 cm×35 cm;矩阵160×160,8个b值(0、25、50、100、200、500、800、1 000 s/mm2)。

1.3 临床及影像学特征的采集

       临床特征:采集纳入患者的性别、年龄、是否黄疸、术前糖类抗原19-9(carbohydrate antigen19-9, CA19-9)、术前血清总胆红素值等临床特征。

       原发肿瘤影像学特征包括(1)肿瘤大小;(2)肿瘤部位;(3)胰管扩张;(4)胰体尾萎缩;(5)强化程度(对比正常胰腺实质);(6)边缘强化:定义为不规则环状强化,在动态增强图像上具有相对低血供的中心区域;(7)胰腺周围肿瘤浸润;(8)动脉侵犯;(9)静脉侵犯。

       淋巴结评估:测量最大淋巴结的短径和长径。

       ADC值测量:以常规和动态增强MRI扫描为参考,在DWI图像(b=1 000 s/mm2)上原发肿瘤最大横截面积手动勾画感兴趣区(region of interest, ROI),最大限度地排除囊变坏死区域。由两名分别具有7年和5年腹盆影像诊断经验的放射科医师不知晓病理淋巴结状态结果的情况分别独立测量,以平均值作为最终ADC值。

1.4 影像组学分析

       影像组学分析流程包括:ROI勾画;影像组学特征提取;影像组学特征选择;模型构建和评价。本研究中ROI勾画和特征提取在Research Portal V1.1平台(上海联影智能有限公司)完成。

       ROI勾画:沿着原发肿瘤边界逐层勾画,尽量避开明显囊变区域。最初由一名5年诊断经验的放射科住院医师勾画所有图像ROI。一个月后,随机选择其中30例患者图像,由另一位7年诊断经验的放射科主治医师重新绘制ROI,计算组间相关系数(inter-class correlation coefficient, ICC),以评估从两名放射科医生绘制的ROI中提取影像组学特征的可重复性。两名医师除了知晓勾画对象是PDAC外,对LNM状态不知晓。

       影像组学特征提取:图像预处理步骤包括像素归一化和重采样。每个ROI中提取出1 094个特征,包括从原始图像中提取的104个特征(基于形态和大小特征、一阶特征和纹理特征)和通过小波滤波器及高斯-拉普拉斯滤波器转换得到的990个特征。本研究共提取了来自T1WI、T2WI、动脉期、门脉期及ADC图的共5 470(1 094×5)个影像组学特征。

       影像组学特征选择:保留其中ICC≥0.75的影像组学特征。采用最小绝对收缩和选择算子(least absolute shrinkage and selection operator, LASSO)回归选出最相关的特征用于下一步的模型构建。根据LASSO回归筛出的影像组学特征及其对应的相关系数相乘之总和得到Radscore。

       模型构建:分别建立基于常规MRI图像(T1WI、T2WI、动脉期及门脉期图像)影像组学模型(影像组学模型1)和联合常规MRI及ADC图的影像组学模型(影像组学模型2)。在训练集中,通过单因素和多因素向前逐步逻辑回归分析筛选出与LNM相关的临床及影像学特征,建立LNM的术前诊断临床影像学(clinical and radiographic, CR)模型。

       模型效能评价:计算训练集和验证集中模型预测LNM的受试者工作特征曲线下面积(area under the curve, AUC)值。以最大约登指数为最佳诊断截断值,计算其对应的敏感度、特异度、阳性预测值、阴性预测值和准确率等评价指标。使用DeLong检验比较各预测模型的诊断效能。采用校准曲线和Hosmer-Lemeshow检验评估模型的校准度。决策曲线分析(decision curve analysis, DCA)来评估模型的临床实用性。

1.5 统计分析

       统计分析软件包括SPSS 22.0(SPSS Inc, Chicago, USA)、MedCalc Version 15.0(MedCalc Software Ltd., Ostend, Belgium)和R软件包(version 3.5.2)。分类资料用频数(百分比)表示,其组间差异比较采用卡方检验或Fisher精确检验。采用Kolmogorov-Smimov检验连续性变量的正态性。正态分布数据用均值±标准差来表示,组间差异比较采用独立样本检验。非正态分布数据表示为中位数(四分位数间距),组间差异比较采用Mann-Whitaey U检验。AUC评估模型的诊断效能。P<0.05为差异具有统计学意义。

2 结果

2.1 一般资料

       本研究最终218例PDAC纳入,伴LNM 97例,不伴LNM 121例。按照7∶3的比例随机分为训练集(152例)和验证集(66例)。病例入组流程详见图1

图1  病例纳排流程图。
Fig. 1  A flowchart of patient inclusion.

2.2 模型构建

       在训练集中,临床及影像学特征单因素及多因素逻辑回归结果显示,术前CA19-9升高、肿瘤大小、胰体尾萎缩及ADC值是术前预测LNM的独立因素(表1),根据这4个因素构建出预测LNM的CR模型,并绘制了列线图(图2)。LASSO回归选择出7个与LNM最相关的影像组学特征(图3A),包括3个来自T2WI图像,3个来自动脉期图像和1个来自门脉期图像的组学特征构建影像组学模型1;选择出11个与LNM相关的影像组学特征,包括2个来自T2WI图像,3个来自动脉期图像,1个来自门脉期图像和5个来自ADC图的影像组学特征构建影像组学模型2(图3B)。

图2  临床影像学模型术前预测胰腺导管腺癌淋巴结转移列线图。
图3  LASSO回归分析结果。3A:基于T1WI、T2WI、动脉期和门脉期图像影像组学特征进行LASSO回归,筛选出与淋巴结转移最相关的影像组学特征,构建影像组学模型1;3B:基于T1WI、T2WI、动脉期、门脉期图像和ADC图的影像组学特征进行LASSO回归,筛选出与淋巴结转移最相关的影像组学特征,构建影像组学模型2。CA19-9:糖类抗原19-9;ADC:表观扩散系数;LASSO:最小绝对收缩和选择算子。
Fig. 2  Nomogram of clinical radiological model for preoperatively predicting lymph node metastasis in pancreatic ductal adenocarcinoma.
Fig. 3  Results of LASSO regression analysis. 3A: LASSO regression is used to select radiomics features from T1WI, T2WI, arterial phase and portal venous phase images to develop radiomics model 1; 3B: LASSO regression is performed on the radiomics features derived from T1WI, T2WI, arterial phase, portal venous phase images and ADC maps to select to develop radiomics model 2. CA19-9: carbohydrate antigen 19-9; ADC: apparent diffusion coefficient; LASSO: least absolute shrinkage and selection operator.
表1  预测胰腺导管腺癌淋巴结转移的临床及影像学特征单因素和多因素分析
Tab. 1  Univariate and multivariate analysis of clinical and radiological features of predicting lymph node metastasis in pancreatic ductal adenocarcinoma

2.3 模型效能评价

       影像组学模型2术前预测LNM的AUC值最高,高于CR模型(训练集:0.854 vs. 0.741,P=0.009;验证集:0.839 vs. 0.674,P=0.023)及影像组学模型1(训练集:0.854 vs. 0.818,P=0.044;验证集:0.839 vs. 0.702,P=0.041),差异具有统计学意义;影像组学模型1的预测效能高于CR模型(训练集:0.818 vs. 0.741,P=0.095;验证集:0.702 vs. 0.674,P=0.759),但差异不具有统计学意义(表2图4)。

       校准曲线分析显示单个模型术前预测LNM时预测概率的估计值和实际观察值之间均具有较好的一致性(图5)。三个模型预测LNM的DCA显示,训练集和验证集中的大部分阈值概率范围内,影像组学模型2比影像组学模型1和CR模型具有更高的净效益(图6)。图7显示一例PDAC患者,病理诊断伴有LNM,影像组学模型2评估正确,而影像组学模型1及CR模型评估错误。图8显示一例PDAC患者,病理不伴LNM,影像组学模型2、影像组学模型1以及CR模型均评估正确。

图4  不同模型术前预测胰腺导管腺癌淋巴结转移的受试者工作特征曲线。4A:训练集;4B:验证集。AUC:曲线下面积;CR model(CR模型):临床及影像学模型;radiomics model 1(影像组学模型1):基于常规MRI影像组学特征建立的模型;radiomics model 2(影像组学模型2):联合常规MRI和表观扩散系数图影像组学特征建立的模型。
图5  不同模型预测胰腺导管腺癌淋巴结转移的校准曲线图。ADC:表观扩散系数。
Fig. 4  Receiver operating characteristic curves of different models for preoperatively predicting lymph node metastasis in pancreatic ductal adenocarcinoma. 4A: training set; 4B: validation set. AUC: area under the curve; CR model: clinical and radiological model; radiomics model 1: the model based on conventional MRI radiomics features; radiomics model 2: combined radiomics features of conventional MRI and apparent diffusion coefficientmap.
Fig. 5  Calibration curves of different models for predicting lymph node metastasis in pancreatic ductal adenocarcinoma. ADC: apparent diffusion coefficient.
图6  不同模型术前预测胰腺导管腺癌淋巴结转移的决策曲线图。6A:训练集;6B:验证集。CR model(CR模型):临床及影像学模型;radiomics model 1(影像组学模型1):基于常规MRI影像组学特征建立的模型;radiomics model 2(影像组学模型2):联合常规MRI和表观扩散系数图影像组学特征建立的模型。
Fig. 6  Decision curves of different models for preoperatively predicting lymph node metastasis in pancreatic ductal adenocarcinoma. 6A: training set; 6B: validation set. CR model: clinical and radiological model; radiomics model 1: the model based on conventional MRI radiomics features; radiomics model 2: combined radiomics features of conventional MRI and apparent diffusion coefficientmap.
图7  男,66岁,胰颈后部见稍长T2信号(7A,箭),动脉期低强化(7B,箭),扩散加权成像图像呈稍高信号(7C,箭),表观扩散系数图像呈稍低信号(7D,箭),肿瘤大小2.4 cm,病理诊断为胰腺导管腺癌,伴有淋巴结转移。联合常规MRI和表观扩散系数图影像组学特征建立的影像组学模型2预测淋巴结转移阳性,基于常规MRI影像组学特征建立的影像组学模型1和临床影像学模型预测淋巴结阴性。
Fig. 7  A 66-year-old male patient presented with slightly hyperintense T2 signal in the posterior part of the pancreatic neck (7A, arrow), hypoenhancement in the arterial phase (7B, arrow), slightly hyperintensity on diffusion-weighted imaging (7C, arrow), and slightly hypointensity on apparent diffusion coefficient images (7D, arrow). The tumor size was 2.4 cm. The tumor was pathologically diagnosed as pancreatic ductal adenocarcinoma with lymph node metastasis. Radiomics model 2 based on conventional MRI radiomics features and apparent diffusion coefficient map radiomics features predicted positive lymph node metastasis, and radiomics model 1 based on conventional MRI radiomics features and clinical and radiological model predicted negative lymph nodes.
图8  男,47岁,胰头钩突见稍长T2信号(8A,箭),动脉期低强化(8B,箭),扩散加权成像图像呈高信号(8C,箭),表观扩散系数图像呈低信号(8D,箭),肿瘤大小2.2 cm,病理诊断胰腺导管腺癌,不伴有淋巴结转移。联合常规MRI和表观扩散系数图影像组学特征建立的影像组学模型2、基于常规MRI影像组学特征的影像组学模型1和临床影像学模型均预测为淋巴结阴性。
Fig. 8  A 47-year-old male patient presented with slightly hyperintense T2 signal in the uncinate process of pancreatic head (8A, arrow), hypoenhancement in the arterial phase (8B, arrow), hyperintensity on diffusion-weighted imaging (8C, arrow), and hypointensity on apparent diffusion coefficient images (8D, arrow). The tumor size was 2.2 cm. The tumor was pathologically diagnosed as pancreatic ductal adenocarcinoma without lymph node metastasis. Radiomics model 2 based on conventional MRI radiomics features and apparent diffusion coefficient map radiomics features, radiomics model 1 based on conventional MRI radiomics features, and clinical and radiological model predicted negative lymph nodes.
表2  不同模型术前预测胰腺导管腺癌淋巴结转移的效能
Tab. 2  The efficacy of different models in preoperative prediction of lymph node metastasis in pancreatic ductal adenocarcinoma

3 讨论

       本研究发现联合常规MRI及ADC图影像组学特征构建的多参数MRI影像组学模型具有术前预测PDAC患者LNM的潜能,且其预测效能优于基于常规MRI的影像组学模型和临床模型,可作为术前评估PDAC患者淋巴结状态的方法。

3.1 临床及影像学特征术前预测PDAC LNM

       本研究中,肿瘤大小、术前CA19-9升高、胰体尾萎缩及ADC值是术前预测LNM独立的预测因素。LNM与原发肿瘤大小有关,原发肿瘤增大是LNM的危险因素,这与报道的文献结果一致[8]。术前CA19-9对LNM预测有一定价值,但目前文献报道CA19-9预测LNM的截断值并不统一[9, 10]

       LEE等[11]测量胰腺癌区域淋巴结的ADC值,发现转移性淋巴结的ADC值明显低于良性淋巴结。转移性淋巴结的肿瘤细胞数量和核质比的增加以及恶性肿瘤细胞外间隙的减少限制了水的扩散,导致ADC减低。但是在实际临床工作中,测量淋巴结ADC值来诊断LNM难以操作。因为通常区域内有多发淋巴结,难以确定以测量哪个淋巴结为标准。其次转移淋巴结和正常淋巴结均较小,容易导致ADC值测量可重复性差。LNM已被证明与多种情况下原发肿瘤的组织学特征和影像学表现有关[12]。本研究是通过测量PDAC原发肿瘤的ADC值来预测LNM,发现伴有LNM患者原发肿瘤ADC值显著性低于不伴有LNM患者,与SHI等[13]研究一致。

3.2 影像组学术前预测PDAC LNM

       目前一些研究表明,基于CT的影像组学在预测PDAC的LNM方面有价值[14, 15, 16, 17]。与CT相比,MRI无电离辐射暴露,提供了更好的软组织分辨率,在诊断早期PDAC方面优于CT[18]。本研究勾画了整个肿瘤并从和多序列、多期MRI图像中提取影像组学特征,综合反映肿瘤异质性[19]。本研究提取常规MRI图像的影像组学特征来构建影像组学模型,发现术前预测LNM具有良好的准确性,训练集诊断准确度为76.3%,验证集诊断准确度为71.2%,这与既往研究[20, 21, 22, 23]的研究结果一致。但SHI等[20]研究是将≤3枚阳性淋巴结定义为阴性LNM,>3枚阳性淋巴结定义为阳性LNM。因此,他们建立的模型实际上是区分PDAC的N0/N1和N2期LNM。本研究发现基于常规MRI影像组学模型有助于无创地术前预测PDAC淋巴结状态,诊断效能高于本研究建立的CR模型,但差异无统计学意义。

       本研究提取ADC图的影像组学特征,发现基于联合常规MRI和ADC功能图构建的影像组学模型对LNM的预测能力最高,显著性高于基于常规MRI图像构建的影像组学模型和CR模型,这进一步说明了ADC对预测LNM的重要价值。基于ADC图的影像组学模型对LNM的预测能力已在宫颈癌、直肠癌、子宫内膜癌及前列腺癌等肿瘤中得到验证[24, 25, 26, 27, 28, 29, 30]。WU等[26]在研究影像组学评价宫颈鳞状细胞癌病理特征时发现,功能图对宫颈鳞状细胞癌的病理特征鉴别价值优于解剖图,ADC图对LNM的预测效能最好,与本研究一致。既往有研究构建基于DWI图像影像组学特征来预测LNM[31, 32]。本研究是提取ADC图的影像组学特征,而不是DWI图像。这是考虑到DWI的信号强度变化取决于组织的ADC状态和运动敏感梯度。b值的选择会显著影响DWI图像的纹理分析[24]。ADC目前已广泛运用于临床,ADC影像组学特征有望将来在协助LNM评估中发挥作用。

3.3 局限性及展望

       首先,本研究影像组学模型是单中心回顾性研究,仅进行了内部验证,未来可以进一步扩大样本量,纳入多中心的数据,进行外部验证以及前瞻性验证,增加模型的泛化性;其次,仅研究了ADC图的影像组学特征,未来可以进一步探究体素内不相干运动、扩散峰度成像等功能成像的影像组学特征预测PDAC患者LNM的价值;最后,本研究由于样本量较少,未进一步探索N1、N2期分类以及N0、N1、N2三分类模型;另外,目前胰腺影像组学研究病灶的勾画多是基于人工手动勾画,费时费力,未来准确的半自动或全自动分割方式使影像组学操作尽可能简化,增加影像组学在临床应用的可能性。

4 结论

       综上所述,联合常规MRI及ADC图建立的多参数MRI影像组学模型具有术前预测LNM的潜能,且效能显著性优于常规MRI影像组学模型及CR模型,为影像组学在胰腺癌的个性化综合治疗提供参考。

[1]
CHEN W, ZHENG R, BAADE P D, et al. Cancer statistics in China, 2015[J]. CA Cancer J Clin, 2016, 66(2): 115-132. DOI: 10.3322/caac.21338.
[2]
ZHENG Z J, WANG M J, TAN C L, et al. Prognostic impact of lymph node status in patients after total pancreatectomy for pancreatic ductal adenocarcinoma: a strobe-compliant study[J/OL]. Medicine, 2020, 99(8): e19327 [2024-02-05]. https://pubmed.ncbi.nlm.nih.gov/32080152/. DOI: 10.1097/MD.0000000000019327.
[3]
TEMPERO M A, MALAFA M P, AL-HAWARY M, et al. Pancreatic adenocarcinoma, version 2.2021, NCCN clinical practice guidelines in oncology[J]. J Natl Compr Canc Netw, 2021, 19(4): 439-457. DOI: 10.6004/jnccn.2021.0017.
[4]
国家卫生健康委办公厅. 胰腺癌诊疗指南(2022年版)[J]. 临床肝胆病杂志, 2022, 38(5): 1006-1015. DOI: 10.3969/j.issn.1001-5256.2022.05.007.
General Office of National Health Commission. Standard for diagnosis and treatment of pancreatic cancer (2022 edition)[J]. J Clin Hepatol, 2022, 38(5): 1006-1015. DOI: 10.3969/j.issn.1001-5256.2022.05.007.
[5]
SHIN J, SHIN S, LEE J H, et al. Lymph node size and its association with nodal metastasis in ductal adenocarcinoma of the pancreas[J]. J Pathol Transl Med, 2020, 54(5): 387-395. DOI: 10.4132/jptm.2020.06.23.
[6]
LE O, JAVADI S, BHOSALE P R, et al. CT features predictive of nodal positivity at surgery in pancreatic cancer patients following neoadjuvant therapy in the setting of dual energy CT[J]. Abdom Radiol, 2021, 46(6): 2620-2627. DOI: 10.1007/s00261-020-02917-5.
[7]
LAMBIN P, LEIJENAAR R T H, DEIST T M, et al. Radiomics: the bridge between medical imaging and personalized medicine[J]. Nat Rev Clin Oncol, 2017, 14(12): 749-762. DOI: 10.1038/nrclinonc.2017.141.
[8]
PU N, CHEN Q D, GAN W, et al. Lymph node metastatic patterns and survival predictors based on tumor size in pancreatic ductal adenocarcinoma[J]. Adv Ther, 2021, 38(8): 4258-4270. DOI: 10.1007/s12325-021-01819-2.
[9]
WANG S Y, SHI H C, YANG F X, et al. The value of 18F-FDG PET/CT and carbohydrate antigen 19-9 in predicting lymph node micrometastases of pancreatic cancer[J]. Abdom Radiol, 2019, 44(12): 4057-4062. DOI: 10.1007/s00261-019-02248-0.
[10]
HUA J, CHEN X M, CHEN Y J, et al. Development and multicenter validation of a nomogram for preoperative prediction of lymph node positivity in pancreatic cancer (NeoPangram)[J]. Hepatobiliary Pancreat Dis Int, 2021, 20(2): 163-172. DOI: 10.1016/j.hbpd.2020.12.020.
[11]
LEE J H, HAN S S, HONG E K, et al. Predicting lymph node metastasis in pancreatobiliary cancer with magnetic resonance imaging: a prospective analysis[J]. Eur J Radiol, 2019, 116: 1-7. DOI: 10.1016/j.ejrad.2019.04.007.
[12]
WANG F, WANG Y X, ZHOU Y, et al. Apparent diffusion coefficient histogram analysis for assessing tumor staging and detection of lymph node metastasis in epithelial ovarian cancer: correlation with p53 and ki-67 expression[J]. Mol Imaging Biol, 2019, 21(4): 731-739. DOI: 10.1007/s11307-018-1295-7.
[13]
SHI Y J, LIU B N, LI X T, et al. Establishment of a multi-parameters MRI model for predicting small lymph nodes metastases (<10 mm) in patients with resected pancreatic ductal adenocarcinoma[J]. Abdom Radiol, 2022, 47(9): 3217-3228. DOI: 10.1007/s00261-021-03347-7.
[14]
BIAN Y, GUO S W, JIANG H, et al. Radiomics nomogram for the preoperative prediction of lymph node metastasis in pancreatic ductal adenocarcinoma[J/OL]. Cancer Imaging, 2022, 22(1): 4 [2024-02-05]. https://pubmed.ncbi.nlm.nih.gov/34991733/. DOI: 10.1186/s40644-021-00443-1.
[15]
GAO J H, HAN F, JIN Y Y, et al. A radiomics nomogram for the preoperative prediction of lymph node metastasis in pancreatic ductal adenocarcinoma[J/OL]. Front Oncol, 2020, 10: 1654 [2024-02-05]. https://pubmed.ncbi.nlm.nih.gov/32974205/. DOI: 10.3389/fonc.2020.01654.
[16]
LI K, YAO Q D, XIAO J J, et al. Contrast-enhanced CT radiomics for predicting lymph node metastasis in pancreatic ductal adenocarcinoma: a pilot study[J/OL]. Cancer Imaging, 2020, 20(1): 12 [2024-02-05]. https://pubmed.ncbi.nlm.nih.gov/32000852/. DOI: 10.1186/s40644-020-0288-3.
[17]
LIU P, GU Q B, HU X L, et al. Applying a radiomics-based strategy to preoperatively predict lymph node metastasis in the resectable pancreatic ductal adenocarcinoma[J]. J Xray Sci Technol, 2020, 28(6): 1113-1121. DOI: 10.3233/XST-200730.
[18]
JANG K M, KIM S H, KIM Y K, et al. Missed pancreatic ductal adenocarcinoma: assessment of early imaging findings on prediagnostic magnetic resonance imaging[J]. Eur J Radiol, 2015, 84(8): 1473-1479. DOI: 10.1016/j.ejrad.2015.05.012.
[19]
YANG L F, YANG J B, ZHOU X B, et al. Development of a radiomics nomogram based on the 2D and 3D CT features to predict the survival of non-small cell lung cancer patients[J]. Eur Radiol, 2019, 29(5): 2196-2206. DOI: 10.1007/s00330-018-5770-y.
[20]
SHI Z S, MA C L, HUANG X M, et al. Magnetic resonance imaging radiomics-based nomogram from primary tumor for pretreatment prediction of peripancreatic lymph node metastasis in pancreatic ductal adenocarcinoma: A multicenter study[J]. J Magn Reson Imaging, 2022, 55(3): 823-839. DOI: 10.1002/jmri.28048.
[21]
SHI L, WANG L, WU C Y, et al. Preoperative prediction of lymph node metastasis of pancreatic ductal adenocarcinoma based on a radiomics nomogram of dual-parametric MRI imaging[J/OL]. Front Oncol, 2022, 12: 927077 [2024-02-05]. https://pubmed.ncbi.nlm.nih.gov/35875061/. DOI: 10.3389/fonc.2022.927077.
[22]
刘琢玉, 黄小华, 陈雨薇, 等. 基于T1WI动脉期影像组学模型预测胰腺导管腺癌淋巴结转移的研究[J]. 磁共振成像, 2022, 13(8): 30-35. DOI: 10.12015/issn.1674-8034.2022.08.006.
LIU Z Y, HUANG X H, CHEN Y W, et al. Prediction of lymph node metastasis of pancreatic ductal adenocarcinoma based on radiomics model of T1WI arterial phase[J]. Chin J Magn Reson Imag, 2022, 13(8): 30-35. DOI: 10.12015/issn.1674-8034.2022.08.006.
[23]
ZENG P E, QU C, LIU J F, et al. Comparison of MRI and CT-based radiomics for preoperative prediction of lymph node metastasis in pancreatic ductal adenocarcinoma[J]. Acta Radiol, 2023, 64(7): 2221-2228. DOI: 10.1177/02841851221142552.
[24]
YU Y Y, ZHANG R, DONG R T, et al. Feasibility of an ADC-based radiomics model for predicting pelvic lymph node metastases in patients with stage IB-IIA cervical squamous cell carcinoma[J/OL]. Br J Radiol, 2019, 92(1097): 20180986 [2024-02-05]. https://pubmed.ncbi.nlm.nih.gov/30888846/. DOI: 10.1259/bjr.20180986.
[25]
JAJODIA A, GUPTA A, PROSCH H, et al. Combination of radiomics and machine learning with diffusion-weighted MR imaging for clinical outcome prognostication in cervical cancer[J]. Tomography, 2021, 7(3): 344-357. DOI: 10.3390/tomography7030031.
[26]
WU Q X, SHI D P, DOU S W, et al. Radiomics analysis of multiparametric MRI evaluates the pathological features of cervical squamous cell carcinoma[J]. J Magn Reson Imaging, 2019, 49(4): 1141-1148. DOI: 10.1002/jmri.26301.
[27]
XIAO M L, MA F H, LI Y, et al. Multiparametric MRI-based radiomics nomogram for predicting lymph node metastasis in early-stage cervical cancer[J]. J Magn Reson Imaging, 2020, 52(3): 885-896. DOI: 10.1002/jmri.27101.
[28]
HOU L N, ZHOU W, REN J L, et al. Radiomics analysis of multiparametric MRI for the preoperative prediction of lymph node metastasis in cervical cancer[J/OL]. Front Oncol, 2020, 10: 1393 [2024-02-05]. https://pubmed.ncbi.nlm.nih.gov/32974143/. DOI: 10.3389/fonc.2020.01393.
[29]
LI C L, YIN J D. Radiomics based on T2-weighted imaging and apparent diffusion coefficient images for preoperative evaluation of lymph node metastasis in rectal cancer patients[J/OL]. Front Oncol, 2021, 11: 671354 [2024-02-05]. https://pubmed.ncbi.nlm.nih.gov/34041033/. DOI: 10.3389/fonc.2021.671354.
[30]
LIU X, WANG X P, ZHANG Y F, et al. Preoperative prediction of pelvic lymph nodes metastasis in prostate cancer using an ADC-based radiomics model: comparison with clinical nomograms and PI-RADS assessment[J]. Abdom Radiol, 2022, 47(9): 3327-3337. DOI: 10.1007/s00261-022-03583-5.
[31]
LIU X, TIAN J Y, WU J Y, et al. Utility of diffusion weighted imaging-based radiomics nomogram to predict pelvic lymph nodes metastasis in prostate cancer[J/OL]. BMC Med Imaging, 2022, 22(1): 190 [2024-02-05]. https://pubmed.ncbi.nlm.nih.gov/36333664/. DOI: 10.1186/s12880-022-00905-3.
[32]
WANG T, GAO T T, YANG J B, et al. Preoperative prediction of pelvic lymph nodes metastasis in early-stage cervical cancer using radiomics nomogram developed based on T2-weighted MRI and diffusion-weighted imaging[J]. Eur J Radiol, 2019, 114: 128-135. DOI: 10.1016/j.ejrad.2019.01.003.

上一篇 基于LI-RADS v2018及MRI特征对肝细胞癌CK19表达的风险预测及预后评估
下一篇 子宫性不孕的多参数MRI、MR-HSG及临床特征分析
  
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