分享:
分享到微信朋友圈
X
临床研究
DCE-MRI定量参数直方图特征联合ADC值对局部晚期宫颈癌放化疗疗效的预测价值
何雨琪 杜云霞 徐文翔 李飞翔 孙赟 彭乐平 王莉莉 黄刚

Cite this article as: HE Y Q, DU Y X, XU W X, et al. Predictive value of quantitative parameters from DCE-MRI histogram combined with ADC value for chemoradiotherapy efficacy in locally advanced cervical cancer[J]. Chin J Magn Reson Imaging, 2025, 16(6): 93-99, 109.本文引用格式:何雨琪, 杜云霞, 徐文翔, 等. DCE-MRI定量参数直方图特征联合ADC值对局部晚期宫颈癌放化疗疗效的预测价值[J]. 磁共振成像, 2025, 16(6): 93-99, 109. DOI:10.12015/issn.1674-8034.2025.06.014.


[摘要] 目的 探讨动态对比增强磁共振成像(dynamic contrast-enhanced magnetic resonance imaging, DCE-MRI)定量参数直方图特征联合表观弥散系数(apparent diffusion coefficient, ADC)预测局部晚期宫颈癌(locally advanced cervical cancer, LACC)放化疗疗效的价值。材料与方法 于甘肃省人民医院回顾性分析2017年1月至2023年12月88例LACC同步放化疗患者的临床及影像资料,前瞻性收集2023年12月至2024年5月15例LACC患者。按照实体瘤临床疗效评价标准(response evaluation criteria in solid tumor, RECIST)v1.1将患者分为显著反应组与非显著反应组。在DCE-MRI图像上选择肿瘤最大层面全肿瘤轮廓作为感兴趣区(region of interest, ROI)获得转运常数(volume transport constant, Ktrans)、血管外细胞外间隙容积分数(extravascular extracellular volume fraction, Ve)、速率常数(rate constant, Kep)原始频数表,导入IBM SPSS Statistics 27软件计算直方图特征,103例患者基于时间序列分层分割策略分为训练集88例,验证集15例,利用机器学习筛选最优DCE-MRI定量参数直方图特征并计算灌注参数评分(DCEscore);同时在ADC图测量ADC值。构建DCE直方图特征模型、ADC值及联合模型预测LACC放化疗疗效。采用受试者工作特征(receiver operating characteristic, ROC)曲线、校准曲线和决策曲线评估模型效能。比较临床参数及直方图特征在LACC患者放化疗疗效显著反应组及非显著反应组间差异,单因素及多因素回归分析筛选宫颈癌放化疗疗效独立危险因素。结果 基于DCE-MRI定量参数直方图特征模型预测LACC患者放化疗疗效训练集、验证集ROC曲线下面积(area under the curve, AUC)分别为0.922、0.841;ADC值预测LACC患者放化疗疗效训练集、验证集AUC为0.835、0.705;DCEscore联合ADC值预测LACC患者放化疗疗效效能最佳,训练集、验证集AUC为0.943、0.909。临床参数中,身体质量指数(body mass index, BMI)在显著反应组及非显著反应组之间差异具有统计学意义(P=0.032)。单因素逻辑回归分析结果表明BMI、DCEscore、ADC是LACC放化疗疗效的影响因素(OR值分别为1.264、277.9、0.001;P值分别为0.008、<0.001、0.002),多因素逻辑回归筛选DCEscore及ADC值是宫颈癌放化疗疗效的独立危险因素(OR值分别为518.2、0.002;P值分别为<0.001、0.007)。结论 基于DCE-MRI定量参数直方图特征联合ADC值构建的联合模型能够治疗前预测宫颈癌放化疗疗效,提示DCE-MRI定量参数直方图特征联合ADC值可能为LACC患者精准医疗提供一种无创评估方法。
[Abstract] Objective To investigate the predictive value of quantitative histogram features from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) combined with apparent diffusion coefficient (ADC) in assessing the efficacy of radiotherapy for locally advanced cervical cancer (LACC).Materials and Methods A retrospective analysis was conducted on the clinical and imaging data of 88 patients with concurrent chemoradiotherapy for LACC in Gansu Provincial People's Hospital from January 2017 to December 2023. Prospectively, 15 patients with LACC in Gansu Provincial People's Hospital from December 2023 to May 2024 were collected. According to response evaluation criteria in solid tumors (RECIST) v1.1 standard, the patients were divided into significant response group and non-significant tumors group. On the DCE-MRI images, the contour of the entire tumor at the largest layer of the tumor was selected as the region of interest (ROI) to obtain the original frequency tables of the transport constant (Ktrans), volume fraction (Ve), and rate constant (Kep). The IBM SPSS Statistics 27 software was imported to calculate the histogram characteristics. A total of 103 patients were divided into 88 cases in the training set and 15 cases in the validation set based on the hierarchical segmentation strategy of time series. Machine learning was used to screen the optimal histogram characteristics of quantitative parameters of DCE-MRI and calculate the perfusion parameter score (DCEscore). Meanwhile, measure the ADC value on the ADC diagram. DCE histogram feature model, ADC value and combined model were constructed to predict the efficacy of LACC chemoradiotherapy. Receiver operating characteristic (ROC) curves, calibration curves and decision curves were used to evaluate the model performance. The difference of clinical parameters and histogram features between the significant response group and the non-significant response group in LACC patients with radiotherapy and chemotherapy was compared and analyzed. Univariate and multivariate regression analysis was used to screen independent risk factors for radiotherapy and chemotherapy for cervical cancer.Results The area under the curve (AUC) of the training set and the validation set were 0.922 and 0.841, respectively, for the treatment of LACC patients based on the DCE-MRI quantitative parameter histogram feature model. ADC values to predict radiotherapy efficacy in LACC patients training set, validation set AUC of 0.835, 0.705. DCEscore combined with ADC values predicted the best efficacy of radiotherapy efficacy in LACC patients, with training set and validation set AUC of 0.943, 0.909. Among clinical parameters, body mass index (BMI) showed a statistically significant difference between the significant response group and the non-significant response group (P = 0.032). The results of univariate logistic regression analysis showed that BMI, DCEscore, and ADC were the influencing factors for the efficacy of radiotherapy for locally advanced cervical cancer (OR values of 1.264, 277.9, and 0.001, respectively; P values of 0.008, < 0.001, and 0.002, respectively), and multivariate logistic regression screened that the DCEscore and ADC values were the independent risk factors (OR 518.2, 0.002; P values < 0.001, 0.007, respectively).Conclusions The combined model based on DCE-MRI quantitative parameter histogram features combined with ADC values can predict the efficacy of radiotherapy and chemotherapy for cervical cancer before treatment, suggesting that DCE-MRI quantitative parameter histogram features combined with ADC values may provide a non-invasive evaluation method for precision medical treatment of locally advanced cervical cancer patients.
[关键词] 宫颈癌;放化疗;动态对比增强磁共振成像;表观弥散系数;机器学习
[Keywords] cervical cancer;chemoradiotherapy;dynamic contrast-enhanced magnetic resonance imaging;apparent diffusion coefficient;machine learning

何雨琪 1   杜云霞 1   徐文翔 1   李飞翔 1   孙赟 1   彭乐平 1   王莉莉 2   黄刚 2*  

1 甘肃中医药大学第一临床医学院,兰州 730000

2 甘肃省人民医院放射科,兰州 730000

通信作者:黄刚,E-mail:huang_g2024@163.com

作者贡献声明::黄刚设计本研究的方案,对稿件重要内容进行了修改;何雨琪起草和撰写稿件,获取、分析和解释本研究的数据;杜云霞、徐文翔、李飞翔、孙赟、彭乐平、王莉莉获取、分析或解释本研究的数据,对稿件重要的内容进行了修改;全体作者都同意发表最后的修改稿,同意对本研究的所有方面负责,确保本研究的准确性和诚信。


收稿日期:2024-10-26
接受日期:2025-05-10
中图分类号:R445.2  R737.33 
文献标识码:A
DOI: 10.12015/issn.1674-8034.2025.06.014
本文引用格式:何雨琪, 杜云霞, 徐文翔, 等. DCE-MRI定量参数直方图特征联合ADC值对局部晚期宫颈癌放化疗疗效的预测价值[J]. 磁共振成像, 2025, 16(6): 93-99, 109. DOI:10.12015/issn.1674-8034.2025.06.014.

0 引言

       宫颈癌是女性第五大常见肿瘤,发病率4.1%~43.1%,死亡率2.5%~20%[1]。美国国立综合癌症网络(National Comprehensive Cancer Network, NCCN)子宫颈癌临床实践指南2024(第一版)[2]推荐Ⅱb期以上的患者首选同步放化疗,但部分患者由于对放化疗不敏感,其5年内总生存率仅15%[3]。研究表明局部晚期宫颈癌(locally advanced cervical cancer, LACC)放化疗敏感性与肿瘤缺氧有关[4, 5, 6],而肿瘤缺氧涉及缺氧诱导因子-1α(hypoxia-inducible factor-1α, HIF-1α)引导的多种机制[7]。如果治疗前能准确评估宫颈癌缺氧状态,就能够精准选择敏感患者的治疗方案。HIF-1α可通过术后病理测定,但是所获取的组织样本比较局限,常常不能反映肿瘤的整体状况。研究显示动态对比增强磁共振成像(dynamic contrast-enhanced magnetic resonance imaging, DCE-MRI)定量参数能反映肿瘤的血管生成及血流灌注,这些参数与HIF-1α表达存在相关性,能够间接反映宫颈肿瘤组织缺氧状况[8, 9]。有研究发现治疗前肿瘤的转运常数(volume transport constant, Ktrans)值能够预测宫颈癌放化疗疗效(AUC为0.76~0.813)[10, 11]。磁共振弥散加权成像(magnetic resonance diffusion-weighted imaging, MR-DWI)是一种常见的功能成像,其定量参数表观弥散系数(apparent diffusion coefficient, ADC)能表征肿瘤细胞密度,在评价宫颈癌同步放化疗疗效方面有一定价值[12, 13]。对于治疗前低ADC值的患者能在放化疗疗效中获益还是高ADC值的患者在放化疗疗效中获益,目前存在争议[12, 14]。DCE-MRI直方图能够提供多维度信息,可以对数据分布进行描述,如平均值、峰度、分位数等,这些参数能够可视化肿瘤异质性[15]。目前有一项研究对不同疗效的宫颈癌DCE-MRI直方图特征进行差异分析,但诊断效能并不高(AUC为0.664~0.740)[16];该研究特征的筛选并未采用机器学习方法,可能并未完全挖掘特征间潜在的关联性。由于机器学习能够对高通量特征进行最大程度的发掘与拟合,已经在肿瘤的鉴别、分层、预测预后等方面显示了很高的价值[17, 18, 19]。因此,本研究应用机器学习的方法对DCE-MRI定量参数直方图特征进行筛选以期充分挖掘特征间潜在的关联性,将筛选后的特征进行模型构建,并联合与之生物学意义不同的ADC值预测LACC放化疗疗效,以期为临床提供一种更高效的LACC患者分层方法。

1 材料与方法

1.1 研究对象

       本研究遵守《赫尔辛基宣言》,回顾性收集2017年1月至2023年12月在甘肃省人民医院因宫颈癌接受放化疗患者影像及临床资料,已申请免除受试者知情同意,前瞻性收集2023年12月至2024年5月在甘肃省人民医院因宫颈癌接受放化疗患者,受试者均已签署知情同意书。本研究经甘肃省人民医院伦理委员会批准,批准文号:2023-708。患者纳排流程详见图1

图1  患者纳排流程图。1A:回顾性研究患者纳排流程图;1B:前瞻性研究患者纳排流程图。
Fig. 1  Flow chart of patient scheduling. 1A: Retrospective study; 1B Prospective study.

1.2 图像分析

1.2.1 图像采集

       图像扫描采用SIEMENS 3.0 T Skyra(德国西门子公司)、1.5 T Amira MRI(德国西门子公司)、3.0 T Elition(荷兰飞利浦公司),分别采用18、13、16通道体部相控阵线圈行常规MRI和DCE-MRI扫描,对比剂采用Gd-GTPA(佳迪显,钆特酸葡胺注射液,江苏恒瑞医药股份有限公司,中国),剂量0.1 mmol/kg,以3.5 mL/s速率从上肢静脉注入,再以相同速率注入生理盐水。启动扫描18 s注入对比剂,扫描间隔时间约9 s,以宫颈癌肿瘤组织为中心重复采集35组图像,总扫描时间约5 min。DWI采用单次激发平面回波成像(single-shotecho planar imaging, SS-EPI)序列行横断面扫描,3 T MRI b值取50、1000 s/mm2,1.5 T MRI b值取50、800 s/mm2,将数据进行标准化处理。具体参数如表1所示。

表1  MRI扫描序列详细参数
Tab. 1  Detailed parameters of MRI scan sequence

1.2.2 图像处理

       由1名具有5年诊断经验的主治医师进行图像分析。DCE-MRI数据处理采用Extended Tofts Linear双室模型,首先按顺序导入2°和15°翻转角数据,将DCE-MRI序列35期图像调入MMWP version:VE4.0工作站,应用Tissue4D由髂内动脉的时间-信号强度曲线(time-intensity curve, TIC)确定血流峰值期,并在该期选择肿瘤最大层面全肿瘤轮廓作为感兴趣区(region of interest, ROI)进行勾画,尽量避开坏死,出血等区域,获得ROI内定量参数Ktrans、速率常数(rate constant, Kep)、血管外细胞外间隙容积分数(extravascular extracellular volume fraction, Ve)的原始频数表,利用IBM SPSS Statistics 27软件计算出部分直方图特征(均值、平均标准误差、中值、众数、标准差、方差、偏度、偏度标准误差、峰度、峰度标准误差、最大值、最小值、总和、第25%和75%分位数)。DWI序列扫描完成后,系统自动生成相应ADC图,在ADC图上放置一个类圆形ROI,选择肿瘤最大层面DWI呈高信号、ADC呈低信号区域作为ROI勾画测得ADC值,每个ROI的面积大于等于40 mm2。1个月后随机选择20名患者,由同一影像医师重新勾画ROI,计算所选特征的组内相关系数(intra-class correlation coefficient, ICC),以评估观察者前后两次提取特征之间的一致性,保留ICC>0.75的特征。

1.3 DCE-MRI定量参数直方图特征筛选

       利用pyradiomics的开源软件FAE(0.5.12版本;https://github.com/salan668/FAE)将原始频数表计算的直方图特征进行MinMax标准化处理,采用主成分分析(principal component analysis, PCA)、皮尔森相关系数(Pearson correlation coefficient, PCC)对特征进行降维,使用方差分析(analysis of variance, ANOVA)、Relief算法、递归式特征消除(recursive feature elimination, RFE)及Kruskal-Wallis测试进行特征筛选。选择的分类器有支持向量机(support vector machine, SVM)、线性判别分析(linear discriminant analysis, LDA)、逻辑回归(logistic regression, LR)、Lasso逻辑回归(logistic regression via Lasso, LR-Lasso)等。特征筛选流程见图2

图2  机器学习流程图。标蓝部分为本研究筛选特征所采用的方法。PCA:主成分分析;PCC:皮尔森相关系数;ANOVA:方差分析;RFE:递归式特征消除;SVM:支持向量机;LDA:线性判别分析;LR:逻辑回归。
Fig. 2  Machine learning flowchart. The highlighted blue section represents the method used for feature selection in this study. PCA: principal component analysis; PCC: Pearson correlation coefficient; ANOVA: analysis of variance; RFE: recursive feature elimination; SVM: support vector machine; LDA: linear discriminant analysis; LR: logistic regression.

1.4 模型构建及评估效能

       将筛选并标准化的灌注参数(Ktrans、Kep、Ve)直方图特征建模,将模型生成的概率作为DCEscore(DCEscore=-0.524+∑i=0nCi×Xi,i是选择的特征的数量,C是第i个特征的系数,X是第i个特征)。基于DCE-MRI直方图特征和ADC值构建单一及联合模型预测宫颈癌放化疗疗效。采用ROC曲线下面积(area under the curve, AUC)、校准曲线及决策曲线评价效能。

1.5 筛选影响宫颈癌放化疗疗效因素

       比较所筛选DCE-MRI直方图特征、ADC值及临床参数[中性粒细胞计数、淋巴细胞计数、中性粒细胞计数/淋巴细胞计数、白蛋白、BMI、糖类抗原125(carbohydrate antigen 125, CA125)、鳞状细胞癌抗原(squamous cell carcinoma antigen, SCC)、肿瘤直径大小]两组间有无差异。通过单因素及多因素logistic回归分析筛选影响宫颈癌放化疗疗效的因素。

1.6 LACC患者同步放化疗方案及疗效评估标准

       所有患者均给予同步放化疗,放疗采用盆腔外照射(2 Gy/次,5~6次/周,肿瘤剂量为60 Gy/30次)联合腔内后装放疗(6 Gy/次,1次/周,肿瘤剂量为30 Gy/5次);同步化疗(顺铂,40 mg/m2/周)。按实体瘤临床疗效评价标准(response evaluation criteria in solid tumor, RECIST)v1.1[20]评估疗效:完全缓解(complete response, CR),所有目标病变消失;部分缓解(partial response, PR),目标病变的总直径减少至少30%;疾病进展(progressive disease, PD),目标病变的总直径增加至少20%;疾病稳定(stable disease, SD),既不能达到PR,又不符合PD。显著反应(significant response, SR)组为完全缓解或部分缓解的患者,非显著反应(non-significant response, Non-SR)组为疾病进展或疾病稳定的患者。

1.7 统计学分析

       采用IBM SPSS Statistics 27软件进行数据的统计学分析。分别采用Kolmogorov-Smirnov检验和Levene检验进行连续型数据正态性及方差齐性检验,符合正态分布和方差齐性的数据采用独立样本t检验分析组间差异、非正态分布或方差不齐的数据采用Mann-Whitney U检验分析组间差异,分类数据采用卡方检验进行分析。单因素及多因素logistic回归分析筛选LACC患者放化疗疗效的独立危险因素。P<0.05认为差异具有统计学意义。

2 结果

2.1 临床资料

       本研究共纳入88例LACC患者,其中显著反应组69例,非显著反应组19例。中性粒细胞计数、淋巴细胞计数、中性粒细胞计数/淋巴细胞计数、白蛋白、CA125、SCC、肿瘤直径大小差异均无统计学意义(P>0.05)。BMI在显著反应组和非显著反应组之间差异具有统计学意义(P=0.032),非显著反应组患者BMI高于显著反应组患者(表2)。

表2  显著反应组与非显著反应组临床参数比较
Tab. 2  Comparison of clinical parameters between significant response group and non-significant response group

2.2 DCE-MRI直方图特征及ADC值在两组间的差异

       Ktrans(平均值、平均标准误差、中位数、众数、标准偏差、方差、最大值、最小值、总和、第25百分位数、第75百分位数)Kep(均值、中位数、众数、最小值、第25百分位数、第75百分位数)Ve(均值、平均标准误差、中位数、众数、标准偏差、偏度、峰值、最小值、第25百分位数、第75百分位数)及ADC值两组间差异具有统计学意义Ve(峰度、偏度)及Ktrans第25百分位数非显著反应组高于显著反应组,余特征非显著反应组值均低于显著反应组(表3)。

表3  显著反应组及非显著反应组DCE直方图特征及ADC值比较
Tab. 3  Comparison of DCE histogram features and ADC values between significant response group and non-significant response group

2.3 宫颈癌放化疗疗效的影响因素筛选

       单因素logistic回归分析BMI、DCEscore、ADC值差异具有统计学意义(P<0.05),多因素logistic回归DCEscore及ADC值是宫颈癌放化疗疗效的独立危险因素(P<0.05)见表4

表4  逻辑回归筛选预测宫颈癌放化疗疗效因素
Tab. 4  Logistic regression screening of factors predicting curative effect of radiotherapy and chemotherapy for cervical cancer

2.4 各模型构建及效能评估

       DCE-MRI定量参数(Ktrans,Kep,Ve)的直方图特征经过筛选以及降维后,最终筛选出3个定量参数直方图特征:Ktrans平均值、Ktrans中位数、Ktrans第75百分位数。基于筛选得出的直方图特征建立预测LACC放化疗疗效模型并计算DCEscore,训练集中该模型AUC为0.922(95% CI:0.845~0.969),验证集AUC为0.841(95% CI:0.565~0.974);ADC值训练集AUC为0.835(95% CI:0.741~0.906),验证集AUC为0.705(95% CI:0.420~0.905);DCEscore联合ADC模型训练集AUC为0.943(95% CI:0.872~0.981),验证集AUC为0.909(95% CI:0.648~0.995);三种模型敏感度分别为94.74%,78.95%,100.00%,特异度分别为81.16%,88.41%,81.16%。

       通过决策曲线对各模型预测宫颈癌放化疗疗效分析,结果表明联合模型净收益最大,通过ROC曲线及决策曲线分析联合模型预测宫颈癌放化疗疗效性能最好。DeLong检验结果表明,DCE直方图特征与ADC值、DCE直方图特征与联合模型、ADC值与联合模型ZP分别为1.422与0.155、1.339与0.181、2.078与0.038。详见图345678图9

图3  DCE-MRI直方图特征、ADC值及联合模型预测宫颈癌放化疗疗效ROC曲线。3A:训练集;3B:验证集。
图4  基于DCE直方图特征及ADC值所构建的列线图。
图5  校准曲线。对角虚线参考线表明理想模型基于DCE直方图特征联合ADC具有完美的性能。实线表示联合模型的性能,对角虚线参考线与实线越接近表示性能越好。
图6  DCE直方图特征、ADC、联合模型的决策曲线分析。联合模型净收益最大。
图7  DeLong检验结果。
图8  女,41岁,宫颈癌放化疗疗效显著反应患者。DCEscore为0.014,ADC值为1.09 mm2/s,总分约为28,对应Risk约为<1%,表明列线图模型预测该患者放化疗非显著组的概率<1%。
图9  女,45岁,宫颈癌放化疗疗效非显著反应患者。DCEscore为0.983,ADC值为0.67 mm2/s,总分约为124,对应Risk约为91%,表明列线图模型预测该患者放化疗非显著组的概率约91%。DCE:动态对比增强;ADC:表观弥散系数;ROC:受试者工作特征;DCEscore:DCE灌注参数评分。
Fig. 3  DCE-MRI histogram features, ADC values and ROC curve of combined model predicting the curative effect of radiotherapy and chemotherapy for cervical cancer. 3A: Training set; 3B: Validation set.
Fig. 4  A nomograph based on DCE histogram features and ADC values.
Fig. 5  Diagram calibration curve. The diagonal dotted reference line indicates that the ideal model based on DCE histogram features combined with ADC has perfect performance. The solid line represents the performance of the joint model, and the closer the diagonal dotted reference line is to the solid line, the better the performance is.
Fig. 6  Histogram features, ADC and decision curve analysis (DCA) of the combined model show that the combined model has the largest net benefit.
Fig. 7  The results of DeLong test.
Fig. 8  Female, 41 years old, the patients with significant response to radiotherapy and chemotherapy for cervical cancer. DCEscore of 0.014, ADC value of 1.09 mm2/s, total points of about 28, and corresponding risk of about < 1%.
Fig. 9  Female, 45 years old, the patients with non-significant response to radiotherapy and chemotherapy for cervical cancer. DCEscore of 0.983, ADC value of 0.67 mm2/s, total points of about 124, and the corresponding risk of about 91%. DCE: dynamic contrast-enhanced; ADC: apparent diffusion coefficient; ROC: receiver operating characteristic; DCEscore: DCE parameter scoring.

3 讨论

       本研究探讨了DCE-MRI定量参数直方图特征联合ADC值对LACC患者同步放化疗疗效的预测价值,结果显示联合模型具有良好的预测效能,能够为临床更好地分层LACC患者放化疗治疗提供一种高效的方法。

3.1 DCE-MRI直方图特征在宫颈癌放化疗疗效方面的价值

       肿瘤缺氧是宫颈癌患者接受放化疗治疗失败的主要原因[21, 22],在放化疗前识别这部分肿瘤氧合不良的患者,能够及时地调整治疗策略,从而改善患者预后。DEC-MRI参数能够表征肿瘤内的微循环灌注和血管生成[23],其中Ktrans值被认为与肿瘤内的乏氧状态密切相关[8, 9, 23]。研究显示Ktrans与宫颈癌放化疗后肿瘤消退率呈正相关[24],也可以独立预测LACC放化疗预后(HR=0.016,P<0.023)及评估宫颈癌同步放化疗早期变化[25, 26]。直方图是一种描述数据分布的可视化统计图,通过计算基于全肿瘤轮廓DCE-MRI参数图像素的直方图特征可以显示肿瘤内DCE-MRI参数分布的特点,通过值的映射甚至可以对不同数值频段的区域进行直观显示,因此直方图特征能够反映肿瘤内的异质性[15]。由于DCE-MRI参数是基于数学模型计算获得的,具有比较确切的病理生理学意义,因此也可以更好地解释直方图特征的生物学意义。本研究基于DCE-MRI直方图特征并通过机器学习筛选,由于特征之间存在交互作用,最终筛选了3个最优特征(Ktrans平均值、Ktrans中位数、Ktrans第75百分位数)构建预测宫颈癌放化疗疗效的模型,这些特征均来自Ktrans参数图。Ktrans为容积转移常数,反映组织的血流灌注和血管通透性。既往研究显示Ktrans值可以表征肿瘤缺氧及反映放疗抗性[27, 28],Ktrans值降低可以反映治疗引起的肿瘤缺氧[23],本研究构建的模型所包含的直方图特征反映了肿瘤内Ktrans整体水平(均数和中位数)和Ktrans离散程度(第75%分位数),这些特征在非显著反应组更低,提示了肿瘤内的缺氧可能是疗效不佳的原因,与既往研究结果一致。

3.2 ADC值在宫颈癌放化疗疗效方面的价值

       本研究也探讨了ADC值预测宫颈癌放化疗疗效的价值,通过测量肿瘤组织ADC值也能较好地预测放化疗疗效。ADC值能反映细胞密度,恶性肿瘤细胞密度高,更易发生肿瘤缺氧坏死[29, 30]。但既往研究ADC值预测宫颈癌放化疗疗效结果并不一致,可能与ADC值测量时样本量大小、b值的选择、ADC值数学算法模型及ROI的选择等有关[13, 31]。虽然ADC值预测宫颈癌的放化疗疗效的效能比DCE-MRI直方图模型预测的效能低(AUC:0.835 vs. 0.922),但是由于它们成像的原理和反映的生物学意义不相同,联合后能够提升整体的预测效能。

3.3 BMI在宫颈癌放化疗疗效方面的价值

       另外本研究发现BMI是LACC放化疗疗效的影响因素,非显著反应组BMI高于显著反应组。研究表明肥胖是许多癌症进展和转移的危险因素,高BMI会增加宫颈癌患病风险[32]。高BMI促进肿瘤进展可能与肥胖相关的肿瘤坏死因子、胰岛素和棕榈酸酯驱动肿瘤相关巨噬细胞表达PD-1[33],从而诱导肿瘤免疫逃逸等相关,另外高BMI患者由于脂蛋白累积会促进肿瘤转移[34],这些原因均可能导致宫颈癌放化疗疗效差。

3.4 局限性及展望

       本文存在以下局限性:第一,本研究选择最大层面勾画全肿瘤轮廓作为ROI,虽有研究显示肿瘤最大层面作为ROI与全肿瘤整体作为ROI并无明显差异[35],但仍需进一步分析全肿瘤以期获取更多信息;第二,本研究为单中心回顾性研究,样本量少,虽然进行了前瞻性外部验证,但结果可能仍然存在偏倚,未来需要扩大样本量,多中心进一步研究;第三,由于不同机器影响,尽管数据进行了标准化及一致性分析显示结果良好,但由于机器间的差异,未来应制定统一扫描标准,以确保结果更加稳定,未来进一步扩大样本量,以分析不同机器间的影响。

4 结论

       综上所述,基于DCE-MRI定量参数直方图特征联合ADC值构建的模型能够预测LACC放化疗疗效,该模型可能是通过表征肿瘤内的缺氧状态来实现对放化疗疗效的预测,为临床提供了一种无创的、可解释的宫颈癌放化疗疗效预测方法。

[1]
BRAY F, FERLAY J, SOERJOMATARAM I, et al. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries[J]. CA Cancer J Clin, 2018, 68(6): 394-424. DOI: 10.3322/caac.21492.
[2]
周晖, 刘昀昀, 罗铭, 等. 《2024 NCCN子宫颈癌临床实践指南(第1版)》更新解读[J]. 中国实用妇科与产科杂志, 2023, 39(11): 1119-1121. DOI: 10.19538/j.fk2023110113.
ZHOU H, LIU Y Y, LUO M, et al. Update and interpretation of 2024 NCCN clinical practice guidelines for cervical cancer (1st edition)[J]. Chin J Pract Gynecol Obstet, 2023, 39(11): 1119-1121. DOI: 10.19538/j.fk2023110113.
[3]
LIU Y, WU L, TONG R, et al. PD-1/PD-L1 inhibitors in cervical cancer[J/OL]. Front Pharmacol, 2019, 10: 65 [2023-03-29]. https://pubmed.ncbi.nlm.nih.gov/30774597/. DOI: 10.3389/fphar.2019.00065.
[4]
YOON M, LEE H K, PARK E Y, et al. Randomized multicenter phase II trial of prophylactic irradiation of Para-aortic lymph nodes in advanced cervical cancer according to tumor hypoxia: Korean Radiation Oncology Group (KROG 07-01) study[J]. Int J Cancer, 2022, 151(12): 2182-2194. DOI: 10.1002/ijc.34190.
[5]
HAN K, FYLES A, SHEK T, et al. A phase II randomized trial of chemoradiation with or without metformin in locally advanced cervical cancer[J]. Clin Cancer Res, 2022, 28(24): 5263-5271. DOI: 10.1158/1078-0432.CCR-22-1665.
[6]
YE Y Q, HU Q S, CHEN H, et al. Characterization of hypoxia-associated molecular features to aid hypoxia-targeted therapy[J]. Nat Metab, 2019, 1(4): 431-444. DOI: 10.1038/s42255-019-0045-8.
[7]
WU Q H, YOU L, NEPOVIMOVA E, et al. Hypoxia-inducible factors: master regulators of hypoxic tumor immune escape[J/OL]. J Hematol Oncol, 2022, 15(1): 77 [2024-03-29]. https://pubmed.ncbi.nlm.nih.gov/35659268/. DOI: 10.1186/s13045-022-01292-6.
[8]
HAUGE A, WEGNER C S, GAUSTAD J V, et al. DCE-MRI of patient-derived xenograft models of uterine cervix carcinoma: associations with parameters of the tumor microenvironment[J/OL]. J Transl Med, 2017, 15(1): 225 [2023-03-19]. https://pubmed.ncbi.nlm.nih.gov/29100521/. DOI: 10.1186/s12967-017-1331-4.
[9]
LI X S, WU S D, LI D C, et al. Intravoxel incoherent motion combined with dynamic contrast-enhanced perfusion MRI of early cervical carcinoma: correlations between multimodal parameters and HIF-1α expression[J]. J Magn Reson Imaging, 2019, 50(3): 918-929. DOI: 10.1002/jmri.26604.
[10]
秦凤英, 马锦涛, 赵明丽, 等. 多参数MRI联合临床分期预测ⅢCr期宫颈癌同步放化疗后复发[J]. 放射学实践, 2021, 36(6): 767-772. DOI: 10.13609/j.cnki.1000-0313.2021.06.015.
QIN F Y, MA J T, ZHAO M L, et al. The value of multi-parameter MRI combined with clinical stage in predicting recurrence of stage ⅢCr cervical cancer after concurrent chemo-radiotherapy[J]. Radiol Pract, 2021, 36(6): 767-772. DOI: 10.13609/j.cnki.1000-0313.2021.06.015.
[11]
ZHENG X M, GUO W Q, DONG J N, et al. Prediction of early response to concurrent chemoradiotherapy in cervical cancer: Value of multi-parameter MRI combined with clinical prognostic factors[J]. Magn Reson Imaging, 2020, 72: 159-166. DOI: 10.1016/j.mri.2020.06.014.
[12]
YANG W, QIANG J W, TIAN H P, et al. Multi-parametric MRI in cervical cancer: early prediction of response to concurrent chemoradiotherapy in combination with clinical prognostic factors[J]. Eur Radiol, 2018, 28(1): 437-445. DOI: 10.1007/s00330-017-4989-3.
[13]
ZHANG H, ZHOU Y Y, LI J, et al. The value of DWI in predicting the response to synchronous radiochemotherapy for advanced cervical carcinoma: comparison among three mathematical models[J/OL]. Cancer Imaging, 2020, 20(1): 8 [2024-04-09].https://pubmed.ncbi.nlm.nih.gov/31937371/. DOI: 10.1186/s40644-019-0285-6.
[14]
WANG Y C, HU D Y, HU X M, et al. Assessing the early response of advanced cervical cancer to neoadjuvant chemotherapy using intravoxel incoherent motion diffusion-weighted magnetic resonance imaging: a pilot study[J]. Chin Med J (Engl), 2016, 129(6): 665-671. DOI: 10.4103/0366-6999.177995.
[15]
JUST N. Improving tumour heterogeneity MRI assessment with histograms[J]. Br J Cancer, 2014, 111(12): 2205-2213. DOI: 10.1038/bjc.2014.512.
[16]
ZHANG A N, SONG J C, MA Z L, et al. Combined dynamic contrast-enhanced magnetic resonance imaging and diffusion-weighted imaging to predict neoadjuvant chemotherapy effect in FIGO stage IB2-IIA2 cervical cancers[J]. La Radiol Med, 2020, 125(12): 1233-1242. DOI: 10.1007/s11547-020-01214-x.
[17]
SWANSON K, WU E, ZHANG A, et al. From patterns to patients: Advances in clinical machine learning for cancer diagnosis, prognosis, and treatment[J]. Cell, 2023, 186(8): 1772-1791. DOI: 10.1016/j.cell.2023.01.035.
[18]
COTTER G, BEAL E W, POULTSIDES G A, et al. Using machine learning to preoperatively stratify prognosis among patients with gallbladder cancer: a multi-institutional analysis[J]. HPB (Oxford), 2022, 24(11): 1980-1988. DOI: 10.1016/j.hpb.2022.06.008.
[19]
CHEN D L, CAI J H, WANG C C N. Identification of key prognostic genes of triple negative breast cancer by LASSO-based machine learning and bioinformatics analysis[J/OL]. Genes (Basel), 2022, 13(5): 902 [2024-04-19]. https://pubmed.ncbi.nlm.nih.gov/35627287/. DOI: 10.3390/genes13050902.
[20]
TIAN X, SUN C X, LIU Z Y, et al. Prediction of response to preoperative neoadjuvant chemotherapy in locally advanced cervical cancer using multicenter CT-based radiomic analysis[J/OL]. Front Oncol, 2020, 10: 77 [2024-04-19]. https://pubmed.ncbi.nlm.nih.gov/32117732/. DOI: 10.3389/fonc.2020.00077.
[21]
VAUPEL P, MAYER A. Hypoxia in cancer: significance and impact on clinical outcome[J]. Cancer Metastasis Rev, 2007, 26(2): 225-239. DOI: 10.1007/s10555-007-9055-1.
[22]
SUNDFØR K, LYNG H, TROPÉ C G, et al. Treatment outcome in advanced squamous cell carcinoma of the uterine cervix: relationships to pretreatment tumor oxygenation and vascularization[J]. Radiother Oncol, 2000, 54(2): 101-107. DOI: 10.1016/s0167-8140(99)00175-9.
[23]
GAUSTAD J V, HAUGE A, WEGNER C S, et al. DCE-MRI of tumor hypoxia and hypoxia-associated aggressiveness[J/OL]. Cancers (Basel), 2020, 12(7): 1979 [2024-05-02]. https://pubmed.ncbi.nlm.nih.gov/32698525/. DOI: 10.3390/cancers12071979.
[24]
赵晓艳. DCE-MRI定量参数在宫颈癌放化疗效果评价及其与病理特征的相关性分析[J]. 医学影像学杂志, 2020, 30(11): 2069-2073.
ZHAO X Y. Quantitative parameters of DCE-MRI in evaluating the efficacy of radiotherapy and chemotherapy for cervical cancer and their correlation with pathological features[J]. J Med Imag, 2020, 30(11): 2069-2073.
[25]
ZHANG Q, GUO J X, OUYANG H, et al. Added-value of dynamic contrast-enhanced MRI on prediction of tumor recurrence in locally advanced cervical cancer treated with chemoradiotherapy[J]. Eur Radiol, 2022, 32(4): 2529-2539. DOI: 10.1007/s00330-021-08279-w.
[26]
KIM J H, KIM C K, PARK B K, et al. Dynamic contrast-enhanced 3-T MR imaging in cervical cancer before and after concurrent chemoradiotherapy[J]. Eur Radiol, 2012, 22(11): 2533-2539. DOI: 10.1007/s00330-012-2504-4.
[27]
ELLINGSEN C, HOMPLAND T, GALAPPATHI K, et al. DCE-MRI of the hypoxic fraction, radioresponsiveness, and metastatic propensity of cervical carcinoma xenografts[J]. Radiother Oncol, 2014, 110(2): 335-341. DOI: 10.1016/j.radonc.2013.10.018.
[28]
HILLESTAD T, HOMPLAND T, FJELDBO C S, et al. MRI distinguishes tumor hypoxia levels of different prognostic and biological significance in cervical cancer[J]. Cancer Res, 2020, 80(18): 3993-4003. DOI: 10.1158/0008-5472.CAN-20-0950.
[29]
JAIN R K. Determinants of tumor blood flow: a review[J]. Cancer Res, 1988, 48(10): 2641-2658.
[30]
VAUPEL P, KALLINOWSKI F, OKUNIEFF P. Blood flow, oxygen and nutrient supply, and metabolic microenvironment of human tumors: a review[J]. Cancer Res, 1989, 49(23): 6449-6465.
[31]
CIOLINA M, VINCI V, VILLANI L, et al. Texture analysis versus conventional MRI prognostic factors in predicting tumor response to neoadjuvant chemotherapy in patients with locally advanced cancer of the uterine cervix[J]. Radiol Med, 2019, 124(10): 955-964. DOI: 10.1007/s11547-019-01055-3.
[32]
SAND F L, URBUTE A, RING L L, et al. The influence of overweight and obesity on participation in cervical cancer screening: a systematic review and meta-analysis[J/OL]. Prev Med, 2023, 172: 107519 [2024-06-10]. https://pubmed.ncbi.nlm.nih.gov/37080502/. DOI: 10.1016/j.ypmed.2023.107519.
[33]
BADER J E, WOLF M M, LUPICA-TONDO G L, et al. Obesity induces PD-1 on macrophages to suppress anti-tumour immunity[J]. Nature, 2024, 630(8018): 968-975. DOI: 10.1038/s41586-024-07529-3.
[34]
SHANG C L, LI Y, HE T H, et al. The prognostic miR-532-5p-correlated CeRNA-mediated lipid droplet accumulation drives nodal metastasis of cervical cancer[J]. J Adv Res, 2021, 37: 169-184. DOI: 10.1016/j.jare.2021.09.009.
[35]
TAMADA T, HUANG C C, REAM J M, et al. Apparent diffusion coefficient values of prostate cancer: comparison of 2D and 3D ROIs[J]. AJR Am J Roentgenol, 2018, 210(1): 113-117. DOI: 10.2214/AJR.17.18495.

上一篇 基于心脏磁共振探析扩张型心肌病左心房应变与左心室纤维化的相关性
下一篇 MR影像组学列线图对胎盘植入性疾病的产前诊断及不良临床结局预测
  
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