基于T2WI和DWI的磁共振影像组学在术前预测直肠癌壁外血管侵犯的价值研究

doi:10.16150/j.1671-2870.2024.01.007

摘要/Abstract

摘要：

目的: 探讨基于磁共振（magnetic resonance imaging，MRI）T2加权成像（T2-weighted imaging，T2WI）和弥散加权成像（diffusion-weighted imaging，DWI）的影像组学，在术前预测直肠癌壁外血管侵犯（extramural vascular invasion, EMVI）的诊断效能。方法: 回顾性收集2010年1月至2023年6月经术后病理证实为直肠腺癌且术前行直肠MRI扫描的患者168例，按7∶3随机分为训练集和验证集。提取T2WI、DWI的影像组学特征，采用最大相关最小冗余 (the maximum relevance minimum redundancy，mRMR)和十倍交叉验证的最小绝对收缩与选择算子（the least absolute shrinkage and selection operator，LASSO）回归分析降维并选择影像组学特征，计算每例患者的影像组学总评分（Radscore），使用Radscore建立影像组学模型。在训练集中，研究纳入了3个临床特征[年龄、性别、术前癌胚抗原（carcinoembryonic antigen，CEA）]和6个磁共振影像学特征[ADC值、浸润深度、肿瘤长度、肿瘤部位、T分期、MRI壁外血管侵犯(magnetic resonance imaging-defined EMVI, mrEMVI)评分]，通过单因素、多因素Logistic回归分析建立临床模型。联合Radscore和临床模型的独立危险因素，建立临床-影像组学模型（联合模型）。采用受试者操作特征（receiver operating characteristic, ROC）曲线评估各模型的诊断效能，通过DeLong检验比较不同模型的效能差异，采用校准曲线评估列线图术前预测结果与术后病理真实状况的拟合度，运用决策曲线分析（decision curve analysis, DCA）评价3种模型的临床应用价值。结果: 联合模型、临床模型、影像组学模型ROC曲线在训练集和验证集中AUC分别为0.926、0.888、0.756和0.917、0.896、0.782，联合模型的诊断效能最佳。Delong检验显示，在训练集和验证集中，联合模型诊断效能高于影像组学模型（P<0.05）；在训练集中，联合模型的诊断效能高于临床模型（P<0.05），但在验证集中差异无统计学意义（P>0.05）。校准曲线显示列线图术前预测结果与术后病理结果一致性良好（P<0.05）。DCA结果表明，当风险阈值概率在0.24~0.77时，联合模型在临床上的获益高于临床模型和影像组学模型。结论: 基于T2WI和DWI的MRI影像组学模型术前预测直肠癌EMVI有较高的诊断效能，联合临床模型中独立危险因素构建的临床-影像组学MRI模型（联合模型）进一步提高了诊断效能。

关键词: 直肠癌, 壁外血管侵犯, 磁共振成像, 影像组学, 预测模型, 列线图

Abstract:

Objective To investigate the diagnostic performance of radiomics based on T2-weighted imaging (T2WI) and diffusion-weighted imaging (DWI) of MRI for preoperative prediction of extramural vascular invasion (EMVI) in rectal cancer. Methods A total of 168 patients with pathology-confirmed rectal adenocarcinoma were enrolled during January 2010 to June 2023. The patients underwent preoperative rectal MRI scans, and they were randomly divided into training set and validation set at a 7∶3 ratio. Radiomic features from T2WI and DWI were extracted and selected by dimensionality reduction using the maximum relevance minimum redundancy (mRMR) method and the least absolute shrinkage and selection operator (LASSO) regression analysis with ten-fold cross-validation. The radiomic total score (Radscore) for each patient was calculated to make radiomics model. The training set enrolled three clinical features [gender, age and preoperative level of carcinoembryonic antigen (CEA)] and six magnetic resonance imaging features [ADC value, depth of infiltration, tumor length, tumor location, T staging and magnetic resonance imaging-defined extramural vascular invasion (mrEMVI)].The clinical model was established through univariable and multivariable logistic regression analysis based on above clinical and imaging features， and the clinical-radiomics model (combined model) was established with Radscore and independent risk factors from the clinical model. The diagnostic efficacy of each model was assessed using receiver operating characteristic (ROC) curve. The differences in performance among the models were compared using the DeLong test. The Calibration curves were employed to evaluate the consistence between the preoperative predictive results obtained from the nomogram and the postoperative pathological results. Additionally, decision curve analysis (DCA) was applied to evaluate the clinical utility of the three models. Results The area under the curve (AUC) of the ROC curve for the combined model, clinical model, and radiomics model in the training were 0.926, 0.888, 0.756, and were 0.917, 0.896, 0.782 in validation sets, respectively. The DeLong test showed that the diagnostic efficacy of combined model was higher than that of radiomics model in both training and validation sets (P<0.05). The diagnostic efficacy of combined model was better than that of clinical model in the training set (P<0.05). The calibration curve showed the consistency between the preoperative predictive results obtained from the nomogram and the postoperative pathological findings was satisfied. The DCA showed that the risk threshold probabilities between 0.24 and 0.77, the clinical benefit of combined model was higher than those of clinical model and the radiomics model. Conclusions For preoperative prediction of EMVI in rectal cancer,the radiomics model based on T2WI and DWI of MRI has a satisfied diagnostic efficiency, while the clinical-radiomics model (combined model) may further enhance the diagnostic efficiency.

Key words: Rectal cancer, Extramural vascular invasion, Magnetic resonance imaging, Radiomics, Prediction model, Nomogram

中图分类号:

丁景峰, 敖炜群, 朱珍, 孙静, 徐良根, 郑世保, 俞晶晶, 胡金文. 基于T2WI和DWI的磁共振影像组学在术前预测直肠癌壁外血管侵犯的价值研究[J]. 诊断学理论与实践, 2024, 23(01): 46-56.

DING Jingfeng, AO Weiqun, ZHU Zhen, SUN Jing, XU Lianggen, ZHENG Shibao, YU Jingjing, HU Jinwen. The value of radiomics based on T2WI and DWI of MRI in preoperative prediction of extramural vascular invasion in rectal cancer[J]. Journal of Diagnostics Concepts & Practice, 2024, 23(01): 46-56.

图/表 12

表1

图1

表2

表3

图2

图3

表4

图4

表5

图5

图6

图7

参考文献 29

[1]	XIA C, DONG X, LI H, et al. Cancer statistics in China and United States, 2022: profiles, trends, and determinants[J]. Chin Med J (Engl), 2022, 135(5):584-590.
[2]	VAN DEN BROEK J J, VAN DER WOLF F S W, HEIJNEN L A, et al. The prognostic importance of MRI detected extramural vascular invasion (mrEMVI) in locally advanced rectal cancer[J]. Int J Colorectal Dis, 2020, 35(10):1849-1854. doi: 10.1007/s00384-020-03632-9 pmid: 32488420
[3]	SCHAAP D P, VOOGT E L K, BURGER J W A, et al. Prognostic implications of mri-detected emvi and tumor deposits and their response to neoadjuvant therapy in cT3 and cT4 rectal cancer[J]. Int J Radiat Oncol Biol Phys, 2021, 111(3):816-825.
[4]	TAN J J, CARTEN R V, BABIKER A, et al. Prognostic importance of MRI-detected extramural venous invasion in rectal cancer: a literature review and systematic meta-analysis[J]. Int J Radiat Oncol Biol Phys, 2021, 111(2):385-394.
[5]	杨鋆, 辛城霖, 张忠涛. 中低位直肠癌的精准诊断与规范治疗[J]. 中华消化外科杂志, 2024, 23(1):85-90.
	YANG Y, XIN C L, ZHANG Z T. Precision diagnosis and standardized treatment of mid-low rectal cancer[J]. Chin J Dig Surg, 2024, 23(1):85-90.
[6]	中华医学会外科学分会腹腔镜与内镜外科学组, 中华医学会外科学分会结直肠外科学组, 中国医师协会外科医师分会结直肠外科专家工作组, 等. 腹腔镜结直肠癌根治术操作指南(2023版)[J]. 中华消化外科杂志, 2024, 23(1):10-22.
	Laparoscopic & Endoscopic Surgery Group, Branch of Surgery, Chinese Medical Association; Colorectal Surgery Group, Branch of Surgery, Chinese Medical Association; Chinese Society of Colon and Rectal Surgeons, Chinese Medical Doctor Association, et al. Guideline for ope-rative procedure of laparoscopic radical surgery for colorectal cancer (2023 edition)[J]. Chin J Dig Surg, 2024, 23(1):10-22.
[7]	肖体先, 侯文运, 梅世文, 等. 早发性结直肠癌的临床病理特征[J]. 中华消化外科杂志, 2023, 22(12):1476-1483.
	XIAO T X, HOU W Y, MEI S W, et al. Clinical characteristics of early-onset colorectal cancer[J]. Chin J Dig Surg, 2023, 22(12):1476-1483.
[8]	应俊, 孙亚煌, 王安琪, 等. 基于SEER数据库直肠癌肝转移预后列线图预测模型的构建及其应用价值[J]. 中华消化外科杂志, 2023, 22(S1):51-57.
	YING J, SUN Y H, WANG A Q, et al. Construction and application value of nomogram predictive model for the prognosis of rectal cancer liver metastases based on SEER database[J]. Chin J Dig Surg, 2023, 22(S1):51-57.
[9]	SMITH N J, BARBACHANO Y, NORMAN A R, et al. Prognostic significance of magnetic resonance imaging-detected extramural vascular invasion in rectal cancer[J]. Br J Surg, 2008, 95(2):229-236. doi: 10.1002/bjs.5917 pmid: 17932879
[10]	BAE J S, KIM S H, HUR B Y, et al. Prognostic value of MRI in assessing extramural venous invasion in rectal cancer: multi-readers' diagnostic performance[J]. Eur Radiol, 2019, 29(8):4379-4388. doi: 10.1007/s00330-018-5926-9 pmid: 30617483
[11]	BROWN G, RADCLIFFE A G, NEWCOMBE R G, et al. Preoperative assessment of prognostic factors in rectal cancer using high-resolution magnetic resonance imaging[J]. Br J Surg, 2003, 90(3):355-364.
[12]	LAMBIN P, LEIJENAAR R T H, DEIST T M, et al. Radiomics: the bridge between medical imaging and persona-lized medicine[J]. Nat Rev Clin Oncol, 2017, 14(12):749-762.
[13]	YU X, SONG W, GUO D, et al. Preoperative prediction of extramural venous invasion in rectal cancer: comparison of the diagnostic efficacy of radiomics models and quantitative dynamic contrast-enhanced magnetic resonance imaging[J]. Front Oncol, 2020,10:459.
[14]	LIU S, YU X, YANG S, et al. Machine learning-based radiomics nomogram for detecting extramural venous invasion in rectal cancer[J]. Front Oncol, 2021,11:610338.
[15]	SHU Z, MAO D, SONG Q, et al. Multiparameter MRI-based radiomics for preoperative prediction of extramural venous invasion in rectal cancer[J]. Eur Radiol, 2022, 32(2):1002-1013.
[16]	KANDA T, MATSUDA M, OBA H, et al. Gadolinium deposition after contrast-enhanced MR imaging[J]. Radio-logy, 2015, 277(3):924-925.
[17]	DUAN C, KALLEHAUGE J F, BRETTHORST G L, et al. Are complex DCE-MRI models supported by clinical data?[J]. Magn Reson Med, 2017, 77(3):1329-1339. doi: 10.1002/mrm.26189 pmid: 26946317
[18]	国家卫生健康委员会医政司, 中华医学会肿瘤学分会. 国家卫健委中国结直肠癌诊疗规范(2023版)[J]. 中国实用外科杂志, 2023, 43(6):602-630.
	Hospital Authority of National Health Commission ofthe People's Republic of China; Chinese Society ofOncology, Chinese Medical Association. Chinese protocol of diagnosis and treatment of colorectal cancer of the National Health Commission(2023 edition)[J]. Chin J Pract Surg, 2023, 43(6):602-630.
[19]	TALBOT I C, RITCHIE S, LEIGHTON M, et al. Invasion of veins by carcinoma of rectum: method of detection, histological features and significance[J]. Histopathology, 1981, 5(2):141-163. doi: 10.1111/j.1365-2559.1981.tb01774.x pmid: 7216178
[20]	BATES D D B, HOMSI M E, CHANG K J, et al. MRI for rectal cancer: staging, mrCRM, EMVI, lymph node staging and post-treatment response[J]. Clin Colorectal Cancer, 2022, 21(1):10-18.
[21]	GLYNNE-JONES R, WYRWICZ L, TIRET E, et al. Rectal cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up[J]. Ann Oncol, 2017, 28(suppl_4):iv22-iv40.
[22]	ZHAO L, LIANG M, WANG S, et al. Preoperative evaluation of extramural venous invasion in rectal cancer using radiomics analysis of relaxation maps from synthetic MRI[J]. Abdom Radiol (NY). 2021 Aug; 46(8):3815-3825. doi: 10.1007/s00261-021-03021-y pmid: 33743017
[23]	WANG H, CHEN X, DING J, et al. Novel multiparametric MRI-based radiomics in preoperative prediction of perirectal fat invasion in rectal cancer[J]. Abdom Radiol (NY), 2023, 48(2):471-485.
[24]	ROY S, MAJI P. Multispectral co-occurrence of wavelet coefficients for malignancy assessment of brain tumors[J]. PLoS One, 2021, 16(6):e0250964.
[25]	PENG Y, LUO Y, HU X, et al. Quantitative T2-weighted imaging and reduced field-of-view diffusion-weighted imaging of rectal cancer: correlation of R2 and apparent diffusion coefficient with histopathological prognostic factors[J]. Front Oncol, 2021,11:670156.
[26]	AO W, ZHANG X, YAO X, et al. Preoperative prediction of extramural venous invasion in rectal cancer by dynamic contrast-enhanced and diffusion weighted MRI: a preliminary study[J]. BMC Med Imaging, 2022, 22(1):78. doi: 10.1186/s12880-022-00810-9 pmid: 35484509
[27]	吴德生, 梁烨鑫, 陈秀婵, 等. 高分辨T2WI在直肠癌壁外血管侵犯诊断中的应用价值[J]. 中国中西医结合影像学杂志, 2020, 18(6):548-551.
	WU D S, LIANG Y X, CHEN X C, et al. Application value of high-resolution MRI T2WI in extramural vascular invasion of rectal cancer[J]. Chin Imag J Integr Tradit West Med, 2020, 18(6):548-551.
[28]	张益飞, 李月玥, 杨彦松, 等. 基于高分辨率T2WI的影像组学列线图预测直肠癌脉管侵袭的研究[J]. 中国临床医学影像杂志, 2021, 32(7):500-505.
	ZHANG Y F, LI Y Y, YANG Y S, et al. High resolution T2WI-based radiomics nomogram for prediction of lymphovascular invasion in rectal cancer[J]. J China Clin Med Imag, 2021, 32(7):500-505.
[29]	王可欣, 余静, 徐青. 基于RESOLVE ADC的影像组学列线图在预测直肠癌壁外血管侵犯中的应用价值[J]. 肿瘤影像学, 2023, 32(2):138-147.
	WANG K X, YU J, XU Q. Radiomics based on RESOLVE ADC in identification of extramural venous invasion in rectal cancer[J]. Oncoradiol, 2023, 32(2):138-147.

Equipment	Parameters Plane	T2WI			DWI Axial
Equipment	Parameters Plane	Axial oblique	Sagittal	Coronal	DWI Axial
Siemens Verio 3.0T	TR/TE,ms	4 000/97	4 000/97	4 000/97	9 700/93
	FOV,mm	240×240	220×220	220×220	280×350
	Thickness,mm	3	3	3	3
	b values	-	-	-	0,800,1 500
Siemens Avanto 1.5T	TR/TE,ms	4 120/97	3 940/85	4 990/96	4 912/95
	FOV,mm	200×200	250×250	240×240	250×250
	Thickness,mm	2.5	4	4	5
	b values	-	-	-	50,400,800

Characteristics	Training set （n=123）	Validation set (n=45)	t/χ²	P value
Age（years）	63.82±10.38	64.47±10.96	0.352	0.726
Gender（%）			0.164	0.686
Male	78（63.4）	27（60.0）
Female	45（36.6）	18（40.0）
CEA（%）			0.323	0.570
≤5 ng/mL	77(62.6)	26(57.8)
>5 ng/mL	46(37.4)	19(42.2)
ADC Value （×10^-3mm²/s）	0.81(0.72,0.91)	0.78(0.71,0.87)	-1.091	0.275
Infiltration depth （mm）	15.34±5.73	16.61±8.09	1.130	0.260
Length,（cm）	43.14±15.78	44.50±15.56	0.496	0.621
mrT stage（%）			0.235	0.628
T1~2	46(37.4)	15(33.3)
T3~4	77(62.6)	30(66.7)
mrEMVI（%）			0.362	0.547
Negative	80(65.0)	27(60.0)
Positive	43(35.0)	18(40.0)
pEMVI（%）			0.016	0.900
Negative	89(72.4)	33(73.3)
Positive	34(27.6)	12(26.7)
Radscore	-2.71(-3.87,-1.32)	-2.09(-3.87,-1.20)	-1.067	0.286

Characteristics	Training set			Validation set
Characteristics	EMVI（-）（n=89）	EMVI（+）(n=34)	P value	EMVI（-）(n=33)	EMVI（+）(n=12)	P value
Age（years）	64.31±10.42	62.53±10.30	0.396	63.7±12.1	66.7±6.9	0.423
Gender（%）			0.814			0.063
Male	32(36.0)	13(38.2)		10(30.3)	8(66.7)
Female	57(64.0)	21(61.8)		23(69.7)	4(33.3)
CEA (%)			0.171			0.097
≤5 ng/mL	59(66.3)	18(52.9)		22(66.7)	4(33.3)
>5 ng/mL	30(33.7)	16(47.1)		11(33.3)	8(66.7)
ADC Value（×10^-3mm²/s）	0.85(0.74,0.99)	0.75(0.61,0.80)	<0.001	0.82(0.72,0.93)	0.75(0.68,0.79)	0.066
Infiltration Depth（mm）	14.63±6.03	17.20±4.39	0.011	14.9±5.1	21.4±12.3	0.101
Length,（cm）	41.18±15.51	48.27±15.54	0.025	43.3±14.2	47.9±19.0	0.387
Location (%)			0.683			0.060^#
Upper	27(30.3)	8(23.5)		11(33.3)	0(0.0)
Middle	35(39.3)	16(47.1)		12(36.4)	6(50.0)
Low	27(30.3)	10(29.4)		10(30.3)	6(50.0)
mrT stage（%）			<0.001			0.074
T1-2	42(47.2)	4(11.8)		14(42.4)	1(8.3)
T3-4	47(52.8)	30(88.2)		19(57.6)	11(91.7)
mrEMVI（%）			<0.001			<0.001
Negative	71(79.8)	9(26.5)		26(78.8)	1 (8.3)
Positive	18(20.2)	25(73.5)		7 (21.2)	11(91.7)
Radscore	-3.14(-4.44,-2.08)	-1.40(-2.69,-0.69)	<0.001	-2.46(-4.19,-1.39)	-1.25(-1.74,-0.18)	0.004

Variables	Univariate		Multivariate（clinical model）		Multivariate（combined model）
Variables	OR(95% CI)	P value	OR(95% CI)	P value	OR(95% CI)	P value
Age	0.983（0.946~1.022）	0.393
Gender
Female
Male	0.907（0.401~2.051）	0.814
CEA
≤5 ng/mL
>5 ng/mL	1.748（0.782~3.907）	0.173
ADC	0.000（0.000~0.012）	<0.001	0.000（0.000~0.016）	<0.001	0.000（0.000~0.016）	0.001
Infiltration depth	1.085（1.008~1.167）	0.029
Length	1.030（1.003~1.058））	0.029
Location
upper
middle	1.543（0.576~4.136）	0.389
low	1.250（0.428~3.651）	0.683
mrT stage
T₁~T₂
T₃~T₄	6.702（2.180~20.607）	0.001	2.869(0.778~10.574)	0.113	3.899（0.940~16.172）	0.061
mrEMVI
Negative
Positive	10.957（4.363~27.518）	<0.001	8.643(2.886~25.886)	<0.001	7.928（2.397~26.221）	0.001
Radscore	1.862（1.336~2.596）	<0.001			2.048（1.301~3.226）	0.002

Models	Training set			Validation set
Models	AUC（95% CI）	Sensitivity	Specificity	AUC（95% CI）	Sensitivity	Specificity
Radiomics model	0.756（0.656~0.855）	0.676	0.787	0.782(0.626~0.937）	0.833	0.697
Clinical model	0.888（0.829~0.948）	0.824	0.865	0.896（0.753~1.000）	0.917	0.939
Combined model	0.926（0.879~0.973）	0.882	0.865	0.917（0.813~1.000）	0.917	0.909