基于CT的影像组学诺模图预测结直肠癌肺转移射频消融后的局部肿瘤进展

doi:10.19401/j.cnki.1007-3639.2024.09.006

摘要/Abstract

摘要：

背景与目的： 早期预测结直肠癌（colorectal cancer，CRC）肺转移瘤经过射频消融（radiofrequency ablation，RFA）治疗后的局部肿瘤无进展生存期（local tumor progression-free survival，LTPFS）具有重要的临床意义。影像组学在肿瘤预后预测方面已有一定的探索。本研究旨在构建基于影像组学的诺模图预测CRC肺转移患者RFA治疗后的LTPFS。方法： 本研究回顾性分析2016年8月—2019年1月复旦大学附属肿瘤医院介入治疗科收治的172例伴有401个肺转移病灶的CRC患者。本研究通过复旦大学附属肿瘤医院医学伦理委员会的审查（伦理编号：2402291-24）。将患者的RFA术前及术后即刻计算机体层成像（computed tomography，CT）图像通过数据增强后，人工分割靶病灶和RFA后区域后提取影像组学特征。采用最大相关最小冗余算法（maximum relevance and minimum redundancy algorithm，MRMRA）和最小绝对值收敛（least absolute shrinkage and selection operator，LASSO）回归模型进行特征筛选。基于筛选后的影像组学特征和多因素分析筛选的临床变量，分别构建临床模型、影像组学模型和融合模型。计算一致性指数（concordance index，C-index）和受试者工作特征（receiver operating characteristic，ROC）曲线的曲线下面积（area under curve，AUC）来评估3种模型的预测性能，最后绘制最佳模型对应的诺模图。 结果： 在所有肺转移瘤中，最终复发的病灶有102个（25.4%），完全缓解（complete response，CR）的病灶有299个（74.6%）。中位随访时间为21个月（95% CI：19.466 ~ 22.534），RFA术后的1、2和3年的LTPFS率分别为76.5%（95% CI：72.0 ~ 80.4）、72.1%（95% CI：66.6 ~ 76.9）和69.9%（95% CI：64.0 ~ 75.1）。无论在训练集还是测试集中，基于LASSO回归模型最终筛选的12个影像组学特征和多因素分析筛选的4个临床变量而构建的融合模型预测LTPFS的AUC最高，C-index分别为0.890（95% CI：0.854 ~ 0.927）和0.843（95% CI：0.768 ~ 0.916）。结论： 基于影像组学特征和临床变量的融合模型预测CRC肺转移患者RFA治疗后的LTPFS是可行的，并且其性能优于单一影像组学模型和临床模型。同时，融合模型的诺模图可以更直观地预测CRC肺转移患者RFA治疗的预后，协助临床医师为患者制订个体化的随访复查方案，灵活地调整治疗策略。

关键词: 计算机断层扫描, 影像组学, 诺模图, 结直肠癌, 肺转移, 射频消融

Abstract:

Background and Purpose: The early prediction of local tumor progression-free survival (LTPFS) after radiofrequency ablation (RFA) for colorectal cancer (CRC) lung metastases has significant clinical importance. The application of radiomics in the prediction of tumor prognosis has been explored. This study aimed to construct a radiomics-based nomogram for predicting LTPFS after RFA in CRC patients with lung metastases. Methods: This study retrospectively analyzed 172 CRC patients with 401 lung metastases admitted to Department of Interventional Radiology, Fudan University Shanghai Cancer Center from August 2016 to January 2019. This study was reviewed by the medical ethics committee of Fudan University Shanghai Cancer Center (ethics number: 2402291-24). After augmentation of pre-ablation and immediate post-ablation computed tomography (CT) images, the target metastases and ablation regions were segmented manually to extract the radiomic features. Maximum relevance and minimum redundancy algorithm (MRMRA) and least absolute shrinkage and selection operator (LASSO) regression models were applied for feature selection. The clinical model, the radiomics model, and the fusion model were constructed based on the selected radiomic features and clinical variables screened by the multivariate analysis. The Harrell concordance index (C-index) and area under receiver operating characteristic (ROC) curves (AUC) were calculated to evaluate the prediction performance. Finally, the corresponding nomogram of the best model was drawn. Results: Among all the lung metastases, 102 (25.4%) had final recurrence, and 299 (74.6%) had complete response (CR). The median follow-up time was 21 months (95% CI: 19.466-22.534), and the LTPFS rates at 1, 2, and 3 years after RFA were 76.5% (95% CI: 72.0-80.4), 72.1% (95% CI: 66.6-76.9) and 69.9% (95% CI: 64.0-75.1). In both the training and test dataset, the fusion model based on the final 12 radiomic features through the LASSO regression and 4 clinical variables screened by multivariate analysis achieved the highest AUC values for LTPFS, with C-index values of 0.890 (95% CI: 0.854-0.927) and 0.843 (95% CI: 0.768-0.916), respectively. Conclusion: The fusion model based on radiomic features and clinical variables is feasible for predicting LTPFS after RFA of CRC patients with lung metastases, whose performance is superior to the single radiomic and clinical model. At the same time, the nomogram of the fusion model can intuitively predict the prognosis of CRC patients with lung metastases after RFA, thus assisting clinicians in developing individualized follow-up review plans for patients and adjusting treatment strategies flexibly.

Key words: Computed tomography, Radiomics, Nomogram, Colorectal cancer, Lung metastases, Radiofrequency ablation

中图分类号:

黄浩哲, 陈红, 郑德重, 陈超, 王英, 许立超, 王耀辉, 何新红, 杨媛媛, 李文涛. 基于CT的影像组学诺模图预测结直肠癌肺转移射频消融后的局部肿瘤进展[J]. 中国癌症杂志, 2024, 34(9): 857-872.

HUANG Haozhe, CHEN Hong, ZHENG Dezhong, CHEN Chao, WANG Ying, XU Lichao, WANG Yaohui, HE Xinhong, YANG Yuanyuan, LI Wentao. A CT-based radiomics nomogram for predicting local tumor progression of colorectal cancer lung metastases treated with radiofrequency ablation[J]. China Oncology, 2024, 34(9): 857-872.

图/表 11

图1

图2

表1

图3

表2

图4

表3

图5

表4

图6

图7

参考文献 33

[1]	SUNG H, FERLAY J, SIEGEL R L, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries[J]. CA Cancer J Clin, 2021, 71(3): 209-249.
[2]	NORDHOLM-CARSTENSEN A, KRARUP P M, JORGENSEN L N, et al. Occurrence and survival of synchronous pulmonary metastases in colorectal cancer: a nationwide cohort study[J]. Eur J Cancer, 2014, 50(2): 447-456.
[3]	SIEGEL R L, MILLER K D, JEMAL A. Cancer statistics, 2019[J]. CA, 2019, 69(1): 7-34.
[4]	CHO J H, KIM S, NAMGUNG M, et al. The prognostic importance of the number of metastases in pulmonary metastasectomy of colorectal cancer[J]. World J Surg Oncol, 2015, 13: 222. doi: 10.1186/s12957-015-0621-7 pmid: 26205014
[5]	IBRAHIM T, TSELIKAS L, YAZBECK C, et al. Systemic versus local therapies for colorectal cancer pulmonary metastasis: what to choose and when?[J]. J Gastrointest Cancer, 2016, 47(3): 223-231. doi: 10.1007/s12029-016-9818-4 pmid: 27080402
[6]	QI H, FAN W J. Value of ablation therapy in the treatment of lung metastases[J]. Thorac Cancer, 2018, 9(2): 199-207. doi: 10.1111/1759-7714.12567 pmid: 29193688
[7]	YU W S, BAE M K, CHOI J K, et al. Pulmonary metastasectomy in colorectal cancer: a population-based retrospective cohort study using the Korean national health insurance database[J]. Cancer Res Treat, 2021, 53(4): 1104-1112. doi: 10.4143/crt.2020.1213 pmid: 33494126
[8]	NAJAFI A, BAERE T D, PURENNE E, et al. Risk factors for local tumor progression after RFA of pulmonary metastases: a matched case-control study[J]. Eur Radiol, 2021, 31(7): 5361-5369. doi: 10.1007/s00330-020-07675-y pmid: 33474569
[9]	CERVANTES A, ADAM R, ROSELLÓ S, et al. Metastatic colorectal cancer: ESMO clinical practice guideline for diagnosis, treatment and follow-up[J]. Ann Oncol, 2023, 34(1): 10-32.
[10]	YANG Q X, QI H, ZHANG R, et al. Risk factors for local progression after percutaneous radiofrequency ablation of lung tumors: evaluation based on a review of 147 tumors[J]. J Vasc Interv Radiol, 2017, 28(4): 481-489. doi: S1051-0443(16)30842-9 pmid: 28111196
[11]	DOO F X, VOSSHENRICH J, COOK T S, et al. Environmental sustainability and AI in radiology: a double-edged sword[J]. Radiology, 2024, 310(2): e232030.
[12]	BRANCATO V, ESPOSITO G, COPPOLA L, et al. Standardizing digital biobanks: Integrating imaging, genomic, and clinical data for precision medicine[J]. J Transl Med, 2024, 22(1): 136. doi: 10.1186/s12967-024-04891-8 pmid: 38317237
[13]	TERRANOVA N, VENKATAKRISHNAN K. Machine learning in modeling disease trajectory and treatment outcomes: an emerging enabler for model-informed precision medicine[J]. Clin Pharmacol Ther, 2024, 115(4): 720-726.
[14]	BEIG N, KHORRAMI M, ALILOU M, et al. Perinodular and intranodular radiomic features on lung CT images distinguish adenocarcinomas from granulomas[J]. Radiology, 2019, 290(3): 783-792. doi: 10.1148/radiol.2018180910 pmid: 30561278
[15]	ABTIN F G, ERADAT J, GUTIERREZ A J, et al. Radiofrequency ablation of lung tumors: imaging features of the postablation zone[J]. Radiographics, 2012, 32(4): 947-969. doi: 10.1148/rg.324105181 pmid: 22786987
[16]	刘宝东, 支修益. 影像引导下热消融治疗肺部肿瘤的局部疗效评价[J]. 中国医学前沿杂志(电子版), 2015, 7(2): 11-14.
[17]	LIU B D, ZHI X Y. Evaluation of local curative effect of image-guided thermal ablation for lung tumors[J]. Chin J Front Med Sci Electron Version, 2015, 7(2): 11-14.
[18]	STARMANS M P A, VAN DER VOORT S R, PHIL T, et al. Reproducible radiomics through automated machine learning validated on twelve clinical applications[EB/OL]. (2022-07-29) [2024-05-07]. http://arxiv.org/abs/2108.08618
[19]	RIGATTI S J. Random forest[J]. J Insur Med, 2017, 47(1): 31-39. doi: 10.17849/insm-47-01-31-39.1 pmid: 28836909
[20]	QIAN L Q, ZHOU Y J, ZENG W Q, et al. A random forest algorithm predicting model combining intraoperative frozen section analysis and clinical features guides surgical strategy for peripheral solitary pulmonary nodules[J]. Transl Lung Cancer Res, 2022, 11(6): 1132-1144.
[21]	KHUSHI M, SHAUKAT K, ALAM T M, et al. A comparative performance analysis of data resampling methods on imbalance medical data[J]. IEEE Access, 2021, 9: 109960-109975.
[22]	HALL D L, LLINAS J. An introduction to multisensor data fusion[J]. Proc IEEE, 1997, 85(1): 6-23.
[23]	RASHINKAR P, KRUSHNASAMY V S. An overview of data fusion techniques[C]. Bengaluru: ICIMIA, 2017: 694-697.
[24]	MARKICH R, PALUSSIÈRE J, CATENA V, et al. Radiomics complements clinical, radiological, and technical features to assess local control of colorectal cancer lung metastases treated with radiofrequency ablation[J]. Eur Radiol, 2021, 31(11): 8302-8314. doi: 10.1007/s00330-021-07998-4 pmid: 33954806
[25]	BAÈRE T D, AUPÉRIN A, DESCHAMPS F, et al. Radiofrequency ablation is a valid treatment option for lung metastases: experience in 566 patients with 1037 metastases[J]. Ann Oncol, 2015, 26(5): 987-991. doi: S0923-7534(19)31500-5 pmid: 25688058
[26]	GONZALEZ M, PONCET A, COMBESCURE C, et al. Risk factors for survival after lung metastasectomy in colorectal cancer patients: a systematic review and meta-analysis[J]. Ann Surg Oncol, 2013, 20(2): 572-579. doi: 10.1245/s10434-012-2726-3 pmid: 23104709
[27]	PALUSSI&#XE; RE J, MARCET B, DESCAT E, et al. Lung tumors treated with percutaneous radiofrequency ablation: computed tomography imaging follow-up[J]. Cardiovasc Intervent Radiol, 2011, 34(5): 989-997. doi: 10.1007/s00270-010-0048-z pmid: 21127867
[28]	MAPELLI P, BEZZI C, MUFFATTI F, et al. Preoperative assessment of lymph nodal metastases with[⁶⁸Ga]Ga-DOTATOC PET radiomics for improved surgical planning in well-differentiated pancreatic neuroendocrine tumours[J]. Eur J Nucl Med Mol Imaging, 2024, 51(9): 2774-2783.
[29]	ZHU F D, YANG C, XIA Y, et al. CT-based radiomics models may predict the early efficacy of microwave ablation in malignant lung tumors[J]. Cancer Imaging, 2023, 23(1): 60. doi: 10.1186/s40644-023-00571-w pmid: 37308918
[30]	HUANG H Z, CHEN H, ZHENG D Z, et al. Habitat-based radiomics analysis for evaluating immediate response in colorectal cancer lung metastases treated by radiofrequency ablation[J]. Cancer Imaging, 2024, 24(1): 44. doi: 10.1186/s40644-024-00692-w pmid: 38532520
[31]	BENSON A B, VENOOK A P, AL-HAWARY M M, et al. Colon cancer, version 2.2021, NCCN clinical practice guidelines in oncology[J]. J Natl Compr Canc Netw, 2021, 19(3): 329-359.
[32]	DUFFY M J, LAMERZ R, HAGLUND C, et al. Tumor markers in colorectal cancer, gastric cancer and gastrointestinal stromal cancers: European group on tumor markers 2014 guidelines update[J]. Int J Cancer, 2014, 134(11): 2513-2522. doi: 10.1002/ijc.28384 pmid: 23852704
[33]	LIU B, LI C H, SUN X R, et al. Assessment and prognostic value of immediate changes in post-ablation intratumor density heterogeneity of pulmonary tumors via radiomics-based computed tomography features[J]. Front Oncol, 2021, 11: 615174.

Characteristic	Training dataset (N=323)	Test dataset (N=78)	P value	Total (N=401)
Pre-RFA clinical characteristics
Gender n			0.947
Male	185	45		230
Female	138	33		171
Age/year M (Q1, Q3)	57 (50, 65)	61 (52, 66.25)	0.120
Initial tumor location n			0.166
Rectum	224	44		268
Sigmoid-left colon	40	15		55
Transverse-right colon	57	18		75
Caecum	2	1		3
Tumor biomarkers M (Q1, Q3)
CEA/(ng·mL^-1)	4.39 (2.07, 13.33)	3.96 (2.08, 11.38)	0.628
CA 19-9/(U·mL^-1)	10.71 (7.58, 19.50)	12.60 (6.83, 23.25)	0.928
Number of recurrence n			0.351
At 1 year	75	17		92
At 2 years	82	17		99
At 3 years	85	17		102
Follow-time/month M (95% CI)	25 (22.49, 27.51)	19 (18.43, 19.57)
Pre-RFA characteristics of lung metastases
Nodule size/mm n			0.064
<10	156	35		191
10-19	118	31		149
20-30	49	12		61
Location n			0.183
RUL	76	15		91
RML	36	8		44
RLL	63	10		73
LUL	77	18		95
LLL	71	27		98
Distance 1/cm n			0.454
≥1	266	67		333
<1	57	11		68
Distance 2/cm n			0.399
≥1	124	34		158
<1	199	44		243
Immediate post-RFA characteristics
Immediate pneumothorax n			0.851
Yes	82	19		101
No	241	59		300
Intra-alveolar hemorrhage n			0.778
Yes	86	22		108
No	237	56		293
Electrode chosen for RFA n			0.722
Expandable	303	74		377
Straight	20	4		24

Characteristic	Univariate analysis		Multivariate analysis
Characteristic	HR (95% CI)	P value	HR (95% CI)	P value
Clinical features
Gender
Male	1.055 (0.687-1.620)	0.806
Female	1
Age	1.003 (0.983-1.023)	0.748
Initial tumor location
Rectum	1
Sigmoid-left colon	0.663 (0.318-1.383)	0.273
Transverse-right colon	0.950 (0.532-1.695)	0.862
Caecum^*	-	-
Lymphadenopathy at diagnosis
Yes	1.034 (0.659-1.621)	0.884
No	1
Systemic therapy
Yes	0.817 (0.524-1.275)	0.374
No	1
Tumor biomarkers
CEA/(ng·mL^-1)	1.004 (1.001-1.007)	0.002	1.052 (1.021-1.084)	0.001
CA 19-9/(U·mL^-1)	1.003 (1.000-1.007)	0.062	1.047 (1.008-1.088)	0.019
Pre-RFA features of the lung metastases
Location
RUL	1.035 (0.506-2.112)	0.925	1.327 (0.660-2.668)	0.427
RML	1.612 (0.723-3.594)	0.243	1.645 (0.760-3.564)	0.207
RLL	2.588 (1.358-4.929)	0.004	3.055 (1.629-5.732)	0.001
LUL	1		1
LLL	1.732 (0.892-3.363)	0.104	2.236 (1.192-4.193)	0.012
Distance 1/cm
≥1	1
<1	1.326 (0.789-2.230)	0.287
Distance 2/cm
≥1	1
<1	1.206 (0.774-1.879)	0.409
Immediate post-RFA features
Pneumothorax
Yes	1.207 (0.758-1.923)	0.428
No	1
Hemorrhage
Yes	0.812 (0.492-1.339)	0.414
No	1
Electrode
Expandable	0.480 (0.240-0.958)	0.037	2.171 (1.176-4.005)	0.013
Straight	1		1

Pre-RFA radiomics feature	Post-RFA radiomics feature
Shape_Elongation	Shape_Sphericity
Shape_MeshVolume	GLCM_Idn^*
FirstOrder_Energy^*	GLCM_Idmn
FirstOrder_Entropy^*	GLDM_SmallDependenceLowGrayLevelEmphasis^*
FirstOrder_MeanAbsoluteDeviation^*	GLSZM_HighGrayLevelZoneEmphasis
FirstOrder_Skewness	NGTDM_Strength
FirstOrder_RootMeanSquared	NGTDM_Coarseness^*
FirstOrder_RobustMeanAbsoluteDeviation
GLCM_ClusterShade^*
GLCM_ClusterProminence
GLCM_Idmn
GLCM_Imc1^*
GLCM_InverseVariance^*
GLDM_DependenceEntropy^*
GLDM_GrayLevelNonUniformity^*
GLSZM_LargeAreaEmphasis
GLSZM_LargeAreaLowGrayLevelEmphasis^*

Variable	Fusion model	Radiomic model	Clinical model
Training dataset
C-index (95% CI)	0.890 (0.854-0.927)	0.867 (0.829-0.906)	0.665 (0.594-0.737)
Brier score (1-year)	0.092	0.106	0.161
Brier score (2-year)	0.160	0.189	0.180
Brier score (3-year)	0.161	0.198	0.177
AUC (1-year)	0.917	0.899	0.670
AUC (2-year)	0.869	0.850	0.668
AUC (3-year)	0.857	0.839	0.659
Test dataset
C-index (95% CI)	0.843 (0.768-0.916)	0.811 (0.722-0.897)	0.688 (0.565-0.811)
Brier score (1-year)	0.126	0.133	0.172
Brier score (2-year)	0.204	0.241	0.191
Brier score (3-year)	0.174	0.188	0.187
AUC (1-year)	0.857	0.810	0.711
AUC (2-year)	0.709	0.618	0.687
AUC (3-year)	0.757	0.674	0.655