基于胸部增强CT影像组学模型用于胸腺瘤分类的研究

doi:10.19401/j.cnki.1007-3639.2024.06.006

摘要/Abstract

摘要：

背景与目的：胸腺瘤术前分类对治疗决策很有帮助，但获取存在一定困难。本研究旨在建立基于增强对比计算机体层成像（contrast-enhanced computed tomography，CECT）影像组学的胸腺瘤风险分类训练模型，并验证其性能、可靠性和泛化能力。 方法：本回顾性队列研究分析了在上海交通大学医学院附属胸科医院2008年1月—2017年12月接受胸腺瘤切除手术的患者（Masaoka-Koga Ⅰ~Ⅲ期）的临床资料。将患者随机分层为训练组（80%）和测试组（20%），将术后病理学检查结果作为金标准。低风险组织学类型包括A、AB和B1。高风险组织学类型包括B2和B3。手动分割术前CECT图像相关肿瘤区域并提取影像组学特征。使用最小绝对收敛和选择算子（least absolute shrinkage and selection operator，LASSO）回归进行特征选择，并将临床特征添加到联合模型中。模型性能指标包括受试者工作特征（receiver operating characteristic，ROC）曲线的曲线下面积（area under curve，AUC）、灵敏度和特异度。结果：共纳入478例患者（平均年龄51.3±12.3岁，男性占48.1%）。临床模型、基于CECT的影像组学模型和在测试集上使用临床和CT特征的模型AUC分别为0.666、0.831和0.850。性能最佳的模型的灵敏度为0.829，特异度为0.764。结论：基于术前CECT的影像组学模型在胸腺瘤风险分类中表现良好。

关键词: 纵隔腔, 胸腺瘤, 计算机体层成像, 机器学习

Abstract:

Background and purpose: Preoperative risk categorization of thymoma is useful for treatment decisions but remains challenging. This study focused on training radiomics models using contrast-enhanced computed tomography (CECT) images for thymoma risk categorization and validating the model's performance, reliability and generalizability in a relatively large cohort. Methods: This retrospective cohort study analyzed the clinical data of thymoma patients (Masaoka Koga Ⅰ-Ⅲ) who underwent thymectomy surgery at the Affiliated Chest Hospital of Shanghai Jiao Tong University School of Medicine from January 2008 to December 2017. The cohort was divided into a training group (80%) and a test group (20%) using stratified random selection. The gold standard for histologic types was based on surgically resected specimens. Low-risk histologic types included A, AB and B1. High-risk histologic types included B2 and B3. Radiomics features were extracted from manually segmented regions of interest on preoperative CECT images. Interobserver correlation and least absolute shrinkage and selection operator (LASSO) regression were used for feature selection. Model performance metrics included area under the curve (AUC) of receiver operating characteristic (ROC) curve, sensitivity and specificity. Clinical characteristics were added to the combined model. Results: A total of 478 patients (mean age 51.3±12.3 years, 48.1% was male) were included. The AUC of the clinical model, the CECT-based model, and the model using both clinical and CECT features on the test set were 0.666, 0.831 and 0.850, respectively. The best performing model had a sensitivity of 0.829 and a specificity of 0.764. Conclusion: CECT-based radiomics models showed good performance in risk categorization of thymomas.

Key words: Mediastinal, Thymoma, Computed tomography, Machine learning

中图分类号:

R734.2

蒋佻宴, 贾田颖, 张琴. 基于胸部增强CT影像组学模型用于胸腺瘤分类的研究[J]. 中国癌症杂志, 2024, 34(6): 581-589.

JIANG Tiaoyan, JIA Tianying, ZHANG Qin. Contrast-enhanced computed tomography-based radiomics models for the risk categorization of thymoma[J]. China Oncology, 2024, 34(6): 581-589.

图/表 6

图1

表1

表2

图2

图3

表3

参考文献 28

[1]	PRIOLA A M, PRIOLA S M, CARDINALE L, et al. The anterior mediastinum: diseases[J]. Radiol Med, 2006, 111(3): 312-342. doi: 10.1007/s11547-006-0032-5 pmid: 16683081
[2]	MARX A, CHAN J K, COINDRE J M, et al. The 2015 World Health Organization classification of tumors of the thymus: continuity and changes[J]. J Thorac Oncol, 2015, 10(10): 1383-1395. doi: 10.1097/JTO.0000000000000654 pmid: 26295375
[3]	CHEN G, MARX A, CHEN W H, et al. New WHO histologic classification predicts prognosis of thymic epithelial tumors: a clinicopathologic study of 200 thymoma cases from China[J]. Cancer, 2002, 95(2): 420-429. doi: 10.1002/cncr.10665 pmid: 12124843
[4]	GIRARD N, RUFFINI E, MARX A, et al. ESMO Guidelines Committee. Thymic epithelial tumours: ESMO clinical practice guidelines for diagnosis, treatment and follow-up[J]. Ann Oncol, 2015, 26(Suppl 5): v40-55.
[5]	SCORSETTI M, LEO F, TRAMA A, D'ANGELILLO R, et al. Thymoma and thymic carcinomas[J]. Crit Rev Oncol Hematol, 2016, 99: 332-350. doi: 10.1016/j.critrevonc.2016.01.012 pmid: 26818050
[6]	VENUTA F, RENDINA E A, LONGO F, et al. Long-term outcome after multimodality treatment for stage Ⅲ thymic tumors[J]. Ann Thorac Surg, 2003, 76(6): 1866-1872; discussion1872.
[7]	VENUTA F, RENDINA E A, PESCARMONA E O, et al. Multimodality treatment of thymoma: a prospective study[J]. Ann Thorac Surg, 1997, 64(6): 1585-1591; discussion: 1591-1592.
[8]	OTANI Y, YOSHIDA I, ISHIKAWA S, et al. Neoadjuvant intra-arterial infusion chemotherapy for invasive thymoma[J]. Oncol Rep, 1997, 4(1): 23-25. doi: 10.3892/or.4.1.23 pmid: 21590005
[9]	JEONG Y J, LEE K S, KIM J, et al. Does CT of thymic epithelial tumors enable us to differentiate histologic types and predict prognosis?[J]. AJR Am J Roentgenol, 2004, 183(2): 283-289.
[10]	LIU J, YIN P, WANG S C, et al. CT-based radiomics signatures for predicting the risk categorization of thymic epithelial tumors[J]. Front Oncol, 2021, 11: 628534.
[11]	FEDOROV A, BEICHEL R, KALPATHY-CRAMER J, et al. 3D Slicer as an image computing platform for the Quantitative Imaging Network[J]. Magn Reson Imaging, 2012, 30(9): 1323-1341. doi: 10.1016/j.mri.2012.05.001 pmid: 22770690
[12]	FRIEDMAN J, HASTIE T, TIBSHIRANI R. Regularization paths for generalized linear models via coordinate descent[J]. J Statistical Software, 33(1): 1-22.
[13]	BISCHL B, LANG M, KOTTHOFF L, et al. MLR: machine learning in R[J]. J Machine Learning Res, 2016, 17(1): 5938-5942.
[14]	ABDEL RAZEK A A, KHAIRY M, NADA N. Diffusion-weighted MR imaging in thymic epithelial tumors: correlation with world health organization classification and clinical staging[J]. Radiology, 2014, 273(1): 268-275. doi: 10.1148/radiol.14131643 pmid: 24877982
[15]	XIAO G, HU Y C, REN J L, et al. MR imaging of thymomas: a combined radiomics nomogram to predict histologic types[J]. Eur Radiol, 2021, 31(1): 447-457.
[16]	WANG X H, SUN W, LIANG H Y, et al. Radiomics signatures of computed tomography imaging for predicting risk categorization and clinical stage of thymomas[J]. Biomed Res Int, 2019, 2019: 3616852.
[17]	XIAO G, RONG W C, HU Y C, et al. MRI radiomics analysis for predicting the pathologic classification and TNM staging of thymic epithelial tumors: a pilot studyB[J]. AJR Am J Roentgenol, 2020, 214(2): 328-340.
[18]	DONG W T, XIONG S T, LEI P G, et al. Application of a combined radiomics nomogram based on CE-CT in the preoperative prediction of thymomas risk categorization[J]. Front Oncol, 2022, 12: 944005.
[19]	YU C, LI T, YANG X, ZHANG R, et al. Contrast-enhanced CT-based radiomics model for differentiating risk subgroups of thymic epithelial tumors[J]. BMC Med Imaging, 2022, 22(1): 37.
[20]	LIU W, WANG W, ZHANG H Y, et al. Development and validation of multi-omics thymoma risk classification model based on transfer learning[J]. J Digit Imaging, 2023, 36(5): 2015-2024. doi: 10.1007/s10278-023-00855-4 pmid: 37268842
[21]	SHANG L, WANG F, GAO Y, et al. Machine-learning classifiers based on non-enhanced computed tomography radiomics to differentiate anterior mediastinal cysts from thymomas and low-risk from high-risk thymomas: a multi-center study[J]. Front Oncol, 2022, 12: 1043163.
[22]	KAYI CANGIR A, ORHAN K, KAHYA Y, et al. CT imaging-based machine learning model: a potential modality for predicting low-risk and high-risk groups of thymoma: "impact of surgical modality choice"[J]. World J Surg Oncol, 2021, 19(1): 147.
[23]	OZKAN E, ORHAN K, SOYDAL C, et al. Combined clinical and specific positron emission tomography/computed tomography-based radiomic features and machine-learning model in prediction of thymoma risk groups[J]. Nucl Med Commun, 2022, 43(5): 529-539. doi: 10.1097/MNM.0000000000001547 pmid: 35234213
[24]	RAJAMOHAN N, GOYAL A, KANDASAMY D, et al. CT texture analysis in evaluation of thymic tumors and thymic hyperplasia: correlation with the international thymic malignancy interest group (ITMIG) stage and WHO grade[J]. Br J Radiol, 2021, 94(1128): 20210583.
[25]	IANNARELLI A, SACCONI B, TOMEI F, et al. Analysis of CT features and quantitative texture analysis in patients with thymic tumors: correlation with grading and staging[J]. Radiol Med, 2018, 123(5): 345-350. doi: 10.1007/s11547-017-0845-4 pmid: 29307077
[26]	FENG X L, WANG S Z, CHEN H H, et al. Optimizing the radiomics-machine-learning model based on non-contrast enhanced CT for the simplified risk categorization of thymic epithelial tumors: a large cohort retrospective study[J]. Lung Cancer, 2022, 166: 150-160.
[27]	NAKAJO M, TAKEDA A, KATSUKI A, et al. The efficacy of ¹⁸F-FDG-PET-based radiomic and deep-learning features using a machine-learning approach to predict the pathological risk subtypes of thymic epithelial tumors[J]. Br J Radiol, 2022, 95(1134): 20211050.
[28]	MAHMOUDI S, GRUENEWALD L D, EICHLER K, et al. Multiparametric evaluation of radiomics features and dual-energy CT iodine maps for discrimination and outcome prediction of thymic masses[J]. Acad Radiol, 2023, 30(12): 3010-3021. doi: 10.1016/j.acra.2023.03.034 pmid: 37105804

Characteristics	Low-risk thymomas (n=273)	High-risk thymomas (n=205)	P value
Gender n (%)
Male	122 (44.69)	107 (52.2)	0.10
Female	151 (55.31)	98 (47.8)
Age/year
x±s	54.5±10.8	50.9±13.7	0.01
Maximal diameter D/cm
x±s	6.18±2.97	6.07±2.57	0.98
Comorbidity n(%)
Present	9 (3.3)	36 (17.56)	<0.001
Absent	264 (96.70)	169 (82.44)
Masaoka-Koga stage n (%)
Ⅰ	110 (40.29)	28 (13.66)	<0.001
Ⅱ	148 (54.21)	117 (57.07)
Ⅲ	15 (5.49)	60 (29.27)

Characteristics	Training set (n=382)	Test set (n=96)	P value
Gender
Male	218 (57.07)	55 (57.29)	0.97
Female	164 (42.93)	41 (42.71)
Age/year
x±s	52.9±12.3	53.1±12.0	0.66
Maximal diameter D/cm
x±s	6.03±2.74	6.36±3.09	0.38
Concomitant disease n (%)
Present	35 (9.16)	10 (10.42)	0.71
Absent	347 (90.84)	86 (89.58)
Masaoka-Koga stage n(%)
Ⅰ	111 (29.06)	27 (28.13)	0.91
Ⅱ	211 (55.24)	54 (56.25)
Ⅲ	60 (15.71)	15 (15.63)
Risk category n (%)
Low risk	218 (57.07)	55 (57.29)	0.97
High risk	164 (42.93)	41 (42.71)

Data	AUC	95% CI	P value	Sensitivity	Specificity	Accuracy
Clinical	0.666	0.554-0.777	0.0028	0.341	0.945	0.688
CECT-1^*	0.800	0.711-0.890	<0.001	0.707	0.782	0.750
CECT-2^**	0.831	0.747-0.914	<0.001	0.756	0.818	0.792
Combined^***	0.850	0.773-0.927	<0.001	0.829	0.764	0.792