Preoperative prediction of thymoma risk classification based on enhanced CT deep learning radiomics
-
摘要:
目的 探讨基于增强CT深度学习影像组学术前预测胸腺瘤风险分类的价值。 方法 收集2015年1月至2023年1月皖南医学院弋矶山医院139例胸腺瘤患者为训练集,池州市人民医院与芜湖市第二人民医院81例患者为外部验证集。其中低风险组(A、AB、B1型)137例、高风险组(B2、B3型)83例。基于CT静脉期图像分别提取手工影像组学(hand-crafted radiomics,HCR)特征与深度学习(deep learning,DL)特征,构建影像组学标签评分(Radscore)。单因素与多因素Logistic回归分析筛选预测胸腺瘤高风险的影像学独立影响因素,利用逻辑回归(Logistic regression,LR)、随机森林(random forest,RF)、决策树(decision tree,DT)及支持向量机(support vector machine,SVM)机器学习建模。选择最佳模型为输出模型,构建模型列线图、校准曲线及临床决策曲线(decision curve analysis,DCA)。受试者工作特征(receiver operating characteristics,ROC)曲线下面积(area under the curve,AUC)及净重新分类指数(net reclassification index,NRI)用于评估模型效能。 结果 LR、RF、DT及SVM构建预测模型的AUC分别为训练集0.899、0.835 、0.897、0.860,外部验证集0.889、0.830、0.905、0.858。由于训练集LR模型效能最佳,作为本研究输出模型,校准曲线及DCA表明模型具有较高的校准度及临床适用性。列线图与Radscore比较效能有改善,NRI训练集为7.5%(P=0.007),外部验证集为5.3%(P=0.020)。 结论 基于增强CT深度学习影像组学术前实现了胸腺瘤准确风险分类,列线图能够提供个性化预测结果。 Abstract:Objective To explore the value of using deep learning radiomics to preoperatively predict thymoma risk classification based on enhanced CT images. Methods 139 patients with thymoma from Yijiashan Hospital of Wannan Medical College from January 2015 to January 2023 were selected as the training set, and 81 patients from The People's Hospital of Chizhou and The Second People's Hospital of Wuhu as the external validation set. There were 137 cases in the low-risk group (types A, AB, B1) and 83 in the high-risk group (types B2, B3). Based on the images from the CT venous phase, we extracted hand-crafted radiomics (HCR) and deep learning (DL) features and constructed Radscore. We used single- and multi-factor Logistic regression analyses to screen independently influencing factors in imaging, predicting a high risk of thymoma. We used four types of machine learning modeling methods, Logistic regression (LR), random forest (RF), decision tree (DT), and support vector machine (SVM), for modeling. We selected the best model as the output model and constructed the model nomogram, calibration curve, and clinical decision curve analysis (DCA). We used the area under the receiver operating characteristic (ROC) curve (AUC) and net reclassification index (NRI) to evaluate the model performance. Results The AUCs of LR, RF, DT, and SVM for constructing prediction models were 0.899, 0.835, 0.897, and 0.860 for the training set and 0.889, 0.830, 0.905, and 0.858 for the external validation set, respectively. We chose the LR model in the training set as the output model for this study owing to its outstanding performance. Calibration curves and DCA indicated that the model had a high degree of calibration and clinical applicability. The performance of the nomogram improved compared to Radscore, with an NRI training set of 7.5% (P=0.007) and an external validation set of 5.3% (P=0.020). Conclusions Based on the radiomics of the enhanced CT deep learning imaging panel, accurate risk classification of thymomas has been achieved. The nomogram can provide personalized prediction outcomes. -
Key words:
- computer tomography (CT) /
- deep learning /
- radiomics /
- thymoma /
- risk classification
-
表 1 训练集组内及训练集与外部验证集组间临床影像特征比较
变量 训练集(n=139) x2 P 外部验证集(n=81) x2 P 低风险组(n=86) 高风险组(n=53) 低风险组(n=51) 高风险组(n=30) 性别 0.950 0.330 0.042 0.838 男 43 31 26 16 女 43 22 25 14 位置 0.513 0.774 1.611 0.447 右 26 18 21 10 居中 24 12 12 11 左 36 23 18 9 形态 10.404 0.001 5.985 0.014 规则 55 19 33 11 不规则 31 34 18 19 边界 6.342 0.012 5.544 0.019 清晰 75 37 45 20 不清晰 11 16 6 10 钙化 0.503 0.778 2.395 0.302 弧形 9 4 2 3 簇状 8 4 8 2 无 69 45 41 25 坏死囊变 0.040 0.841 1.711 0.191 有 23 15 18 13 无 63 38 43 17 增强程度 5.890 0.015 5.931 0.015 中低强化 51 42 27 24 明显强化 35 11 24 6 增强均匀性 0.209 0.647 0.106 0.745 均匀 44 25 24 13 不均匀 42 28 27 17 周围侵犯 13.779 <0.001 6.953 0.008 有 4 14 3 8 无 82 39 48 22 表 2 影像特征预测胸腺瘤高风险的影响因素分析变量
项目 单因素分析 多因素分析 B OR(95%CI) P B OR(95%CI) P 形态 1.155 3.175(1.556~6.480) 0.002 0.903 2.466(1.124~5.410) 0.024 边界 1.081 2.948(1.244~6.987) 0.014 0.873 2.395(0.904~6.343) 0.079 增强程度 −0.963 0.382(0.173~0.842) 0.017 −0.977 0.377(0.160~0.888) 0.026 周围侵犯 1.996 7.359(2.273~23.824) 0.001 1.888 6.605(1.850~23.579) 0.004 表 3 LR、RF、DT及SVM机器学习算法构建模型的效能评价
模型 AUC(95%CI) 准确度(%) 敏感度(%) 特异度(%) 阳性预测值(%) 阴性预测值(%) 训练集 LR 0.899(0.848~0.950) 82.0 84.9 79.1 72.6 89.6 RF 0.835(0.761~0.908) 80.6 66.0 90.7 79.5 81.1 DT 0.897(0.844~0.950) 83.5 77.4 87.2 78.8 86.2 SVM 0.860(0.796~0.923) 78.4 83.0 76.7 68.3 86.8 外部验证集 LR 0.889(0.817~0.961) 82.7 86.7 80.4 72.2 91.1 RF 0.830(0.736~0.925) 74.1 93.3 62.7 59.6 94.1 DT 0.905(0.839~0.970) 82.7 80.0 84.3 75.0 87.8 SVM 0.858(0.772~0.944) 85.2 70.0 94.1 87.5 84.2 表 4 机器学习算法构建模型的效能比较
模型 Z P 训练集 LR vs. RF 3.459 <0.001 LR vs. DT 0.112 0.911 LR vs. SVM 1.864 0.062 RF vs. DT −2.318 0.020 RF vs. SVM −0.787 0.431 DT vs. SVM 1.549 0.121 外部验证集 LR vs. RF 2.365 0.018 LR vs. DT −0.702 0.483 LR vs. SVM 1.484 0.138 RF vs. DT −2.045 0.041 RF vs. SVM −0.914 0.361 DT vs. SVM 1.318 0.188 -
[1] 杜鑫,于磊,杨凌,等.NPTX1在胸腺瘤患者中的表达及诊断价值[J].中国肺癌杂志,2021,24(1):1-6. doi: 10.3779/j.issn.1009-3419.2021.102.03 [2] Roden AC, Yi ES, Jenkins SM, et al. Reproducibility of 3 histologic classifications and 3 staging systems for thymic epithelial neoplasms and its effect on prognosis[J]. Am J Surg Pathol, 2015, 39(4):427-441. [3] 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. doi: 10.3389/fonc.2022.1043163 [4] Ge YX, Xu WB, Wang Z, et al. Prognostic value of CT radiomics in evaluating lymphovascular invasion in rectal cancer: diagnostic performance based on different volumes of interest[J]. J Xray Sci Technol, 2021, 29(4):663-674. [5] 郑华龙,沈莉莉,郑巧灵,等.基于新辅助化疗前CT影像组学的列线图预测进展期胃腺癌远期疗效[J].中国肿瘤临床,2023,50(7):336-344. doi: 10.12354/j.issn.1000-8179.2023.20220831 [6] Pei Q, Yi XP, Chen C, et al. Pre-treatment CT-based radiomics nomogram for predicting microsatellite instability status in colorectal cancer[J]. Eur Radiol, 2022, 32(1):714-724. [7] Dong D, Fang MJ, Tang L, et al. Deep learning radiomic nomogram can predict the number of lymph node metastasis in locally advanced gastric cancer: an international multicenter study[J]. Ann Oncol, 2020, 31(7):912-920. [8] Yuan ZX, Xu TY, Cai J, et al. Development and validation of an image-based deep learning algorithm for detection of synchronous peritoneal carcinomatosis in colorectal cancer[J]. Ann Surg, 2022, 275(4):e645-e651. [9] Zheng XY, Yao Z, Huang YN, et al. Deep learning radiomics can predict axillary lymph node status in early-stage breast cancer[J]. Nat Commun, 2020, 11(1):1236. [10] Tuan PA, Vien MV, Dong HV, et al. The value of CT and MRI for determining thymoma in patients with myasthenia gravis[J]. Cancer Control, 2019, 26(1):1-10. [11] Yamazaki M, Oyanagi K, Umezu H, et al. Quantitative 3D shape analysis of CT images of thymoma: a comparison with histological types[J]. AJR Am J Roentgenol, 2020, 214(2):341-347. [12] 吴树剑,俞咏梅,范莉芳,等.MSCT影像组学结合机器学习预测直径2~5 cm胃胃肠间质瘤危险度分级研究[J].中国实用外科杂志,2022,42(12):1401-1407. [13] Sadohara J, Fujimoto K, Müller NL, et al. Thymic epithelial tumors: comparison of CT and MR imaging findings of low-risk thymomas, high-risk thymomas, and thymic carcinomas[J]. Eur J Radiol, 2006, 60(1):70-79. [14] Yakushiji S, Tateishi U, Nagai, et al. Computed tomographic findings and prognosis in thymic epithelial tumor patients[J]. J Comput Assist Tomogr, 2008, 32(5):799-805. [15] Strange CD, Ahuja J, Shroff GS, et al. Imaging evaluation of thymoma and thymic carcinoma[J]. Front Oncol, 2021, 11:810419. [16] Hu YC, Wu L, Yan LF, et al. Predicting subtypes of thymic epithelial tumors using CT: new perspective based on a comprehensive analysis of 216 patients[J]. Sci Rep, 2014, 4:6984. [17] Pan CC, Chen WYK, Chiang H. Spindle cell and mixed spindle/lymphocytic thymomas[J]. Am J Surg Pathol, 2001, 25(1):111-120. [18] Liu GB, Poon M, Zapala MA, et al. Incorporating radiomics into machine learning models to predict outcomes of neuroblastoma[J]. J Digit Imaging, 2022, 35(3):605-612. [19] Yan M, Zhang X, Zhang B, et al. Deep learning nomogram based on Gd-EOB-DTPA MRI for predicting early recurrence in hepatocellular carcinoma after hepatectomy[J]. Eur Radiol, 2023, 33(7):4949-4961. [20] Zou Y, Shi Y, Liu JH, et al. A comparative analysis of six machine learning models based on ultrasound to distinguish the possibility of central cervical lymph node metastasis in patients with papillary thyroid carcinoma[J]. Front Oncol, 2021, 11:656127. [21] Xiao G, Hu YC, Ren JL, et al. MR imaging of thymomas: a combined radiomics nomogram to predict histologic subtypes[J]. Eur Radiol, 2021, 31(1):447-457. [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] Jiang YM, Zhang ZC, Yuan QY, et al. Predicting peritoneal recurrence and disease-free survival from CT images in gastric cancer with multitask deep learning: a retrospective study[J]. Lancet Digit Health, 2022, 4(5):e340-e350. -