Wei Lu1, Lianzhen Zhong2, Di Dong3, Mengjie Fang4, Qi Dai5, Shaoyi Leng6, Liwen Zhang7, Wei Sun8, Jie Tian9, Jianjun Zheng10, Yinhua Jin11. 1. Department of Medical Imaging, Hwa Mei Hospital, University of Chinese Academy of Sciences, Ningbo, Zhejiang, China; CAS Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, Beijing, China. Electronic address: luwei19@ucas.ac.cn. 2. CAS Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, Beijing, China; University of Chinese Academy of Sciences, Beijing, China. Electronic address: zhonglianzhen2018@ia.ac.cn. 3. CAS Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, Beijing, China; University of Chinese Academy of Sciences, Beijing, China. Electronic address: di.dong@ia.ac.cn. 4. CAS Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, Beijing, China; University of Chinese Academy of Sciences, Beijing, China. Electronic address: fangmengjie2015@ia.ac.cn. 5. Department of Medical Imaging, Hwa Mei Hospital, University of Chinese Academy of Sciences, Ningbo, Zhejiang, China. Electronic address: yxdaiqi@163.com. 6. Department of Medical Imaging, Hwa Mei Hospital, University of Chinese Academy of Sciences, Ningbo, Zhejiang, China. Electronic address: lengshaoyi@163.com. 7. CAS Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, Beijing, China; University of Chinese Academy of Sciences, Beijing, China. Electronic address: zhangliwen2015@ia.ac.cn. 8. Department of Medical Imaging, Hwa Mei Hospital, University of Chinese Academy of Sciences, Ningbo, Zhejiang, China. Electronic address: sunwei@ucas.ac.cn. 9. CAS Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, Beijing, China; University of Chinese Academy of Sciences, Beijing, China; Beijing Advanced Innovation Center for Big Data-Based Precision Medicine, School of Medicine, Beihang University, Beijing, 100191, China. Electronic address: jie.tian@ia.ac.cn. 10. Department of Medical Imaging, Hwa Mei Hospital, University of Chinese Academy of Sciences, Ningbo, Zhejiang, China. Electronic address: zhengjianjun@ucas.ac.cn. 11. Department of Medical Imaging, Hwa Mei Hospital, University of Chinese Academy of Sciences, Ningbo, Zhejiang, China. Electronic address: jinyh@ucas.ac.cn.
Abstract
PURPOSE: Cervical lymph node (LN) metastasis of papillary thyroid carcinoma (PTC) is critical for treatment and prognosis. We explored the feasibility of using radiomics to preoperatively predict cervical LN metastasis in PTC patients. METHOD: Total 221 PTC patients (training cohort: n = 154; validation cohort: n = 67; divided randomly at the ratio of 7:3) were enrolled and divided into 2 groups based on LN pathologic diagnosis (N0: n = 118; N1a and N1b: n = 88 and 15, respectively). We extracted 546 radiomic features from non-contrast and venous contrast-enhanced computed tomography (CT) images. We selected 8 groups of candidate feature sets by minimum redundancy maximum relevance (mRMR), and obtained 8 radiomic sub-signatures by support vector machine (SVM) to construct the radiomic signature. Incorporating the radiomic signature, CT-reported cervical LN status and clinical risk factors, a nomogram was constructed using multivariable logistic regression. The nomogram's calibration, discrimination, and clinical utility were assessed. RESULTS: The radiomic signature was associated significantly with cervical LN status (p < 0.01 for both training and validation cohorts). The radiomic signature showed better predictive performance than any radiomic sub-signatures devised by SVM. Addition of radiomic signature to the nomogram improved the predictive value (area under the curve (AUC), 0.807 to 0.867) in the training cohort; this was confirmed in an independent validation cohort (AUC, 0.795 to 0.822). Good agreement was observed using calibration curves in both cohorts. Decision curve analysis demonstrated the radiomic nomogram was worthy of clinical application. CONCLUSIONS: Our radiomic nomogram improved the preoperative prediction of cervical LN metastasis in PTC patients.
PURPOSE: Cervical lymph node (LN) metastasis of papillary thyroid carcinoma (PTC) is critical for treatment and prognosis. We explored the feasibility of using radiomics to preoperatively predict cervical LN metastasis in PTC patients. METHOD: Total 221 PTC patients (training cohort: n = 154; validation cohort: n = 67; divided randomly at the ratio of 7:3) were enrolled and divided into 2 groups based on LN pathologic diagnosis (N0: n = 118; N1a and N1b: n = 88 and 15, respectively). We extracted 546 radiomic features from non-contrast and venous contrast-enhanced computed tomography (CT) images. We selected 8 groups of candidate feature sets by minimum redundancy maximum relevance (mRMR), and obtained 8 radiomic sub-signatures by support vector machine (SVM) to construct the radiomic signature. Incorporating the radiomic signature, CT-reported cervical LN status and clinical risk factors, a nomogram was constructed using multivariable logistic regression. The nomogram's calibration, discrimination, and clinical utility were assessed. RESULTS: The radiomic signature was associated significantly with cervical LN status (p < 0.01 for both training and validation cohorts). The radiomic signature showed better predictive performance than any radiomic sub-signatures devised by SVM. Addition of radiomic signature to the nomogram improved the predictive value (area under the curve (AUC), 0.807 to 0.867) in the training cohort; this was confirmed in an independent validation cohort (AUC, 0.795 to 0.822). Good agreement was observed using calibration curves in both cohorts. Decision curve analysis demonstrated the radiomic nomogram was worthy of clinical application. CONCLUSIONS: Our radiomic nomogram improved the preoperative prediction of cervical LN metastasis in PTC patients.