Locally advanced rectal cancer:an MRI radiomics study on lymph node re-evaluation after neoadjuvant chemoradiotherapy
10.3760/cma.j.issn.1005-1201.2017.12.008
- VernacularTitle:基于MRI影像组学模型预测局部进展期直肠癌新辅助放化疗后淋巴结状态的研究
- Author:
Xiaoyan ZHANG
1
;
Haitao ZHU
;
Lin WANG
;
Xiaoting LI
;
Yanjie SHI
;
Huici ZHU
;
Yingshi SUN
Author Information
1. 100142,北京大学肿瘤医院暨北京市肿瘤防治研究所医学影像科 恶性肿瘤发病机制及转化研究教育重点实验室
- Keywords:
Rectal neoplasms;
Radiomics;
Lymph node;
re-evaluation;
Neoadjuvant therapy
- From:
Chinese Journal of Radiology
2017;51(12):926-932
- CountryChina
- Language:Chinese
-
Abstract:
Objective To develop and validate one optimal MR radiomics model for lymph node (LN) re-evaluation of locally advanced rectal cancer (LARC) after neoadjuvant chemoradiotheray (NCRT). Methods Four hundred and seven patients with clinicopathologically confirmed LARC in Beijing Cancer Hospital were included in this study from July 2010 to June 2015. All patients received NCRT before surgery,and underwent T2WI and DWI before and after NCRT.These patients were chronologically divided in the primary cohort(300 patients)and independent validation cohort(107 patients).The predicting model was trained and validated using postoperative pathological findings as truth values. By using radiomics method, we extracted the features of the tumor and the largest LN before and after neoadjuvant therapy, combined different features of the tumor and/or the largest LN before and/or after neoadjuvant therapy,and constructed 4 different prediction models,compared the performance of four predicting models.The optimal model with the highest accuracy was validated in the independent cohort. Decision curve analysis was conducted to determine the clinical usefulness of the radiomics nomograms by quantifying the net benefits at different threshold probabilities in the validation dataset. Results In the primary cohort, the radiomics signatures from 4 models provided an AUC of 0.637, 0.709, 0.753, 0.835, respectively in LN re-evaluation after chemoradiotheray. The diagnostic efficacy of model 4 was much better than that of 1, 2 and 3 model. In the validation cohort, the radiomics signatures provided an AUC of 0.795 for LN re-evaluation after chemoradiotheray. The sensitivity, specificity, positive predictive value, negative predictive value were 0.813, 0.693, 0.531, 0.897, respectively (95% CI: 0.694 to 0.896, 0.647 to 0.911, 0.582 to 0.786, 0.361 to 0.621, 0.792 to 0.952). While the probability of predicting N+ ranges from 17% to 80%, using the proposed radiomics model to predict N+ shows a greater advantage than either the scheme in which all patients were assumed to N+ or the scheme in which all patients are N-. Decision curve analysis demonstrated that the radiomics nomograms were clinically useful. Conclusion With a systematic analysis and comparison of both pre-and post-NCRT MRI data, we constructed an optimal individualized LN re-evaluation model based on MR radiomics, combing primary tumor and the largest LN features, compared with other models (only with pre/post tumor or pre/post largest LN features).
