Machine learning-based classification model of lymph node metastasis in nasopharyngeal carcinoma(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2019年第11期
- Page:
- 1350-1355
- Research Field:
- 其他（激光医学等）
- Publishing date:
Info
- Title:
- Machine learning-based classification model of lymph node metastasis in nasopharyngeal carcinoma
- Author(s):
- LIU Yuan1; 2; CHENG Yuyu3; HE Ruimin4; ZHOU Weibing1; HE Qiudong4; XIAO Ruobing4; HE Yang4; XIE Changjun4; XIE Haihui4; WEN Hongyong4; CHEN Juan4; HE Yaolin4
- 1. Department of Oncology, Xiangya Hospital, Central South University, Changsha 410008, China; 2. Department of Nuclear Medicine, Chenzhou No.1 People’s Hospital, Chenzhou 423000, China; 3. Chuanshan College, University of South China, Hengyang 421001, China; 4. Department of Radiotherapy, the Second Affiliated Hospital, University of South China, Hengyang 421001, China
- Keywords:
- Keywords: machine learning; nasopharyngeal carcinoma; metastatic lymph node; auxiliary diagnosis; model
- PACS:
- R730.49；R318
- DOI:
- DOI:10.3969/j.issn.1005-202X.2019.11.020
- Abstract:
- To establish a model for identifying metastatic lymph nodes in CT images of nasopharyngeal carcinoma with the use of machine learning and radiomics. Methods The plain CT, intravenous perfusion enhancement and 18F-FGD-PET images of 50 pre-treatment patients with nasopharyngeal carcinoma were analyzed in the study. All patients were confirmed by pathological and PET examinations as nasopharyngeal carcinoma with local lymph node metastasis. The lymph nodes with a volume >1 cm3 in CT image were manually delineated, and the maximum standardized uptake value (SUVmax)>2.5 of the corresponding area in 18F-FGD-PET image and the current standard were used as the diagnose criteria. A total of 143 lymph nodes were obtained, of which 103 were metastatic lymph nodes. The above classification results were trained and verified by a machine learning model. There were 100 lymph nodes in the training data and 43 in the test data. The lymph nodes were randomly grouped in order to avoid system errors caused by specific grouping. Results During the machine learning, a model was constructed from several imaging ensembles (lymph node volume, maximum cross-sectional minor axis and several radiomics features). The accuracy of the model for the classification of metastatic lymph nodes reached 86%. The result of feature selection showed that the maximum cross-sectional diameter, mean breadth, intensity direct energy, number of voxels, busyness and shape-compactness could be used as important features in the classification of metastatic lymph nodes. Conclusion The established classification model can be used for the auxiliary diagnose of metastatic lymph nodes in CT images, which provides a new idea for the rapid determination of metastatic lymph node in patients with nasopharyngeal carcinoma, and it’s helpful for the accurate segmentation of target areas in indi vidualized radiotherapy.
Last Update: 2019-11-28