The value of a decision tree model based on multiparametric MRI in the diagnosis of parotid tumors
10.3760/cma.j.cn112149-20231229-00503
- VernacularTitle:基于MRI多参数成像的决策树模型在腮腺肿瘤诊断中的价值
- Author:
Gongxin YANG
1
;
Xiaoqing DAI
;
Ling ZHU
;
Xiaofeng TAO
Author Information
1. 上海交通大学医学院附属第九人民医院放射科，上海 200011
- Keywords:
Parotid neoplasms;
Magnetic resonance imaging;
Decision tree
- From:
Chinese Journal of Radiology
2024;58(5):503-509
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To establish a diagnostic decision tree model for parotid tumors closely related to clinical treatment decisions based on multiparametric MRI and to explore and validate its clinical value in parotid tumor diagnosis.Methods:This study was a cross-sectional study that retrospectively collected MRI data from 461 patients with pathologically confirmed parotid tumors from June 2018 to December 2022 at the Ninth People′s Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, including 364 cases of benign tumors, 82 cases of malignant epithelial tumors, and 15 cases of lymphoma. Stratified random sampling was performed according to pathological results to divide the data into a training set (326 cases) and a validation set (135 cases). In the training set, there were 256 cases of benign tumors, 59 cases of malignant epithelial tumors, and 11 cases of lymphoma, while in the validation set, there were 108 cases of benign tumors, 23 cases of malignant epithelial tumors, and 4 cases of lymphoma. Based on MRI and clinical features, a decision tree model was established using the Chi-squared Automatic Interaction Detector (CHAID) algorithm, and the model was used for classification diagnosis. The diagnostic accuracy for benign tumors, malignant epithelial tumors, and lymphoma was calculated, and receiver operating characteristic curves were plotted to evaluate the diagnostic performance for each tumor type individually.Results:In the training set, four optimal diagnostic indicators for tumors were obtained through the CHAID algorithm, including tumor capsule, shape, apparent diffusion coefficient value, and time-signal curve type. A decision tree model was established based on these indicators. The overall diagnostic accuracy of the model for benign tumors, malignant epithelial tumors, and lymphoma was 90.8% (296/326) in the training set and 94.1% (127/135) in the validation set. The area under the curve for independent diagnosis of benign tumors, malignant epithelial tumors, and lymphoma in the training set was 0.964, 0.957, and 0.980, respectively, while in the validation set, it was 0.958, 0.944, and 0.992, respectively.Conclusion:The decision tree predictive model based on multi-parameter MRI demonstrates high efficacy in diagnosing benign tumors, malignant epithelial tumors, and lymphoma of the parotid gland.
