Multi-Parameter Magnetic Resonance Machine Learning Model in the Differential Diagnosis of Primary Central Nervous System Lymphoma and Atypical Glioblastoma
10.3969/j.issn.1005-5185.2024.11.001
- VernacularTitle:多参数磁共振机器学习模型鉴别诊断原发性中枢神经系统淋巴瘤与不典型胶质母细胞瘤的价值
- Author:
Mingxiao WANG
1
;
Guoli LIU
;
Yanhua LI
;
Shuo SUN
;
Lin MA
Author Information
1. 解放军医学院,北京 100853;解放军总医院第一医学中心放射诊断科,北京 100853
- Keywords:
Primary central nervous system lymphoma;
Atypical glioblastoma;
Magnetic resonance imaging;
Radiomics;
Machine learning;
Diagnosis,differential
- From:
Chinese Journal of Medical Imaging
2024;32(11):1089-1096
- CountryChina
- Language:Chinese
-
Abstract:
Purpose To construct the model of differentiating primary central nervous system lymphoma(PCNSL)and atypical glioblastoma(GBM)by combining multi-parameter MRI radiomics and six machine learning algorithms,thus to compare the diagnostic efficacy of different machine learning algorithms.Materials and Methods The clinical and imaging data of 77(125 lesions)PCNSL and 90 atypical GBM(108 lesions)from PLA General Hospital and public databases were retrospectively analyzed from January 2013 to December 2023,and all patients were randomly divided into a training set(163 cases)and a validation set(70 cases)according to 7∶3.T1WI,T2WI and T1-weighted contrast-enhanced sequences were selected for tumor segmentation,and 1 132 radiomics features were extracted from each region of interest.The intraclass correlation coefficient(ICC)was used for the consistency test,and image features with ICC≥0.85 were selected.ICC and recursive feature elimination were used to select the best radiomics features.Six classifiers were used to train and verify three single sequence feature sets,three double-sequence sets and one multi-sequence set.The receiver operating characteristic curve was used to evaluate the diagnostic efficacy of the model.Results The combination model of the support vector machine of radial basis function classifier and multi-sequence feature set were the best model for differentiating PCNSL and atypical GBM.The area under the curve of the training set and the validation set were 0.969 and 0.913,respectively;and the accuracy were both 0.886.Conclusion Noninvasive extraction of multiparametric MRI features combined with machine learning algorithms can effectively differentiate PCNSL and atypical GBM,which provides support for the development of individualized treatment plans for patients.
