Research on glioma grading prediction based on habitat imaging using multimodal magnetic resonance imaging
10.3969/j.issn.1672-8270.2024.10.001
- VernacularTitle:基于生境成像的多模态磁共振成像胶质瘤分级预测研究
- Author:
Tianci LIU
1
;
Yao ZHENG
;
Huan XU
;
Yutao HE
;
Yuefei FENG
;
Xiaoshuo HAO
;
Yang LIU
Author Information
1. 空军军医大学军事生物医学工程学系军事医学信息技术教研室 西安 710032
- Keywords:
Glioma;
Heterogeneity;
Radiomics;
Machine learning;
Habitat imaging
- From:
China Medical Equipment
2024;21(10):1-5,35
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To develop a machine learning algorithm based on habitat imaging(HI),which can be used in the grading of gliomas by using multimodal magnetic resonance imaging(MRI),so as to construct the model of support vector machine(SVM)and the visualized heterogeneous regions of gliomas.Methods:A total of 335 glioma patients were collected from the 2019 brain tumor segmentation(BraTS)challenge competition of World Health Organization(WHO),which included 259 cases with high-grade gliomas(HGG)and 76 cases with low-grade gliomas(LGG).Subregions were divided based on HI technology.The PyRadiomics open-source package was used to extract the image features of region of interest(ROI),and to screen the features that stronger correlated with the high and low-grade gliomas.An SVM model was used to classify and predict the screened feature data.The heterogeneity of gliomas in images was analyzed through visualized characterization.The efficacy of glioma grading was assessed by using the area under curve(AUC)of the receiver operating characteristic(ROC)curve.Results:The AUC of test set exceeded 90%.The average accuracy of the performance indicators of test set was(92.74±2.88)%,and the average sensitivity was(93.90±2.10)%,and the average specificity was(90.36±4.59)%,and the average F1 score was(95.24±0.66)%when the tumors were divided into six habitat regions.The SVM model could showed important sub-regions in glioma grading in three-dimensional space.Conclusion:The study method based on HI has significant advantages in glioma grading,which can effectively realize visualized heterogeneity of tumor and construct model of the heterogeneity of tumor.
