Room: Exhibit Hall | Forum 8
Purpose: The purpose of this study was to predict malignant glioma grades using T2-weighted magnetic resonance (T2WI-MR) images based on a radiomics analysis.
Methods: A database for this study included T2WI-MR images of 70 malignant glioma patients (grade III: 25 and grade IV: 45). The tumor regions were manually delineated on the T2WI-MR images. Radiomics features consisted of size, shape, histogram, and texture features were extracted from tumor regions on filtered T2WI-MR images with wavelet transforms. Total number of the radiomics features was 476. Gray level co-occurrence matrix (GLCM), gray level run length matrix (GLRLM), gray level size zone matrix (GLSZM), and neighborhood gray tone difference matrix (NGTDM) were used for texture features. Prediction models of the malignant glioma grades were reconstructed by using logistic regression (LR), support vector machine (SVM), artificial neural network (ANN), random forest (RF), and naÃ¯ve Bayes (NB). The radiomics features were selected by using a least absolute shrinkage and selection operator (LASSO) regression before constructing the prediction models. Three-fold cross validation was performed 10 times and mean area under the curve (AUC) values for all tests were calculated in order to investigate accuracy of the prediction models.
Results: The mean AUC values for all tests calculated by using LR, SVM, ANN, RF, and NB were 0.719Â±0.105, 0.720Â±0.109, 0.697Â±0.094, 0.715Â±0.079, and 0.702Â±0.097, respectively. The constructed model using the SVM was the best performance for predicting the malignant glioma grades in this study.
Conclusion: The framework for predicting the malignant glioma grades using the T2WI-MR images based on the radiomics was developed in this study. The malignant glioma grades would be moderately predicted in accordance with the results of the mean AUC values in the proposed framework.