. 2019 Dec 19;9:19411. doi: 10.1038/s41598-019-55922-0

Table 5.

Prediction performances for malignant glioma grade identification using a radiomic approach in the proposed framework and in previous studies.

Study	No. of data	MRI sequence	Feature type	Filtering	Feature selection	ML algorithm	Data augmentation	Validation method	Accuracy	Sensitivity	Specificity	AUC value
Proposed framework	Primary dataset: 157 (III: 55, IV: 102)	•CE-T1 •T2	•Shape/size •Intensity •Histogram •GLCM •GLRLM •GLSZM •NGLDM •NGTDM	Wavelet transform high-pass and low-pass filters for all feature types excluding the shape/size	WMW test & LASSO-LR	SVM (rbf kernel)	No	LOOCV	0.866	0.902	0.800	0.932
Proposed framework	Entire dataset: 224 (Primary dataset: 157 & Validation dataset: 67 (III: 22, IV: 45))	•CE-T1 •T2			Using the selected radiomic features for all folds in the LOOCV of the primary dataset	RF	No	Independent validation	0.806	0.822	0.773	0.800
Zacharaki et al.²⁰	52 (III: 18, IV: 34)	•CE-T1 •T1 •T2 •FLAIR •rCBV	•Shape •Intensity •Rotation invariant texture	Gabor filter for rotation invariant texture features	SVM-RFE	SVM (rbf kernel)	No	LOOCV	0.904	1.000	0.722	0.985
Zacharaki et al.²⁰	52 (III: 18, IV: 34)	•CE-T1 •T1 •T2 •FLAIR •rCBV	•Shape •Intensity •Rotation invariant texture	Gabor filter for rotation invariant texture features	t-test with bagging	SVM (rbf kernel)	No	LOOCV	0.942	NR	NR	1.000
Tian et al.²¹	111 (III: 33, IV: 78)	•CE-T1 •T1 •T2 •Diffusion •3D pCASL	•GLCM •GLGCM	No	SVM-RFE	SVM (rbf kernel)	No	100-times 10-fold CV	0.937	0.942	0.927	0.982
Tian et al.²¹	111 (III: 33, IV: 78)	•CE-T1 •T1 •T2 •Diffusion •3D pCASL	•GLCM •GLGCM	No	SVM-RFE	SVM (rbf kernel)	SMOTE	100-times 10-fold CV	0.981	0.987	0.974	0.992

CE-T1: contrast-enhanced T1, FLAIR: fluid attenuated inversion recovery, rCBV: relative blood volume, 3D-pCASL: three-dimensional pseudo-continuous arterial spin labeling, GLCM: gray-level co-occurrence matrix, GLRLM: gray-level run length matrix, GLSZM: gray-level size zone matrix, NGLDM: neighboring gray-level dependence matrix, NGTDM: neighborhood gray-tone difference matrix, GLGCM: gray-level gradient co-occurrence matrix, WMW: Wilcoxon-Mann-Whitney, LASSO-LR: least absolute shrinkage and selection operator logistic regression, RFE: recursive feature elimination, SMOTE: synthetic minority over sampling technique, SVM: support vector machine, rbf: radial basis function, RF: random forest, LOOCV: leave-one-out cross validation, AUC: area under the curve, NR: not reported.