Radiomics-Based Differentiation of Glioblastoma and Metastatic Disease: Impact of Different T1-Contrast-Enhanced Sequences on Radiomics Features and Model Performance

Article Figures & Data

Download figure
Open in new tab
Download powerpoint
FIG 1.
Schematic depicting the overall study workflow.
Download figure
Open in new tab
Download powerpoint
FIG 2.
Representative images from a patient with BM demonstrating the baseline T1 postcontrast MPRAGE (A) and VIBE (D) images, the whole tumor mask (shaded in red) in MPRAGE (B) and VIBE (E), and the necrotic mask in MPRAGE (C) and VIBE (F). The mask for the solid-only enhancing component was obtained by subtracting the corresponding necrotic component mask (shaded in red) from the whole-tumor mask.
Download figure
Open in new tab
Download powerpoint
FIG 3.
Boxplots showing Pearson correlation estimate distributions for the enhancing and necrotic tumor components derived from the 2 sequences for the uncorrected data. Corr. indicates Correlation.
Download figure
Open in new tab
Download powerpoint
FIG 4.
Differences in AUC-ROC performance (MPRAGE-VIBE) between the sequences when using the same ML model and feature-reduction methodology. PCA indicates principal components analysis; KNN, K-nearest neighbors; GBRM, generalized boosted regression model; XBG, extreme gradient boosting.

Table 1:

Demographic and clinical information of the patient cohorts

Table 2:

Performance metrics for the top 5 models using the VIBE and MPRAGE sequences

Model	Feature Selection	AUC-ROC		Log Loss		Brier Score
Model	Feature Selection	Mean	SD	Mean	SD	Mean	SD
VIBE
SVM-RBF	LC	0.907	0.086	0.389	0.165	0.117	0.062
NNET	LC	0.896	0.056	0.678	0.425	0.144	0.041
SVM-Poly	LC	0.895	0.093	0.400	0.178	0.117	0.053
RF	MRMR	0.893	0.056	0.433	0.204	0.128	0.051
ENET	LC	0.893	0.085	0.454	0.138	0.132	0.039
MPRAGE
SVM-Poly	LC	0.890	0.093	0.432	0.129	0.132	0.048
GBRM	LC	0.888	0.060	0.449	0.144	0.132	0.050
SVM-RBF	LC	0.882	0.106	0.430	0.213	0.133	0.071
SVM-RBF	MRMR	0.873	0.073	0.445	0.152	0.136	0.055
XGB	LC	0.871	0.091	0.475	0.226	0.143	0.067

Note:—ENET indicates elastic net; LC, linear combination filter; GBRM, generalized boosted regression model; NNET, neural network; XGB, extreme gradient boosting.