MR Imaging–Based Radiomic Signatures of Distinct Molecular Subgroups of Medulloblastoma

Article Figures & Data

Download figure
Open in new tab
Download powerpoint
Fig 1.
Illustration of 2 strategies used to evaluate the supervised machine learning models to predict the molecular subgroups of medulloblastoma. The upper and lower figures show details of double 10-fold cross-validation and 3-dataset cross-validation schemes, respectively.
Download figure
Open in new tab
Download powerpoint
Fig 2.
Receiver operating characteristic curves with a double 10-fold cross-validation scheme for support vector machine to predict the 4 main molecular subgroups of medulloblastoma with the use of computational MR imaging features.
Download figure
Open in new tab
Download powerpoint
Fig 3.
Receiver operating characteristic curves with the largest mean AUC values for 4 distinct molecular subgroups of medulloblastoma with a 3-dataset cross-validation scheme.
Download figure
Open in new tab
Download powerpoint
Fig 4.
MR imaging appearance of the 4 core molecular subgroups of medulloblastoma on T2-weighted and contrast-enhanced T1-weighted images.

Table 1:

Patient demographics

Table 2:

Predictive performance of 2 machine learning models for the identification of medulloblastoma molecular subgroups

MRI Dataset/Targeted Subgroup	AUC with Double 10-Fold Cross-Validation	AUC with 3-Dataset Cross-Validation
T1
SHH	0.67	0.73
WNT	0.56	0.47
Group 3	0.40	0.54
Group 4	0.79	0.76
T2
SHH	0.70	0.66
WNT	0.63	0.72
Group 3	0.51	0.57
Group 4	0.54	0.59
T1 + T2
SHH	0.79	0.70
WNT	0.45	0.45
Group 3	0.70	0.39
Group 4	0.83	0.80