Table 1.
Comparisons of the performance for prediction of glioma grades and molecular subtypes on different MR modalities by fivefold cross-validation of deep learning
| Tasks | Modality | Accuracy | Sensitivity | Specificity | PPV | NPV | F1-score |
|---|---|---|---|---|---|---|---|
| OGG/GBM | T2 FLAIR | 0.69 ± 0.11 | 0.75 ± 0.22 | 0.60 ± 0.25 | 0.75 ± 0.14 | 0.65 ± 0.18 | 0.60 ± 0.12 |
| T1WI + C | 0.74 ± 0.11 | 0.75 ± 0.22 | 0.73 ± 0.25 | 0.83 ± 0.15 | 0.64 ± 0.08 | 0.70 ± 0.14 | |
| QSM | 0.80 ± 0.07 | 0.86 ± 0.12 | 0.73 ± 0.25 | 0.84 ± 0.13 | 0.75 ± 0.13 | 0.75 ± 0.14 | |
| LGG/HGG | T2 FLAIR | 0.69 ± 0.11 | 0.80 ± 0.19 | 0.53 ± 0.16 | 0.70 ± 0.07 | 0.67 ± 0.18 | 0.61 ± 0.13 |
| T1WI + C | 0.74 ± 0.14 | 0.76 ± 0.16 | 0.73 ± 0.25 | 0.81 ± 0.17 | 0.62 ± 0.10 | 0.72 ± 0.17 | |
| QSM | 0.69 ± 0.11 | 0.70 ± 0.19 | 0.67 ± 0.21 | 0.76 ± 0.14 | 0.59 ± 0.08 | 0.64 ± 0.12 | |
| IDH1(+)/(−) | T2 FLAIR | 0.57 ± 0.20 | 0.65 ± 0.26 | 0.47 ± 0.16 | 0.60 ± 0.17 | 0.48 ± 0.17 | 0.53 ± 0.19 |
| T1WI + C | 0.57 ± 0.20 | 0.60 ± 0.26 | 0.53 ± 0.27 | 0.63 ± 0.22 | 0.57 ± 0.28 | 0.52 ± 0.19 | |
| QSM | 0.77 ± 0.11 | 0.86 ± 0.12 | 0.67 ± 0.30 | 0.81 ± 0.17 | 0.72 ± 0.16 | 0.70 ± 0.18 | |
| ATRX(−)/(+) | T2 FLAIR | 0.54 ± 0.17 | 0.60 ± 0.26 | 0.47 ± 0.27 | 0.62 ± 0.22 | 0.54 ± 0.26 | 0.46 ± 0.13 |
| T1WI + C | 0.46 ± 0.11 | 0.50 ± 0.16 | 0.40 ± 0.13 | 0.52 ± 0.11 | 0.38 ± 0.10 | 0.42 ± 0.09 | |
| QSM | 0.60 ± 0.11 | 0.65 ± 0.20 | 0.53 ± 0.27 | 0.69 ± 0.17 | 0.59 ± 0.22 | 0.52 ± 0.09 |
OGG other grade glioma (grade II + III), GBM glioblastoma multiforme (grade IV), LGG low-grade glioma (grade II), HGG high-grade glioma (grade III + IV), IDH1(+) IDH1 mutation, IDH1(−) IDH1 wild, ATRX(−) ATRX expression loss, ATRX(+) ATRX retention, PPV positive predictive value, NPV negative predictive value
The best results are marked in bold