Table 3.
Performance of multivariable analysis with each modality and fusion method
| Class | Training dataset | Validation dataset |
|---|---|---|
| No fusion-based features | ||
| T1-weighted MR images | 0.8151 (0.6809–0.9493) | 0.8333 (0.5817–0.9999) |
| T2-weighted MR images | 0.8263 (0.6849–0.9677) | 0.6904 (0.3792–0.9999) |
| PET images | 0.8095 (0.6708–0.9483) | 0.8571 (0.5732–0.9999) |
| Image-level fusion-based features | ||
| T1/PET Image Fusion (0.1) | 0.8655 (0.7482–0.9829) | 0.9524 (0.8413–0.9999) |
| T1/PET Image Fusion (0.2) | 0.8711 (0.7606–0.9817) | 0.8571 (0.6376–0.9999) |
| T1/PET Image Fusion (0.3) | 0.8683 (0.7565–0.9802) | 0.8333 (0.6897–0.9999) |
| T1/PET Image Fusion (0.4) | 0.8179 (0.6820–0.9539) | 0.8810 (0.6897–0.9999) |
| T1/PET Image Fusion (0.5) | 0.8599 (0.7422–0.9777) | 0.8571 (0.5771–0.9999) |
| T1/PET Image Fusion (0.6) | 0.8571 (0.7340–0.9803) | 0.7381 (0.4246–0.9999) |
| T1/PET Image Fusion (0.7) | 0.8487 (0.7262–0.9712) | 0.7476 (0.4898–0.9999) |
| T1/PET Image Fusion (0.8) | 0.8207 (0.6872–0.9543) | 0.7381 (0.4388–0.9999) |
| T1/PET Image Fusion (0.9) | 0.8515 (0.7287–0.9743) | 0.7381 (0.4300–0.9999) |
| T2/PET Image Fusion (0.1) | 0.8431 (0.7200–0.9663) | 0.8810 (0.6897–0.9999) |
| T2/PET Image Fusion (0.2) | 0.8627 (0.7438–0.9817) | 0.9048 (0.7090–0.9999) |
| T2/PET Image Fusion (0.3) | 0.8403 (0.7156–0.9651) | 0.9048 (0.7356–0.9999) |
| T2/PET Image Fusion (0.4) | 0.7983 (0.6377–0.9589) | 0.7857 (0.5136–0.9999) |
| T2/PET Image Fusion (0.5) | 0.8319 (0.6913–0.9726) | 0.6905 (0.3643–0.9999) |
| T2/PET Image Fusion (0.6) | 0.8739 (0.7647–0.9832) | 0.6667 (0.3398–0.9935) |
| T2/PET Image Fusion (0.7) | 0.8571 (0.7369–0.9774) | 0.7619 (0.4631–0.9999) |
| T2/PET Image Fusion (0.8) | 0.8347 (0.7087–0.9608) | 0.7381 (0.4214–0.9999) |
| T2/PET Image Fusion (0.9) | 0.8319 (0.6996–0.9643) | 0.7857 (0.5117–0.9999) |
| Matrix-level fusion-based features | ||
| T1/PET Matrix Fusion | 0.8291 (0.7004–0.9579) | 0.7857 (0.5139–0.9999) |
| T2/PET Matrix Fusion | 0.8235 (0.6749–0.9722) | 0.6190 (0.2632–0.9749) |
| Feature-level fusion based features | ||
| T1/PET Feature Concatenation | 0.8459 (0.7245–0.9674) | 0.6905 (0.3788–0.9999) |
| T2/PET Feature Concatenation | 0.8543 (0.7330–0.9757) | 0.9047 (0.7361–0.9999) |
| T1/PET Feature Average | 0.8543 (0.7363–0.9723) | 0.7857 (0.5139–0.9999) |
| T2/PET Feature Average | 0.8347 (0.6963–0.9731) | 0.8571 (0.6132–0.9999) |
For the class column, the number in the parentheses indicated the MR weight. For the training dataset and validation dataset columns, the number in the parentheses indicated the 95% confidence interval of AUC