Table 1 Performance and mean training time comparison of DNN models (U-Net and backbone U-Net) using different MRI image weights for glioma semantic segmentation.
From: Enhanced glioma semantic segmentation using U-net and pre-trained backbone U-net architectures
DNN models | MRI image weights | ACC (%) | Mean IoU | Training time for each epoch |
|---|---|---|---|---|
U-Net | T1 | 97.6 | 0.724 | 17Â s 7 ms |
T2 | 97.27 | 0.759 | 18Â s 7 ms | |
T1Gd | 98.15 | 0.782 | 18Â s 7 ms | |
T2-FLAIR | 97.3 | 0.754 | 17Â s 7 ms | |
ResNet-U-Net | T1 | 98.41 | 0.802 | 82Â s 29 ms |
T2 | 98.45 | 0.810 | 82Â s 29 ms | |
T1Gd | 98.62 | 0.830 | 82Â s 29 ms | |
T2-FLAIR | 98.42 | 0.795 | 82Â s 29 ms | |
Inception-U-Net | T1 | 98.39 | 0.796 | 114Â s 40 ms |
T2 | 98.41 | 0.801 | 115Â s 40 ms | |
T1Gd | 98.31 | 0.810 | 116Â s 41 ms | |
T2-FLAIR | 98.48 | 0.812 | 116Â s 41 ms | |
VGG-U-Net | T1 | 98.26 | 0.761 | 89Â s 31 ms |
T2 | 98.34 | 0.777 | 89Â s 31 ms | |
T1Gd | 98.34 | 0.808 | 89Â s 31 ms | |
T2-FLAIR | 98.39 | 0.787 | 90Â s 32 ms |
