Skip to main content
. 2025 Jul 15;16:1611267. doi: 10.3389/fphys.2025.1611267

TABLE 2.

The evaluation metric score comparison of our method, LRU-Net, with six other methods on the training and validation sets.

Method Evaluation metrics score on training data Evaluation metrics score on validation data
Accuracy F1 Score IoU Dice Coefficient Score Dice Loss Boundary Dice Loss Accuracy F1 Score IoU Dice Coefficient Score Dice Loss Boundary Dice Loss
LRU-Net 98.87% 97.58% 96.45% 98.71% 5.13% 8.19% 98.83% 98.16% 96.40% 97.93% 5.25% 8.60%
Attention-Unet 94.11% 90.61% 83.76% 90.19% 26.0% 44.22% 94.67% 90.72% 83.75% 90.10% 26.36% 42.91%
Attention-ResUnet 95.75% 93.06% 89.72% 92.51% 20.2% 19.70% 95.97% 93.38% 88.14% 92.38% 19.70% 31.88%
InceptionV3-Unet 97.67% 95.92% 89.66% 95.82% 11.1% 18.71% 97.75% 96.40% 93.20% 97.82% 10.48% 17.13%
Swin-Unet 96.88% 95.33% 91.29% 95.71% 11.3% 16.65% 97.03% 96.17% 91.17% 96.21% 13.54% 19.97%
Trans-Unet 98.16% 97.01% 95.28% 97.97% 7.02% 9.72% 97.95% 97.53% 95.82% 96.81% 9.33% 10.04%
Rethinking ResNets 98.70% 97.36% 96.30% 98.53% 6.07% 9.27% 98.73% 97.98% 96.33% 97.36% 5.81% 9.77%

The bold values highlights that our proposed method LRU-Net achieved the highest accuracy, F1 score, IoU, Dice Coefficient Score, Dice loss, and Boundary Dice Loss on the validation data.