Benchmark evaluation demonstrates improved cardiac MRI segmentation accuracy across the ACDC dataset, highlighting superior robustness over existing architectures.
Key Points
The proposed UU-Mamba architecture outperforms state-of-the-art models on the ACDC cardiac dataset, achieving superior Dice similarity coefficient and mean squared error scores.
Assessment using the Sharpness-Aware Minimization optimizer enables convergence to flat minima, reducing overfitting while combining region, distribution, and pixel loss functions.
This uncertainty-aware loss framework addresses computational demands in cardiac MRI segmentation, providing a robust automated alternative to labor-intensive manual delineation.