UU-Mamba is a novel model for cardiac image segmentation that integrates the U-Mamba architecture with an uncertainty-aware loss function and the Sharpness-Aware Minimization (SAM) optimizer. The model improves segmentation accuracy and robustness by combining region-based, distribution-based, and pixel-based losses, which help prioritize confident predictions and reduce the impact of ambiguous data. The SAM optimizer enhances generalization by locating flat minima in the loss landscape, reducing overfitting. Evaluation on the ACDC cardiac dataset shows that UU-Mamba outperforms state-of-the-art models like TransUNet, Swin-Unet, nnUNet, and nnFormer, achieving a high Dice Similarity Coefficient (DSC) and low Mean Squared Error (MSE). The model's performance is further improved by the uncertainty-aware loss and SAM optimization, which together enhance segmentation accuracy and robustness. The UU-Mamba model demonstrates superior performance in cardiac MRI segmentation, achieving a DSC of 92.787%. The model's effectiveness is validated through quantitative evaluation and ablation studies, showing that the combination of uncertainty-aware loss and SAM optimization significantly improves segmentation results. Future work will focus on exploring additional data augmentation techniques and validating the model on larger and more diverse datasets.UU-Mamba is a novel model for cardiac image segmentation that integrates the U-Mamba architecture with an uncertainty-aware loss function and the Sharpness-Aware Minimization (SAM) optimizer. The model improves segmentation accuracy and robustness by combining region-based, distribution-based, and pixel-based losses, which help prioritize confident predictions and reduce the impact of ambiguous data. The SAM optimizer enhances generalization by locating flat minima in the loss landscape, reducing overfitting. Evaluation on the ACDC cardiac dataset shows that UU-Mamba outperforms state-of-the-art models like TransUNet, Swin-Unet, nnUNet, and nnFormer, achieving a high Dice Similarity Coefficient (DSC) and low Mean Squared Error (MSE). The model's performance is further improved by the uncertainty-aware loss and SAM optimization, which together enhance segmentation accuracy and robustness. The UU-Mamba model demonstrates superior performance in cardiac MRI segmentation, achieving a DSC of 92.787%. The model's effectiveness is validated through quantitative evaluation and ablation studies, showing that the combination of uncertainty-aware loss and SAM optimization significantly improves segmentation results. Future work will focus on exploring additional data augmentation techniques and validating the model on larger and more diverse datasets.