Performance Comparison of Liver Segmentation Using U-Net, Attention-Enhanced U-Net, And R2u-Net Models

In medical imaging, precise segmentation is crucial for enhancing diagnostic outcomes. This study compares the efficacy of the conventional U-Net model with an attention-enhanced U-Net model for liver segmentation tasks using the Medical Segmentation Decathlon dataset. Both models were trained and validated on a set of 202 liver CT volumes, assessing their performance using metrics such as Dice Coefficient and Jaccard Similarity. Our findings show that while the attention-enhanced U-Net demonstrated a marginal improvement in segmentation accuracy over the baseline U-Net model, these enhancements did not significantly translate to the validation set, suggesting limitations in generalization. This paper discusses the challenges of integrating attention mechanisms in medical image segmentation and suggests directions for future research in optimizing these models.