我们通过比较我们的模型crossmodel和之前的state-of-the-arts：1transunet;2unext;如表一所示。首先我们能够看到当训练集大小为208时。我们的模型相较于transunetunext在Dice指标上分别提升了2224在Miou指标上分别提升了2641。当训练集大小缩小10大小为186后我们的模型相较于transunetunext在Dice指标上分别提升了2530

We compared our crossmodel with two state-of-the-art models, TransUNet and U-Net with Extra Skip Connections (UNetX), as shown in Table 1. Firstly, when the training set size was 208, our model improved the Dice score by 2.2% and 2.4% and the Mean Intersection over Union (Miou) score by 2.6% and 4.1% compared to TransUNet and UNetX, respectively. When decreasing the training set size by 10% to 186, our model still outperformed TransUNet and UNetX with a 2.5% and 3.0% improvement in Dice score and a 2.9% and 3.9% improvement in Miou score. Furthermore, when the training set size was further decreased by 20% to 162, our model achieved a higher Dice score by 4.1% and 4.5% and a higher Miou score by 5.1% and 5.8% compared to TransUNet and UNetX, respectively. Our results demonstrated that our model performed better than TransUNet and UNetX, particularly when the training set size was small. This is consistent with the design of our model, which leverages the inherent characteristics of the data to provide more supervision signals when the data is limited