Page 75 - AIH-2-2

Artificial Intelligence in Health Improved liver tumor segmentation with dense networks

Figure 10. Three examples of segmentation results derived from the liver tumor segmentation challenge test dataset using our proposed method. The red
regions denote the segmented liver whereas the green ones denote the segmented lesions.

Figure 11. Two examples of segmentation results derived from the 3DIRCADb dataset using our proposed method. The red regions denote the liver
whereas the green ones denote the lesions.

we reproduced its lesion segmentation result on the Table 5. Comparison with Public 2D Methods on the
3DIRCADb dataset, as these results were not provided in 3DIRCADb Dataset
29
Li et al. Our method still outperforms 2D DenseUNet Method Lesion Dice score
on the 3DIRCADb dataset, with a 1% increase in lesion UNet 0.51±0.25
Dice score. The experimental comparison demonstrates
the superiority of the proposed method. Figure 11 presents Cascade UNets 0.56±0.26
some examples of liver tumor segmentation results of our ResNet 0.60±0.12
2
method on the 3DIRCADb dataset. I -DenseFCN 0.62±0.02

Volume 2 Issue 2 (2025) 69 doi: 10.36922/aih.5001