Table 5 Zero-shot cross-domain performance

From: Prompt-mamba filtering networks for accurate hepatocellular carcinoma lesion segmentation in abdominal CT

Method

Dice (%)

HD95 (mm)

Recall (%)

U-Net

67.8

14.2

70.3

TransUNet

72.6

11.1

75.2

Swin-UNet

74.5

9.3

77.0

MedSAM

76.4

8.5

78.1

Prompt-Mamba-AF (Ours)

79.2

7.4

82.5

  1. Models trained on LiTS were evaluated on 3DIRCADb without fine-tuning. Bold indicates best performance.
Back to article page