Table 2 Segmentation results on INbreast, BUSI, and Duke-Breast-MRI datasets

From: Anatomy-guided visual prompt tuning for cross-modal breast cancer understanding

Method

INbreast Dice

BUSI Dice

Duke-MRI Dice

U-Net27

84.2

82.7

80.6

U-Net++28

86.0

83.5

81.5

TransUNet29

88.1

84.2

82.3

Swin-U-Net30

88.9

85.0

83.5

SegFormer-B131

89.2

85.5

84.1

nnU-Net32

89.6

85.8

84.8

Swin-Transformer3

90.1

86.0

85.1

MedT33

90.3

86.2

85.3

LoRA24

90.8

86.7

85.6

VPT14

91.0

87.0

85.8

A-VPT (ours)

92.2

88.1

86.9

  1. Metrics: Dice (%), IoU (%), HD95 (mm, lower is better). All parameter-efficient baselines share the same frozen ViT-B/16 backbone.
  2. The bold values indicate the results of our proposed model.
Back to article page