Table 8.

Impacts of three attention modules on localization performance.

Dataset	CAM	SAM	MAM	IoU	Dice	p-value
BUSI	✗	✓	✓	0.411 ± 0.024	0.521 ± 0.018	**
	✓	✗	✓	0.413 ± 0.014	0.515 ± 0.005	**
	✓	✓	✗	0.419 ± 0.010	0.528 ± 0.007	**
	✓	✓	✓	0.438 ± 0.016	0.543 ± 0.018	–
MBUD	✗	✓	✓	0.341 ± 0.039	0.466 ± 0.056	**
	✓	✗	✓	0.358 ± 0.024	0.468 ± 0.029	**
	✓	✗	✓	0.384 ± 0.051	0.510 ± 0.059	ns
	✓	✓	✓	0.406 ± 0.035	0.534 ± 0.040	–

Notes: The symbol ✗means this attention module was removed in our framework. In this experiment, ResNet50 was used as the FEX, and the input image size was 256 × 256 pixels. The results are represented as a mean value with a ±95% confidence interval. The null hypothesis and meaning of the p-values remain consistent with the description in Table 6.