Diffusion-CSPAM U-Net: A U-Net model integrated hybrid attention mechanism and diffusion model for segmentation of computed tomography images of brain metastases

Table 3 Overall mean results of automated segmentation of brain metastases

	DSC	IoU	Accuracy	Sensitivity	Specificity	HD (mm)
Internal validation
U-Net	0.760 ± 0.138	0.652 ± 0.130	0.907 ± 0.114	0.745 ± 0.129	0.894 ± 0.142	7.526 ± 1.107
CSPAM-U-Net	0.801 ± 0.126	0.700 ± 0.132	0.951 ± 0.110	0.792 ± 0.111	0.940 ± 0.145	6.430 ± 0.986
Diffusion-CSPAM-U-Net	0.844 ± 0.128	0.731 ± 0.125	0.972 ± 0.096	0.838 ± 0.113	0.972 ± 0.138	5.107 ± 0.984
External validation
U-Net	0.698 ± 0.151		0.886 ± 0.111	0.708 ± 0.124	0.821 ± 0.147	8.324 ± 1.225
CSPAM-U-Net	0.756 ± 0.139^a	0.656 ± 0.115	0.937 ± 0.107	0.761 ± 0.126	0.890 ± 0.139^a	6.819 ± 1.104^a
Diffusion-CSPAM-U-Net	0.793 ± 0.133^b	0.692 ± 0.133^b	0.955 ± 0.118	0.803 ± 0.121^b	0.938 ± 0.140^b	5.606 ± 0.990^b

DSC, Dice similarity coefficient; IoU, intersection over union. “a” represents the comparison with U-Net where CSPAM-U-Net is significantly different in the corresponding metrics (P < 0.05). “b” denotes that Diffusion-CSPAM-U-Net is significantly different in the comparison with U-Net on the corresponding metric (P < 0.05)

ISSN: 1748-717X