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 4 Comparison of segmentation results between the proposed model and those of other studies for brain metastases of different sizes in the external validation dataset

	DSC	IoU	Accuracy	Sensitivity	Specificity
Proposed model
< 5 mm	0.652 ± 0.163	0.549 ± 0.158	0.780 ± 0.136	0.657 ± 0.158	0.780 ± 0.133
5–20 mm	0.775 ± 0.141	0.669 ± 0.144	0.936 ± 0.120	0.794 ± 0.127	0.924 ± 0.128
> 20 mm	0.816 ± 0.118	0.710 ± 0.140	0.969 ± 0.113	0.871 ± 0.115	0.961 ± 0.124
PAM-U-Net [21]
< 5 mm	0.624 ± 0.168	0.526 ± 0.146	0.805 ± 0.123	0.630 ± 0.114	0.732 ± 0.139
5-20 mm	0.726 ± 0.121	0.630 ± 0.137	0.927 ± 0.115	0.712 ± 0.106	0.909 ± 0.116
> 20 mm	0.753 ± 0.094	0.698 ± 0.119	0.963 ± 0.106	0.747 ± 0.098	0.974 ± 0.103
GAN + Mask-R CNN + CRF [20]
< 5 mm	0.635 ± 0.090	0.512 ± 0.058	0.776 ± 0.071	0.618 ± 0.102	0.839 ± 0.095
5-20 mm	0.710 ± 0.121	0.624 ± 0.104	0.875 ± 0.101	0.731 ± 0.105	0.912 ± 0.112
> 20 mm	0.749 ± 0.098	0.675 ± 0.082	0.938 ± 0.113	0.850 ± 0.089	0.967 ± 0.107

ISSN: 1748-717X