Channel-wise attention enhanced and structural similarity constrained cycleGAN for effective synthetic CT generation from head and neck MRI images

Table 1 CT synthesis accuracies for different synthesis methods

Range	Parameter	cycleGAN	SSC-cycleGAN	cycleSimulationGAN	P1	P2	P3
Body	MAE	71.32 ± 1.37	63.18 ± 1.80	61.88 ± 1.42	< 0.001	< 0.001	0.077
	RMSE	126.05 ± 2.94	120.03 ± 3.70	116.85 ± 3.42	< 0.001	< 0.001	< 0.001
	PSNR	34.95 ± 0.380	35.76 ± 0.50	36.23 ± 0.52	< 0.001	< 0.001	< 0.001
	SSIM	0.975 ± 0.002	0.982 ± 0.002	0.985 ± 0.002	< 0.001	< 0.001	0.953
Air (HU<-100)	MAE	128.97 ± 6.58	114.48 ± 6.03	111.41 ± 4.78	< 0.001	< 0.001	< 0.001
	RMSE	199.64 ± 11.31	176.97 ± 11.76	174.71 ± 9.26	< 0.001	< 0.001	0.023
	PSNR	31.11 ± 1.27	33.23 ± 0.98	33.00 ± 1.11	< 0.001	< 0.001	< 0.001
	SSIM	0.982 ± 0.004	0.987 ± 0.002	0.987 ± 0.003	< 0.001	< 0.001	0.126
Soft-tissue (150 > HU≥-100)	MAE	39.08 ± 1.32	32.22 ± 1.60	31.99 ± 1.11	< 0.001	< 0.001	< 0.001
	RMSE	65.38 ± 3.58	59.68 ± 4.17	59.51 ± 3.61	< 0.001	< 0.001	0.256
	PSNR	26.19 ± 0.68	27.00 ± 0.49	27.01 ± 0.54	< 0.001	< 0.001	0.637
	SSIM	0.954 ± 0.005	0.962 ± 0.005	0.962 ± 0.005	< 0.001	< 0.001	0.087
Bone (HU ≥ 150)	MAE	170.71 ± 8.36	166.35 ± 9.17	160.69 ± 9.42	< 0.001	< 0.001	< 0.001
	RMSE	222.02 ± 11.50	218.30 ± 12.96	210.92 ± 12.47	0.001	< 0.001	< 0.001
	PSNR	27.27 ± 0.39	27.37 ± 0.37	27.44 ± 0.38	0.877	< 0.001	< 0.001
	SSIM	0.980 ± 0.001	0.982 ± 0.001	0.985 ± 0.001	0.742	0.592	0.830

Notes P1 = cycleGAN vs. SSC-cycleGAN, P2 = cycleGAN vs. cycleSimulationGAN, P3 = SSC-cycleGAN vs. cycleSimulationGAN.

ISSN: 1748-717X