Table 1.

Image metrics calculated between sCT and pCT_deformed per dl-model (Median [range]). The best performance is indicated in bold.

Contour	dl-model	Mean Absolute Error [HU]	Peak Signal-to-Noise Ratio [dB]	Structural Similarity Index Measure [-]
External	Dual-UNet	25 [19–30]	39.4 [39.0–40.8]	0.74 [0.68–0.79]
	Single-UNet	40 [34–56]	36.4 [35.0–37.33]	0.65 [0.52–0.70]
	CycleGAN	41 [37–46]	37.3 [36.6–37.8]	0.67 [0.59–0.70]
Bladder	Dual-UNet	19 [13–29]	40 [39–45]	–
	Single-UNet	40 [30–66]	37 [34–39]	–
	CycleGAN	37 [26–56]	38 [36–40]	–
Rectum	Dual-UNet	18 [12–23]	41 [37–45]	–
	Single-UNet	50 [32–95]	36 [32–39]	–
	CycleGAN	41 [27–55]	37 [33–40]	–