Online prediction for respiratory movement compensation: a patient-specific gating control for MRI-guided radiotherapy

Table 3 Mean and std deviation of gating errors with prediction length of 0.4 s and 0.6 s for 21 liver cancer patients and 10 lung cancer patients, as well as calculation time for each model

PW	Model	Liver organs			Lung tumors			Time (ms)
PW	Model	Crossing (ms)	Gating (ms)	Gat-acc (%)	Crossing (ms)	Gating (ms)	Gat-acc (%)	Time (ms)
0.4 s (j = 2)	Linear	30 ± 25	190 ± 27	93.9 ± 2.1	25 ± 28	183 ± 65	91.6 ± 3.3	0.06
	Ridge	32 ± 27	190 ± 30	93.8 ± 2.1	25 ± 29	184 ± 66	91.4 ± 3.3	0.07
	L2-L1	58 ± 24	184 ± 46	93.3 ± 2.3	44 ± 29	191 ± 76	90.1 ± 3.9	0.03
	LSTM	75 ± 64	232 ± 141	93.0 ± 4.2	75 ± 52	233 ± 111	90.5 ± 4.7	3.71
	Bi-LSTM	73 ± 53	285 ± 261	90.6 ± 9.9	70 ± 52	267 ± 101	90.0 ± 4.8	5.43
	GRU	71 ± 69	283 ± 250	90.5 ± 16	47 ± 44	249 ± 107	90.2 ± 4.7	3.36
	KF	68 ± 32	240 ± 130	91.2 ± 17	52 ± 32	212 ± 94	91.2 ± 3.7	0.02
0.6 s (j = 3)	Linear	56 ± 33	56 ± 33	98.3 ± 1.0	45 ± 30	45 ± 30	98.0 ± 1.8	0.07
	Ridge	58 ± 33	58 ± 33	98.2 ± 1.1	47 ± 31	47 ± 31	97.8 ± 2.5	0.06
	L2-L1	80 ± 50	80 ± 50	97.2 ± 1.2	80 ± 47	80 ± 47	96.4 ± 2.5	0.02
	LSTM	112 ± 85	112 ± 85	96.1 ± 2.6	138 ± 74	138 ± 74	94.7 ± 3.3	4.36
	Bi-LSTM	130 ± 101	130 ± 101	96.0 ± 2.7	131 ± 73	131 ± 73	94.7 ± 4.0	7.25
	GRU	112 ± 72	112 ± 72	96.5 ± 2.3	125 ± 80	125 ± 80	95.1 ± 3.4	4.32
	KF	102 ± 36	102 ± 36	96.9 ± 1.2	115 ± 70	115 ± 70	95.2 ± 1.9	0.02
	AL (30 s)	44 ± 23	44 ± 23	98.3 ± 0.6	45 ± 31	45 ± 31	98.0 ± 1.7	0.99

The best performing model for gating errors is shown in bold.
PW prediction window, Crossing crossing on/off error, Gating gating on/off error, Gat-acc gating accuracy, KF Kalman filter, AL adaptive linear, 30 s using only the first 30 s as the burn-in period.

ISSN: 1748-717X