Table 2.

Prediction accuracy with matching uncertainties based on 8-Hz motion waveforms.

Patient	Tumor motion and volume				AR accuracy (mm) b					LSTM accuracy (mm) b
	SI (mm)	AP (mm)	LR (mm)	Vol (cc)	Prediction			Matched		Prediction			Matched
					AVG a	STD	t (s)	AVG a	STD	AVG a	STD	t (s)	AVG a	STD
1	3.6	6.7	2.1	5.2	0.03	0.03	0.52	0.05	0.06	0.14	0.12	1.87	0.19	0.13
2	9.1	8.0	5.2	16.2	0.05	0.04	0.51	0.06	0.05	0.20	0.17	1.85	0.22	0.16
3	6.6	5.5	8.9	8.8	0.08	0.06	0.50	0.10	0.07	0.15	0.12	1.93	0.19	0.13
4	3.7	4.5	4.1	25.1	0.06	0.05	0.51	0.07	0.06	0.17	0.13	1.89	0.20	0.14
5	3.5	2.1	2.1	3.5	0.04	0.04	0.51	0.06	0.04	0.11	0.09	1.90	0.13	0.10
6	6.0	8.9	9.9	10.0	0.10	0.09	0.50	0.14	0.11	0.17	0.14	1.93	0.20	0.14
7	8.1	8.4	6.5	64.1	0.05	0.04	0.50	0.07	0.05	0.15	0.12	1.92	0.16	0.11
8	5.2	2.7	5.3	1.2	0.06	0.07	0.44	0.09	0.11	0.30	0.45	1.89	0.37	0.51
AVG	5.7	5.9	5.5	16.8	0.06	0.05	0.50	0.08	0.07	0.18	0.17	1.90	0.21	0.18
STD	2.1	2.6	2.9	20.6	0.02	0.02	0.03	0.03	0.03	0.06	0.11	0.03	0.07	0.14

The AR (autoregression) and LSTM (long short-term memory) prediction accuracy and performance, tumor motion, location, and volume, and the library matching errors are provided. The AR method provides more accurate prediction results using less time than that of the LSTM method.

^ap-value <0.001 in both prediction and matched accuracy between AR and LSTM.

^bNo significant difference between prediction and matched accuracy using either AR or LSTM.