Table 2.

Prediction accuracy of ACLR and Poisson Lasso model under different gamma criteria.

		MAE(%)			RMSE(%)
		ACLR	PL	Improved	ACLR	PL	Improved
TV	3%/3 mm	1.76**	2.10	16	2.50	2.99	16
	3%/2 mm	2.60**	3.04	17	3.50	4.23	17
	2%/2 mm	4.66**	5.29	12	6.08	6.95	13
CV	3%/3 mm	1.73	2.07	16	2.61	3.04	14
	3%/2 mm	2.75	2.99	8	3.87	4.12	6
	2%/2 mm	5.93*	7.12	17	7.39	8.91	17

Abbreviations: MAE = mean absolute error, RMSE = root mean squared error; TV = technical validation; CV = clinical validation. ACLR = autoencoder based classification-regression deep learning model; PL = Poisson Lasso. Student t test was performed on MAE values between ACLR and PL in TV and CV. Improved = the percentage improvement between ACLR and PL. Bold number in table 2 = absolute prediction errors have significant differences between ACLR and PL

^**in table 2 = (p < 0.01)

^*= (p < 0.05)