Machine learning adaptation of intraocular lens power calculation for a patient group

Table 2 Means, SDs, MedAEs of prediction errors, and proportions of errors within ± 0.25 D, ± 0.50 D, ± 1.00 D, and greater than + 0.50 D in the use of 5 kinds of power calculations (N = 395 eyes)

Power calculation method	SRK/T	Haigis	BUII	Hill-RBF	Adapted
Mean (SD), D	0.02 (0.45)	− 0.03 (0.44)	0.25 (0.40)	0.12 (0.40)	0.01 (0.38)
MedAE, D	0.27	0.27	0.31	0.27	0.21
Within ± 0.25 D	46.1%	47.1%	41.0%^x	46.6%	54.4%*
Within ± 0.50 D	75.2%	78.2%	70.9%^x	79.5%	83.5%*
Within ± 1.00 D	96.2%	96.7%	96.2%	97.5%	98.5%
Greater than + 0.50 D	24.8%	21.8%	29.1%	20.5%	16.5%*

BUII Barrett Universal II; SD standard deviation; D diopter; MedAE median absolute error
Adapted denotes SRK/T formula adapted with support vector regressor
^* and ^x: significantly larger or smaller than the mean levels, respectively (residual analysis)

ISSN: 2326-0254