Estimation of cardiac output and systemic vascular resistance using a multivariate regression model with features selected from the finger photoplethysmogram and routine cardiovascular measurements

Table 3 CO and SVR estimation performance of single feature models

Feature	CO (L min−1)			Feature	SVR (dyn.s.cm^-5)
	r	Bias	s.d.		r	Bias	s.d.
28	0.42	0	1.49	24	0.55	-0.87	210
7	0.41	-0.01	1.50	15	0.52	-0.45	215
14	0.38	-0.01	1.52	8	0.52	-0.47	215
21	0.34	-0.01	1.55	1	0.51	-0.57	216
24	0.32	0	1.56	3	0.51	1.85	218
22	0.32	0	1.56	22	0.46	-1.13	224
25	0.28	0.01	1.58	13	0.45	0.13	225
1	0.28	0	1.58	6	0.44	-0.46	226
4	0.25	0.01	1.59	20	0.45	0.92	226
23	0.24	0	1.60	27	0.43	-0.89	228

The Bland-Altman bias and s.d. achieved when a univariate model is used. Only the 10 best single features are shown here, in increasing order of s.d.. The best s.d. achieved when estimating CO using a multivariate model constructed using only PPG features was 1.38 L min^-1. For SVR, the multivariate model has an s.d. of 211 dyn.s.cm^-5.

ISSN: 1475-925X