A Kalman Filtering and Nonlinear Penalty Regression Approach for Noninvasive Anemia Detection with Palpebral Conjunctiva Images

Table 1

Performance comparison between the methods with and without KF.

	Sensitivity	Specificity	Number of suspect samples	Sensitivity change	Specificity change	Number of suspect samples change
R + linear regression (adopted from [38])	0.8333	0.8261	65	+2.86%	−6.62%	−32.31%
KF + R + linear regression (adopted from [38])	0.8571	0.7714	44
Erythema index + linear regression [10]	1.0000	1.0000	74	0%	−4.35%	−24.34%
KF + erythema index + linear regression [10]	1.0000	0.9565	56
Hue + nonlinear penalty regression [39]	NAN	NAN	100	NAN	NAN	−42%
KF + hue + nonlinear penalty regression [39]	0.8462	0.5862	58
R + nonlinear penalty regression	0.7647	0.8158	45	−0.36%	−0.89%	−28.89%
KF + R + nonlinear penalty regression (proposed algorithm)	0.7619	0.8085	32

NAN: there is no data, which means all test samples are considered to be suspect samples.