Non-linear estimation methods for hematocrit density based on changing pattern of transduced anodic current curve

Many studies reported that the hematocrit (HCT) is the most highly influencing factor affecting the accuracy of the glucose measurements by portable/handheld devices. It is also known as an important factor for clinical decision-making situations. Therefore, estimation of HCT plays a crucial role for enhancing accuracy of glucose measurements and performance of therapy. In this paper, we present novel methods for hematocrit estimation from the transduced current curve which is produced by glucose-oxidase reaction in strip-type electrochemical biosensors. The proposed methods are nonlinear, including neural networks and support vector machine. Input features are composed of two parts: the sampled points of the time-varying current curve and extended extra features computed from those sampled points.