Page 149 - Control Theory in Biomedical Engineering

134 Control theory in biomedical engineering

Fig. 1 Cross-validation of the handwriting model using a Kalman filter (Okorokova et al.,
2015).

dorsal interosseus, and flexor carpi radialis) (Zhang and Kamavuako, 2019).
The experimental procedure consisted of in-house software that collected
the x and y coordinates provided by the tablet together with the three
EMG channels. The considered database contains four graphic traces com-
posed of different kinds of lines: straight, broken, and curves as shown in
Fig. 2. Each trace graphic was recorded five times at a moderate speed
and participants were asked to maintain the speed as constant as possible.
Three features were explored, mainly the mean absolute average (MAV),
root mean square (RMS), and mean power frequency (MPF), bringing
the number of inputs to nine. The proposed approach is based on a back
propagation (BP) neural network with nine inputs, a single hidden layer
with six neurons and an output layer with two neurons. Thus the total num-
ber of weights to tune was 108, which are tuned during calibration only.
This structure was chosen as a good trade-off between accuracy and speed.
The four basic traces were used for training only. The ability of the model to
reconstruct writing was tested by asking the subjects to perform random

Fig. 2 The four traces used in Zhang-Kamavuako handwriting.