Information Technology Reference
In-Depth Information
Feature Selection
The same principle is applied for the GSR
where V
GSR
is measured through pins RB3 and
RB5, and used to find the conductance:
In selecting the features there is a need to analyze
the physiological indicators' relationship with
agitation. In order to relate heart rate to agita-
tion the Central Nervous System (CNS) must be
considered. CNS consists of two main sections;
the somatic part responsible for all our voluntary
movements and the autonomic part which controls
all the non-voluntary movements. The autonomic
system itself is split to two main sections: the
Sympathetic Nervous System (SNS) (concerned
with all emergency cases we encounter), and the
Parasympathetic Nervous System (PNS) (respon-
sible for our relaxed state). High PNS activity
indicates a relaxed patient while high SNS activity
corresponds to agitation.
High correlation between Heart Rate Variabil-
ity (HRV), PNS and SNS has been demonstrated
in stress studies (Dishman, et al., 2000). In most
clinical applications HRV is analyzed in both time
and frequency domains. The R-R interval is an
important measure of the heart state from which
the IBI is extracted. IBI can be extracted by tak-
ing any point as the reference point but since the
R peak has the highest amplitude, it is immune
against possible sources of noise (Lee, et al., 2004)
(Malik, et al., 1996). In the frequency domain,
studies have shown that power distribution has
four main spectral components: High Frequency
(HF), Low Frequency (LF), Very Low Frequency
(VLF) and Ultra Low Frequency (Murray, 2003).
Large correlation has been demonstrated between
HF and PNS and between LF and SNS (Dishman,
et al., 2000). LF is calculated from the IBI which
is extracted from the normal to normal interval.
Due to this relationship between agitation and
IBI, it is used as one of the features for agitation
detection.
Two additional features are used due to their
direct relation with stress status. Skin conductance
and skin temperature, which can be measured non-
invasively, are monitored and used in the detection
algorithm. The correlation of skin temperature is
V
0 5
.
−
GSR
5
GSR
=
470 000
,
.
V
0 5
.
+
GSR
5
In order to obtain subject independent predic-
tions it is necessary to normalize the data using
the method presented in Sakr et al. (2010). For
example; GSR(normalized)= (GSR-mean)/stdev.
The values of all the averages and standard de-
viations are computed during the first minute.
After the first minute these values are fixed and
used in the normalization procedure. After the
normalization phase, the data is passed to the
decision function that generates the final decision
based on the algorithm described next. Finally the
inter-beat interval, the skin temperature and skin
conductivity are sent along with the decision to the
PC via UART2 pin and Bluetooth. The Bluetooth
air cable creates a virtual serial port on the PC that
communicates using a baud rate of 115200bps.
The oscillator used allows the DSPIC to achieve
this rate with 0% error.
A C# application was developed to receive the
data sent by the device to be stored for possible
further analysis. The data is displayed using a
DataGridView and stored simultaneously in a
text file. This text file is formatted to be compat-
ible with MATLAB to facilitate further analysis
of the data.
Agitation Detection Algorithm
In this section we present the derivation of the
detection algorithm as well as its implementation.
Search WWH ::
Custom Search