Page 252 - ITU Kaleidoscope 2016

2016 ITU Kaleidoscope Academic Conference

troller need not to be a separated device, as in the system
depicted which consists of a 3W solar panel that provides 5V
to the Arduino board and to the 2300 mAh Li-Ion battery.

Figure 2: The Patient Prioritization System.

order to analyze the patient data stored in the cloud database
storage c) a mobile visualization application applying rule
based data analysis in order to perform situation awareness
Figure 3: Solar Powered Sensor Subsystem [8].
and d) a server application interfacing with the cloud storage
and the Mobile Visualization application. The triage scoring
system adopted by our system was described in [1]. It is
3. MACHINE LEARNING ALGORITHMS
based on a model that assigns scores to vital signs used as
Triage parameters in order to quantify their severity level.
The aim of a patient prioritization system also called Triage
As depicted by figure 2, upon determination of the quanti-
system is to build upon the WHO scoring system and ma-
tative measure of a patient medical condition, the feedback
chine learning techniques borrowed from the field of artifi-
from the patient prioritization software can be provided to
cial intelligence to determine a quantitative measure of pa-
the users (patients and/or doctors) through the mobile visu-
tient’s medical conditions (condition recognition) and then
alization application revealing the condition of the patient
give priorities to the most urgent cases. The Triage algo-
in green, red or orange statuses and different actions to be
rithm should be easily interpretable, fast to compute, simple,
taken: do nothing if green status, plan a visit to the doctor if
accurate and scalable in order to be portable to small devices,
orange status and see the doctor immediately if red status.
e.g. smart phones, tablets, iPad, and/or work with biomedical
sensors and smart watches without any problems. Two ma-
2.3. Information Dissemination and Power Supply chine learning algorithms were selected to solve the patient
prioritization problem. Their basic characteristics are de-
Different frequency bands have been recently recommended scribed below and their performance compared in section 4.
by the 802.15.6 standard to mitigate the interference in the
ISM band resulting from co-location of bio-sensors with
3.1. Multivariate Linear Regression
other devices. While interference is very significant in the
urban settings of the developing world, the less crowded
ISM band of the rural settings of the developing world can
tolerate less complex mitigation schemes used to overcome
wireless interference. The focus of our work will therefore
lie on the IEEE802.11 and IEEE802.15.4 communication
standards when operating in the ISM band, as they are
commonly available and have a built in interference mit-
igation mechanism. The power supply model adopted by
a Cyber-healthcare system is as important as its commu-
nication model since both have an impact on its efficiency Figure 4: Multivariate Linear Regression.
and the lifetime of the underlying communication network
infrastructure. The Cyber-healthcare power supply system The Matrix Algebra method (MAM) has been often used as
proposed in this paper is based on photovoltaic energy. It an alternative to the the multivariate linear regression (MLR)
consists of a solar powered sensor subsystem depicted in algorithm. However, the MLR by gradient descent was used
Figure 3 where a photovoltaic panel converts light into vari- in our work because MAM tends to run slower with the in-
able DC power, a charge controller is used to charge the crease of the amount of data.
battery and to supply the proper voltage to the load. The Algorithm description. As illustrated by Figure 4, this al-
same scheme can be applied to provide large amounts of gorithm uses the knowledge based system to score the train-
electricity, but in the case of a sensor board, the charge con- ing data before training. In other words, it is an improved

– 234 –