Algorithms design for improving homecare using Electrocardiogram (ECG) signals and Internet of Things (IoT)

Abstract

Due to the fast growing of population, a lot of hospitals get crowded from the huge amount of
patients visits. Moreover, during COVID-19 a lot of patients prefer staying at home to minimize
the spread of the virus. The need for providing care to patients at home is essential. Internet
of Things (IoT) is widely known and used by different fields. IoT based homecare will help
in reducing the burden upon hospitals. IoT with homecare bring up several benefits such as
minimizing human exertions, economical savings and improved efficiency and effectiveness. One
of the important requirement on homecare system is the accuracy because those systems are
dealing with human health which is sensitive and need high amount of accuracy. Moreover,
those systems deal with huge amount of data due to the continues sensing that need to be
processed well to provide fast response regarding the diagnosis with minimum cost requirements.
Heart is one of the most important organ in the human body that requires high level of caring.
Monitoring heart status can diagnose disease from the early stage and find the best medication
plan by health experts. Continues monitoring and diagnosis of heart could exhaust caregivers
efforts. Having an IoT heart monitoring model at home is the solution to this problem. Electrocardiogram
(ECG) signals are used to track heart condition using waves and peaks. Accurate
and efficient IoT ECG monitoring at home can detect heart diseases and save human lives.
As a consequence, an IoT ECG homecare monitoring model is designed in this thesis for detecting
Cardiac Arrhythmia and diagnosing heart diseases. Two databases of ECG signals are used;
one online which is old and limited, and another huge, unique and special from real patients
in hospital. The raw ECG signal for each patient is passed through the implemented Low
Pass filter and Savitzky Golay filter signal processing techniques to remove the noise and any
external interference. The clear signal in this model is passed through feature extraction stage
to extract number of features based on some metrics and medical information along with feature extraction algorithm to find peaks and waves. Those features are saved in the local database to
apply classification on them. For the diagnosis purpose a classification stage is made using three
classification ways; threshold values, machine learning and deep learning to increase the accuracy.
Threshold values classification technique worked based on medical values and boarder lines. In
case any feature goes above or beyond these ranges, a warning message appeared with expected
heart disease. The second type of classification is by using machine learning to minimize the
human efforts. A Support Vector Machine (SVM) algorithm is proposed by running the algorithm
on the features extracted from both databases. The classification accuracy for online and hospital
databases was 91.67% and 94% respectively. Due to the non-linearity of the decision boundary, a
third way of classification using deep learning is presented. A full Multilayer Perceptron (MLP)
Neural Network is implemented to improve the accuracy and reduce the errors. The number of
errors reduced to 0.019 and 0.006 using online and hospital databases.
While using hospital database which is huge, there is a need for a technique to reduce the amount
of data. Furthermore, a novel adaptive amplitude threshold compression algorithm is proposed.
This algorithm is able to make diagnosis of heart disease from the reduced size using compressed
ECG signals with high level of accuracy and low cost. The extracted features from compressed
and original are similar with only slight differences of 1%, 2% and 3% with no effects on machine
learning and deep learning classification accuracy without the need for any reconstructions. The
throughput is improved by 43% with reduced storage space of 57% when using data compression.
Moreover, to achieve fast response, the amount of data should be reduced further to provide
fast data transmission. A compressive sensing based cardiac homecare system is presented.
It gives the channel between sender and receiver the ability to carry small amount of data.
Experiment results reveal that the proposed models are more accurate in the classification of
Cardiac Arrhythmia and in the diagnosis of heart diseases. The proposed models ensure fast
diagnosis and minimum cost requirements. Based on the experiments on classification accuracy,
number of errors and false alarms, the dictionary of the compressive sensing selected to be 900.
As a result, this thesis provided three different scenarios that achieved IoT homecare Cardiac
monitoring to assist in further research for designing homecare Cardiac monitoring systems. The experiment results reveal that those scenarios produced better results with high level of accuracy
in addition to minimizing data and cost requirements.