Leading cause of heart disorder is Arrhythmia which is categorized into three types namely
premature heart beat, tachycardia and bradycardia where most of the arrhythmias does not
present any risk immediately and happens usually in our daily life. Acute stroke is mainly
caused by atrial fibrillation and sudden shock or cardiac death is caused by mainly due to
ventricular tachycardia. According to statistics taken from worldwide by World Health
Organization, 15% of death has occurred due to arrhythmias and cardiovascular diseases are
the reason for 80% of sudden deaths throughout the world.
Among the cardiovascular diseases, Arrhythmia is most important cause for death hence aged
community people can be given continuous health care by utilizing wearable device for
monitoring and detection of unusual electrocardiogram (ECG) signals such that instant
warning message can be sent to hospital or concerned medical practitioner. Based on alert
message immediate care is given to the patient avoiding happening of tragedies. In this
invention several arrhythmias are focused for building an algorithm based on convolutional
neural network (CNN) for the classification of cardiac disease. This health care platform of
Artificial Intelligence (AI) involves IoT based wearable hardware, cloud database and user
interface application.
Summary of Invention
In this invention Artificial Intelligence of Things (AIoT) is proposed which focuses on real
time analysis of ECG signals continuously for detection of any arrhythmias leading to risk to
life. This system involves front end wearable device for sensing ECG, user interface
application for smart devices, AI based algorithm for analysis of cardiac disease, cloud
database for enabling real time detection, low consumption of power and longer duration of
complete system usage. The sensing hardware device includes analog front end circuits
consuming low power that includes integrated circuit for commercial power management
along with a commercial Bluetooth module.
Hardware device is self designed system on chip (SOC) that includes sigma-delta 10 bit
analog to digital converter, digital signal processing unit and a level shifter. Front end SOC
5
collects ECG signal from the user which is transmitted to the smart device APP instantly via
Bluetooth module that utilizes Bluetooth low energy 4.0. A single lead is provided with
wearable ECG monitoring device that is attached with two wet electrodes of silver chloride to
be fixed on the chest and it can be usable for 24 hours continuously under normal usage.
Smart device APP acts as a user interface which involves mainly three parts name.
Brief description of the system
In this invention, cardiac disease detection is focused using AIoT (Artificial Intelligence of
Things) based on classification of ECG signals.
Cardiac arrhythmias are of four categories namely normal ECG, atrial fibrillation, atrial
flutter and ventricular fibrillation.
Real time ECG signal will be displayed on screen which will be classified using
convolutional neural network based AI algorithm into different types of arrhythmias stored in
the cloud database.
We can also able to compute the classification on smart device itself where it will be able to
classify whether the ECG signal is normal or abnormal but not the type of arrhythmia or the
category where it falls.
Accurate classification will be done by cloud server which will be able to identify the type of
arrhythmia precisely based on the ECG signal of the user obtained from the wearable wireless
sensing device.
ECG signal will be stored in the local smart device transmitted through Bluetooth and also
will be uploaded to cloud database.
Encoding of signal is done for safe transmission of ECG signal such that time stamps are
added with it for maintenance of correctness.
Cloud server consist of big data database which involves three segments namely data storage,
web user interface and AI algorithm for classification.
6
Data packets are received in encoded form by data storage segment from front end smart
devices, converts data packets into decoded ECG signals.
Then these ECG signals will be stored separately in decoded form based on the measuring
time stamps and measured objects.
Web user interface acts as a clear information platform for patients, their care takers and
medical practitioners responsible for them.
These stored ECG signals will enable doctors to diagnose condition of the patient more
specifically while patient along with their family will be able to realize condition of the
patient on daily basis based on ECG signal.
Thirdly, unusual ECG signals are detected by AI based algorithm from a huge set of data in
few minutes using Convolutional neural network.
Generally one hundred thousand heartbeats will be produced by normal human per day where
most of them are normal while few abnormal.This makes the doctor to diagnose, a great
challenge from ECG data of long term hence our invention will be able to detect unusual
signals quickly using cloud platform based on artificial intelligence based algorithm which is
then displayed using web user interface.
AI based algorithm for classification of arrhythmia involves two segments namely data pre
processing and Convolutional neural network model.
Better feature learning is provided by CNN model hence conventional signal processing of
ECG signals such as feature extraction, R-peak, timefrequency analysis and QRS complex detection.
Pre-processing of data in AI algorithm involves three steps namely removal of noise, baseline
removal and generation of image.
First step involves noise removal is done by 8-point moving average filter.
7
Moving average filter involves finite size windows using which convolution is taken along
with the ECG signal obtained from the user.
Also average of output signal is taken in the filter for reducing the range of discrete time
noise in turn enhances the peak value of identification.
Description of the system
The concept of this invention is such that parametric curve is placed for approaching the
location of known data set followed by subtraction of fitting parametric signal from the
original signal for obtaining ECG data with removal of baseline drift. Image generation is
done based on design that imitates how cardiologists are able to ECG signals. Whenever
abnormal ECG occur sustained analysis and non – sustained analysis is done by cardiologist
on ECG.
Cardiologist note unusual ECG signal whenever 3 abnormal heartbeats occurs consecutively
which last more than 30 seconds.
Based on above concept, pre-processing of ECG signal is converted into image with a sample
rate of 1200 Hz with time slot of 4.8 seconds.
These images will be used by CNN model for training process, validation and testing of data
in the next stage where we implement one dimensional CNN for extraction of features and
classification.
CNN model can also be realized using digital IC design. CNN model proposed in this
invention involves four convolutional layers along with three fully connected layers.
Each of the convolutional layer is followed by a leaky rectified linear unit (leaky ReLU)
acting as an active function.
These leaky ReLU have both benefits of traditional ReLU and also prevents several neurons
from dead ReLU. Also partially input data is negative consisting of some vital features hence
should not be ignored.
8
Hence due to these reasons leaky ReLU is selected as active function.
Max pooling for this model involves stride equal to 1 for extracting more precise features.
Also same padding is taken for saving the edge information from each of the input image.
Then three fully connected layers makes the output neurons count from 100 to 10 which is
again shrinks these 10 neurons to 4 divisions as output.
Order of the filter is set as 10 at the first convolutional layer which is then doubled at each of
the following layer.
Desired features are extracted by setting the order of all filter kernel to 180.
This model chooses its optimizer as gradient descent for optimization. Weight decay
parameter, learning rate decay and learning rate of the model is set to 0.0001, 0.9 and 0.1
respectively after every epoch.

We Claims:
1. This invention focuses on platform of Artificial intelligence of Things integrated
for health care system that includes hardware, software and storage cloud database.
2. Artificial Intelligence based algorithm is used for classification of arrhythmia
based on professional advice of cardiologist.
3. ECG of user is sensor by the wearable device and transmitted to Smart device App
using Bluetooth module.
4. IoT is utilized to transmit the sensed data from smart device to cloud database for
classification process.
5. AI based algorithm will be able to classify arrhythmia into four categories namely
normal ECG, atrial fibrillation, atrial flutter and ventricular fibrillation.