=Paper=
{{Paper
|id=Vol-2563/aics_33
|storemode=property
|title=Arrhythmia Detection in ECG Signals Using a Multilayer Perceptron Network
|pdfUrl=https://ceur-ws.org/Vol-2563/aics_33.pdf
|volume=Vol-2563
|authors=Gaurav Kumar,Urja Pawar,Ruairi O'Reilly
|dblpUrl=https://dblp.org/rec/conf/aics/KumarPO19
}}
==Arrhythmia Detection in ECG Signals Using a Multilayer Perceptron Network==
https://ceur-ws.org/Vol-2563/aics_33.pdf
Arrhythmia Detection in ECG Signals Using a
Multilayer Perceptron Network
Gaurav Kumar, Urja Pawar and Ruairi O’Reilly
Cork Institute of Technology, Ireland,
gaurav.kumar@mycit.ie, urja.pawar@mycit.ie, ruairi.oreilly@cit.ie
Abstract. Electrocardiography (ECG) is a form of physiological data
used to record the electrical activity of the heart. Numerous researchers
have proposed and developed methods to extract features from the ECG
signal (for example, R-R segment, P-R segment). These features can be
used to analyse and classify various forms of heart arrhythmia.
In this work, a method for ECG classification that employs a generalised
signal pre-processing technique and uses a Multi-Layer Perceptron net-
work to classify arrhythmia per the AAMI EC57 standard accurately
is presented. The method is trained and evaluated using PhysioNet’s
MIT-BIH dataset, and an average accuracy of 98.72% is achieved. The
proposed methodology is comparable to state-of-the-art CNN models,
both in terms of accuracy and efficiency.
Keywords: Arrhythmia Classification, Multi-Layer Perceptron, Convolutional
Neural Networks
1 Introduction
An Electrocardiogram (ECG) is a time-series signal used for recording the elec-
trical activity of the heart. ECG recordings require a cardiologist to interpret
and detect cardiac abnormalities or arrhythmia. A typical heart beats in a steady
rhythm. A heartbeat varies across individuals and within individuals depending
on a variety of conditions. The segments of a standard ECG signal consist of
waveforms like P, Q, R, S, T and U as depicted in Figure 1.
The QRS complex which represents ventricular depolarisation and contrac-
tion typically begins with a downward movement and is composed of Q, R and
S waves where the Q wave is a larger upwards deflection, a peak at R and then a
downwards S wave as depicted in Figure 1. The PR segment or interval indicates
the time endured by the wave to travel from the sinus node to the ventricles.
The RR interval represents the time between successive QRS complexes and is
used to calculate heart rate.
Electrocardiography (ECG) monitoring is used in diagnosing and treating
patients with heart disorders. In order to detect and precisely categorise abnor-
mal beats in an ECG signal, high-level expertise in the domain are required.
This requirement introduces several constraints concerning the expert analysis
of ECG data: i) It is time-consuming and prone to human errors; ii) There are a
Copyright © 2019 for this paper by its authors. Use permitted under Creative Commons License
Attribution 4.0 International (CC BY 4.0).
Fig. 1. ECG intervals and segments depict-
ing QRS complex [1].
Fig. 2. Normal Vrs Arrhythmic ECG Wave.
limited number of expert cardiologists available to diagnose the millions of pa-
tients suffering from heart disorders; iii) The cost of diagnosis is expensive. These
constraints highlight the need for a reliable and low-cost means of analysing and
diagnosing an individual’s cardiac health [16].
In addressing this need, numerous researchers have investigated the applica-
tion of machine learning techniques to ECG in order to automate the detection
of abnormalities in ECG signals [6] [5]. The resultant models have demonstrated
precision in identifying and classifying the wave morphologies of ECG signals
which plays a significant role in the detection of abnormalities.
In this work, a Multi-Layer Perceptron (MLP) model is proposed which is
trained on a pre-processed version of PhysioNet’s MIT-BIH dataset. The trained
model has achieved an accuracy of 98.72%, which is comparable to state-of-the-
art methods [10][2][12] in ECG classification. It is envisaged that this model will
enable a sufficiently accurate analysis with a low computational cost such that
analysis can be carried out in real-time on low-end devices. The model could,
therefore, contribute to the incorporation of automated analysis solutions for
conditions such as heart disorders into devices such as activity trackers with the
intent of reducing the health impact on the general public.
2 Related Work
Several machine learning algorithms have been proposed and adapted for the
accurate classification of ECG data, an excellent overview of which is presented
in [15]. This section demonstrates the most relevant techniques and their appli-
cations in the field of ECG classification.
Application of Artificial Neural Networks (ANN) In this paper, [11],
sufficient accuracy was achieved with a short learning time. A new arrhythmia
classification algorithm was proposed, which had a fast learning speed and high
accuracy by making use of Morphology Filter, Principle Component Analysis
(PCA) and Extreme Learning Machine (ELM). The accuracy levels 98.00% in
terms of average sensitivity, 97.95% in terms of average specificity was achieved.
Additionally, a comparative study was performed in terms of learning rate using
an ELM, comparing back propagation neural network (BPNN), radial basis func-
tion network (RBFN) and support vector machines (SVM). It was observed that
the learning time of the proposed algorithm using ELM was about 290, 70, and
3 times faster than an algorithm using a BPNN, RBFN and SVM, respectively.
Vishwa et al. [17] implemented an ANN (Artificial Neural Network) based
classification system to detect heart disorders through ECG analysis using es-
timated feed-forward ANN and back-propagation learning algorithms. The ap-
proach was performed on a subset of arrhythmia classes in the MIT-BIH database
resulting in an accuracy of 96.77%.
In order to classify ECG data, automatic extraction of both time interval
and morphological features was carried out in [3]. Linear Discriminant Analysis
(LDA) and Artificial Neural Networks (ANN) were used for classification. The
ANN (in the form of an MLP) proved to be the more accurate of the two clas-
sifiers with a training accuracy of 85.07% and 70.15% on unseen data. Principal
Component Analysis (PCA) was used for feature selection and dimensionality
reduction.
Jadhav et al. [9] used Modular Neural Networks (MNN) to classify ECG
signals into normal and abnormal classes. The UCI arrhythmia dataset was used
for this experiment. The hidden layers in the network were varied, and the model
was trained on different subsets of the training data. The model was capable of
achieving an accuracy of 82.22% on the unseen or test dataset.
Application of Convolutional Neural Networks In [2], researchers used
Convolutional Neural Network (CNN) to detect arrhythmia in ECG heartbeats.
The CNN was trained using the MIT-BIH dataset. An accuracy of 93.5% was
achieved. The dataset was pre-processed by: i) Removing noise from the ECG
signals with the help of wavelet filters; ii) Segmenting the ECG signal into R-
peak beats and; iii) Normalising each segment to scale the amplitude of the
beat. The annotations in the dataset were divided into five categories namely:
non-ectopic (N), supraventricular ectopic (S), ventricular ectopic (V), fusion (F),
and unknown (Q) as per the Association for the Advancement of Medical In-
strumentation (AAMI) standard [13].
Data Augmentation was practised to address the imbalance in the dataset.
Synthetic data was generated for the minority classes to prevent the model from
over-fitting on the majority class. The CNN model contained one input layer,
one output layer and eight hidden layers of Convolutional, max-pooling and
fully-connected layers.
In [10], researchers extracted R-R features from the MIT-BIH dataset and
used these features as input to the model. The data is then subjected to a series
of convolution layers applying 1-D convolution. The predictor network consists of
five residual blocks, followed by two fully-connected layers and a softmax layer to
predict output class probabilities. Each residual block contains two convolutional
layers, two ReLU activation layers, a residual skip connection, and a pooling
layer. In total, the resulting network is a deep network consisting of 13 weight
layers.
In a residual block, a layer can either feed the data into the next layer or the
layers 2-3 steps away. In other words, the model may train the layers in a residual
block or may skip the training of those layers by using skip connection. This
ability makes the model more dynamic and overcomes some of the drawbacks of
having extra layers (such as over-fitting and slow learning) in the network.
The TensorFlow computational library for model training and evaluation was
employed. For the softmax layer, cross-entropy was used as the loss function.
Adam optimiser was used to train the network, with the learning rate, beta-
1 and beta-2 of 0.001, 0.9, and 0.999, respectively. Learning rate is decayed
exponentially with the decay factor of 0.75 every 10000 iterations.
The performance of the arrhythmia classifier was tested on 4079 heartbeats
(about 819 from each class) which were not used in the network training phase.
Data augmentation was used to balance the number of beats in each category.
The final accuracy which they were able to achieve was 93.4% on the MIT-BIH
arrhythmia dataset.
The learned representations or weights (filters in the case of a CNN) were
used to classify MI in the PTB Diagnostic dataset. This experiment involved
freezing the learned weights till the last convolution layer of the CNN network
and training only the last two fully-connected layers with 32 neurons each. The
model achieved an accuracy, precision and recall of 95.9%, 95.2% and 95.1% in
MI classification, respectively. The network was trained for approximately two
hours on a GeForce GTX 1080Ti processor.
In summary, a variety of machine learning algorithms and their application
to ECG data is evident in the literature. CNN’s demonstrate state-of-the-art
accuracy but are expensive in terms of the computation time required to train.
To detect arrhythmias in real-time, the proposed model needs to be accurate as
well as computationally inexpensive. As such, a generalised signal pre-processing
approach, as described in [10], is adopted, and an MLP model is proposed to
classify ECG heartbeats accurately.
3 Methods
This section presents the design and implementation of the proposed MLP net-
work and its comparison with a state-of-the-art Convolutional Neural Network
(CNN) for ECG beat classification. An overview of the proposed MLP network
is depicted in Figure 3.
Fig. 3. Architecture of the proposed MLP network
The first layer is the input layer through which the data will be fed to the
network. The number of neurons in the input layer is 187 (equal to the num-
ber of features/columns in the data). The following network consists of 4 sets
of Fully-connected (Dense), Batch Normalization and ReLU activation layers.
Number of neurons in each of the fully-connected layers are 50, 150, 900 and
400, respectively. The last layer of the network is the fully-connected output
layer of 5 neurons (equal to the number of distinct classes in the dataset) with
an activation function of SoftMax (as this is a multi-class classification problem).
3.1 MIT-BIH Dataset
The MIT-BIH dataset [14] [8] consists of ECG recordings of 49 distinct subjects
recorded at the sampling rate of 360Hz. Each record contains a recording of 30
minutes from two leads namely modified limb lead II (MLII) and one out of the
modified leads V1, V2, V3, V4 or V5. The dataset contains more than 109,000
beats annotated individually, belonging to one of possible 15 beat types.
The R-R interval of the ECG signal is widely used in the literature for the
classification of ECG signals. M. Kachuee etal. [10] extracted the R-R intervals
from the ECG signals of the MIT-BIH dataset. These features are then used
for arrhythmia classification. The annotations available in the MIT-BIH dataset
contains five different beat categories as denoted in Table 1.
Category Annotations
N Normal, Left/Right bundle branch block, Atrial escape, Nodal escape
S Atrial premature, Aberrant atrial premature, Nodal premature, Supra-ventricular premature
V Premature ventricular contraction, Ventricular escape
F Fusion of ventricular and normal
Q Paced, Fusion of paced and normal, Unclassifiable
Table 1. Mapping between heartbeat annotations and AAMI EC57 [13] categories.
The researchers in [10] used 47 recordings for the experiment and down-
sampled the sampling frequency of the MIT-BIH dataset from 360Hz to 125Hz.
The steps followed to extract ECG beats from the original signal are: i) Extract-
ing the R-R intervals by splitting the original continuous ECG signal to windows
of 10 seconds and selecting a 10-second window from the signal; ii) Normalis-
ing the amplitude of the extracted signal to a range between zero and one; iii)
Extracted the set of local maximums with a threshold of 0.9 representing ECG
R-peaks and; iv) Padding the extracted R-R interval with zeros making sure
that all the extracted beats are of identical length.
The advantages of this pre-processing include: i) It is useful in extracting R-R
intervals from signals with distinct morphologies (shapes); ii) No filter is applied
to extract the beats that make an assumption about the signal morphology (for
example, Fourier filter makes an assumption that the actual signal frequencies
fall at low frequencies while noise at high); iii) All the extracted beats have iden-
tical length which is essential for being used as input to the successive processing
parts. The pre-processed MIT-BIH dataset has been made available on Kaggle
[7] by [10] and is used for training the proposed MLP network. It provides the
extracted R-R features from the dataset along with a predefined 80:20 split for
training and testing.
3.2 Pre-processing
The MIT-BIH training data is highly imbalanced i.e. the distribution of instances
per class is not uniform. The number of instances for the “Normal” class is
82.77% of the whole training dataset. Therefore, there is a high probability for
the model to over-fit or gets biased towards the majority class (“Normal”) and
generalises the other classes to be Normal as well. To prevent the problem of over-
fitting, Compute Class Weight function of Class Weight library with “balanced”
as a parameter is used. This function calculates the weights per class by weighing
classes inversely proportional to their frequency:
n
wj =
knj
Here, wj is the weight to class j, n is the number of observations, nj is the
number of observations in class j, and k is the total number of classes.
3.3 Modelling
To build the network Keras (a high-level neural network API), Sequential Model
is used. It is a linear stack of layers. Various layers can be added by specifying a
list of layer instances to the model. The input layer in Keras Sequential model is
defined by specifying the dimension of input data which in this case is 187 (the
number of columns in the dataset). The number of rows is not specified because
it may vary for the training and test dataset.
The first hidden layer of the network is a Dense layer with 50 neurons. The
weights of this layer are initialised with an Identity matrix with a multiplicative
factor (gain) of 1. A Batch Normalisation layer follows this. The Batch Normal-
isation layer is responsible for normalising the output of the hidden layer and
increasing the learning speed of the model. Finally, these normalised values are
passed to a ReLU activation layer which will decide based on polarity (negative
or positive) of the value whether the individual neuron is activated or not.
The activated output from the first hidden layer is then rendered to the next
three sets of Dense, Batch Normalisation and ReLU activation layers (succes-
sively) of the MLP network. The activated output from the last hidden layer is
assigned to the output SoftMax activation layer with five neurons to classify the
input data amongst one of the five distinct classes of the MIT-BIH dataset.
During the training phase of the network, the classifications on input data
made by the model are compared with the actual labels (classes) to compute
the training loss in each iteration. The function used to calculate the loss in
the proposed network is Sparse Categorical Cross-Entropy, as it is a multi-class
classification problem and only one label or class is applicable per instance.
Weights of hidden layers are updated or tuned by an Adam optimiser after the
training loss has been calculated. Tuning helps to decrease the overall training
loss in the next iteration or epoch of training. The learning rate (magnitude by
which the weights are updated) used to update the weights is 0.001.
Apart from calculating loss and optimising weights, the model also evaluates
training performance in each epoch, i.e. determining the number of correctly
classified instances. The metric used for assessing the model’s performance is
Accuracy.
Loss, Optimizer and Metric are specified while compiling the model. The
compiled model is then trained on the training set (containing both the feature
and label data) of 87,554 instances of ECG beats for 100 epochs. At each epoch
predictions made on the validation set (also known as the test set) are evaluated;
this validates the performance of the model. The validation set contains 21,892
ECG beats and is not used in the training phase of the model.
3.4 Architectural Design of the Proposed MLP and the CNN
The proposed MLP network was then compared with a state-of-the-art CNN as
presented in [10]. This CNN was chosen for comparison as it uses the generalised
signal pre-processing technique without any form of filters applied. Therefore,
both networks use the same pre-processed MIT-BIH dataset for training and
validation of the models and so a comparative analysis can be derived.
The significant difference between the two networks is in their architectural
design. The proposed MLP network employs dense or fully-connected layers to
process the input data while CNN makes use of convolutional layers. There
are in total six layers in the MLP network (one input layer, one output layer
and four hidden weighted layers - see Figure 3), whereas the CNN architecture
contains 15 layers (including input and output layer) of which 13 are weighted
(11 convolutional and two fully-connected).
All convolutional layers in the CNN network apply 1-D convolution, and
each layer has 32 kernels or filters of size five, and the two fully-connected layers
have 32 neurons each. However, the four dense hidden layers of the proposed
MLP network have 50, 150, 900 and 400 neurons, respectively. ReLU (Rectified
Linear Unit) activation function is utilised to activate the neurons or filters in
both networks. The weights of the first dense hidden layer of the MLP network
are explicitly initialised by an identity matrix with a multiplicative factor of 1,
whereas the Kernels or filters initialiser in the CNN model is not specified1 .
A batch normalisation layer (BNL) is employed after each hidden layer to
normalise the weighted sum output of each dense hidden layer in the proposed
network. The BNL helped the network to train faster and prevent over-fitting.
No such standardisation technique appears to have been applied in the CNN
network rather 5 Max-Pooling layers one in each residual block is practised.
The output layer of both the networks contains five fully-connected neurons
with SoftMax activation function to classify the given instance amongst one
of the possible five classes. Both the proposed MLP and the CNN is compiled
by employing Accuracy, Adam and Categorical Cross Entropy as the metric,
optimiser and the loss function respectively.
4 Results
This section details the evaluation and testing of the proposed MLP network
and its comparison to a state-of-the-art CNN network.
4.1 MLP: Mini-Batch Training
Propagating the whole training set in a neural network in each iteration (epoch)
is referred to as batch training. It typically increases memory consumption and
1
Authors of [10] e-mailed querying implementation details. No reply to date.
the time to train the model. To address this, mini-batch training is practised.
Mini-batch training propagates fixed subsets of the training data through the
network one by one. For instance, if the training data has 32,000 instances, and
the mini-batch size is 32, then there will be 1000 mini-batches.
The advantages of using mini-batch training are: i) Reduced memory con-
sumption as only one batch of training data is loaded in the memory at a time;
ii) Reduced training time. The network weights are updated with each propaga-
tion of a mini-batch while the weights are updated only once per epoch in batch
training. The default batch size of the Sequential Model API is 32. Depending
on the size of the training data, the batch size can be altered.
Mini-Batch testing on MLP Network
The proposed MLP network is trained with various batch sizes for 50 epochs.
The performance of the model is evaluated based on the time taken to train and
accuracy achieved on the validation or test dataset. Using the batch size of
512 yielded the best performance of the MLP model in terms of accuracy and
speed. The accuracy achieved on validation dataset is 98.28% and the model
took 3.4 minutes to train. Table 2 denotes the performance of the MLP network
for different batch sizes.
Batch Size Training time Accuracy F1 score
32 39.2 min 97.69% 90.64%
64 20.8 min 97.92% 91.46%
128 11.2 min 98.13% 91.73%
256 6.5 min 98.21% 92.05%
512 3.4 min 98.28% 92.24%
Table 2. Results for the MLP network over different batch sizes with regard to training
time accuracy and F1 score.
4.2 MLP: Kernel Initialiser testing on the MLP network
Kernel initialiser (also known as weights initialiser) is a technique that helps to
assign initial values to the weights of hidden layers in the network. By default,
the weights initialiser for a Dense layer is Glorot-Uniform. This technique draws
samples from a uniform distribution within -limit to +limit where the limit is
defined as:
r
6
limit =
f an − in + f an − out
Here, fan-in is the number of neurons in the previous layer and fan-out is
the number of neurons in the current later. The MLP network is tested on
various kernel initialiser techniques like glorot-uniform, glorot-normal, identity,
orthogonal, and random-uniform.
1. Glorot-Normal: This technique draws samples from a truncated normal
distribution centered on 0 with standard deviation defined as:
r
2
stddev =
f an − in + f an − out
Here, fan-in is the number of neurons in the previous layer and fan-out is
the number of neurons in the current later.
2. Identity: This technique generates an identity matrix of weights. It is used
only for 2-Dimensional matrices. If the resulting matrix is not square, it pads
the additional rows/columns with zeros.
3. Orthogonal: This technique generates a random orthogonal matrix of weights.
4. Random-Uniform: This technique initialises the weights with a uniform
distribution. It takes three arguments: (minval: the lower bound of the range
of random values to generate, maxval: the upper bound of the range of
random values to generate and seed: A seed is a python integer used to seed
the random generator. Mainly used for the similar production of values.)
Initializer Accuracy F1 score
Glorot Normal 97.80% 90.11%
Orthogonal 98.22% 91.90%
Random Uniform 97.86% 90.54%
Identity 98.43% 92.46%
Table 3. Evaluating the performance of the MLP network on different weight initial-
izers.
Out of all the above-mentioned kernel initialisers, identity initialisation pro-
duced the best results. An accuracy of 98.43% is achieved on the validation
dataset. The batch size used for this testing is 512. Table 3 denotes the perfor-
mance of the different initializers evaluated on the MLP network.
4.3 MLP: Gradual Decay in Learning Rate
This technique is used to reduce the learning rate of the model while training.
It monitors a metric or quantity, and if no improvement is seen for X number
of epochs, the learning rate is reduced. Validation accuracy is monitored during
the training of MLP network, and if no improvement in the validation accu-
racy is observed for five epochs, the learning rate is decreased by a factor of 1.
Reducing the learning rate by introducing a gradual decay improved the overall
performance of the MLP network (With Gradual Decay — Accuracy: 98.72% F1
score: 93.20%, Without Gradual Decay — Accuracy: 98.43% F1 score: 92.46%).
4.4 Comparative Analysis of the MLP and CNN
To evaluate the performance of the proposed MLP network a state-of-the-art
CNN [10] was selected. The CNN was trained on the pre-processed MIT-BIH
dataset for 50 epochs and the associated validation dataset used to evaluate
the performance of the model. The model took approximately 28 minutes to
train and achieved an accuracy of 98.6% on the validation dataset. In [10], an
accuracy of 93.4% on the MIT-BIH dataset is reported. The increase of 5.4%
in the validation accuracy observed in this experimental is probably due to the
split used between the training and validation dataset.
A slight variant of the replicated CNN was also trained to enable a more
transparent comparison. The techniques utilised by the MLP model to yield
better performance such as Batch-Normalization layer, mini-batch learning, ker-
nel initialisation, class-weight computation, and reducing learning rate were also
utilised by this instance of the CNN, referred to as CNN-REP*.
Method Training time Accuracy F1 score Precision Recall
Proposed MLP 3.4 min 98.72% 93.20% 94.6% 91.0%
CNN-Rep 28.4 min 98.64% 92.58% 94% 91.4%
CNN-Rep* 15.9 min 98.79% 93.01% 94.4% 91.8%
Table 4. Evaluating the performance of the MLP and CNN networks in the classi-
fication of heartbeats. Note: CNN-Rep as replicated from [10] CNN-Rep* with the
performance enhancing techniques applied.
Table 4 denotes the results of the three networks, the proposed MLP, the
replicated CNN and the improved replicated CNN. It is evident that: i) The
MLP network is less computationally expensive when compared to both CNNs
and demonstrates a reduced training time; ii) The proposed MLP outperformed
the replicated CNN network in terms of validation accuracy and F1 score while it
did not outperform CNN-Rep*. Figure 4 depicts the validation loss and accuracy
graph for the MLP and CNN-Rep network. iii) The performance enhancement
techniques utilised by CNN-Rep* resulted in a 44% decreased in training time
when compared to CNN-Rep and a .15% improvement in accuracy. While the
CNN-Rep* demonstrated an accuracy .07% better than the proposed MLP, it
took 4.68 times longer to train.
Fig. 4. Training Loss, Accuracy and Validation Loss and Accuracy visualised for MLP
(Left) and the CNN (Right).
4.5 Classification of ECG data in real-time
This experiment intended to evaluate the classification of ECG data in real-time.
The results are indicative of a models suitability for real-time analysis. For this
experiment, five ECG beats are extracted from the validation dataset and joined
to make a continuous stream of ECG data. A window of size 187 is constructed,
and a sliding window is used to analyse the ECG signal by one frame (column)
width from left to right. At each interval, the window contains 187 bits of ECG
data. The data obtained on sliding the window is analysed and classified by the
network. The proposed MLP network slightly outperformed the CNN network
in terms of average prediction time. The average prediction time for the MLP
Network was 3.12ms and 4.3ms for the CNN Network [10]. Low-end devices, such
as activity trackers and smartwatches, have access to limited compute, memory
and storage. A model running on these devices will be competing for scarce re-
sources with the applications and services being utilised by the device. As such,
it is envisaged that the prediction time may vary, and this warrants the com-
putational complexity of the underlying model being considered as part of the
evaluation. A thorough assessment of trained models for arrhythmia detection
running on low-end and edge devices will be the subject of future work.
Note: All experiments were performed using Google Co-laboratory (RAM:
12GB, Disk: 358GB, GPU: Tesla K80) and Keras computational library [4] for
model training and evaluation.
5 Conclusion
In this paper, an MLP for the real-time detection of arrhythmia in ECG data
is presented. In order to enhance the performance of the proposed model tech-
niques including mini-batch training, gradual reduction in learning rate, batch-
normalisation, kernel initialisation, and class-weight computation, are imple-
mented. The performance of the resultant model is compared with a state-of-the-
art CNN. Classification of the ECG data in real-time is performed to compute
the average prediction time of the models.
The training and validation of the models were carried out using the MIT-
BIH dataset. The proposed MLP outperformed a replicated state-of-the-art CNN
in ECG beat classification. An average accuracy of 98.72% was achieved with an
average time of 3.12 milliseconds to classify an ECG beat in real-time .
The rationale for the comparative accuracy gains experienced by the MLP
is due to a combination of the pre-processing and/or implementation details
omitted from the CNN network. Another instance of the CNN was implemented
(CNN-Rep*) with the same hyper-parameters as those used by the MLP to
enable a more transparent comparison. CNN-Rep* outperformed the MLP con-
cerning accuracy (∼0.07%) but also required approximately 4.6 times the amount
of time to train.
The MLP network demonstrated itself as relatively computationally inex-
pensive approach and while this naturally implies it would take less time than a
CNN to classify an ECG beat in real-time it also highlights its appropriateness
for low-end devices, particularly with the level of accuracy demonstrated.
Acknowledgement: This material is based upon works supported by Science
Foundation Ireland under Grant No. SFI CRT 18/CRT/6222
References
1. A real time ECG signal processing application for arrhythmia detection on
portable devices - Scientific Figure on ResearchGate. Available from: https://www.
researchgate.net/figure/ECG-intervals-and-segments_fig1_321455361, ac-
cessed: 05-2019
2. Acharya, U.R., Oh, S.L., Hagiwara, Y., Tan, J.H., Adam, M., Gertych, A., San Tan,
R.: A deep convolutional neural network model to classify heartbeats. Computers
in biology and medicine 89, 389–396 (2017)
3. Alexakis, C., Nyongesa, H., Saatchi, R., Harris, N., Davies, C., Emery, C., Ireland,
R., Heller, S.: Feature extraction and classification of electrocardiogram (ecg) sig-
nals related to hypoglycaemia. In: Computers in Cardiology, 2003. pp. 537–540.
IEEE (2003)
4. Chollet, F., et al.: Keras (2015)
5. Dastjerdi, A.E., Kachuee, M., Shabany, M.: Non-invasive blood pressure estimation
using phonocardiogram. In: 2017 IEEE International Symposium on Circuits and
Systems (ISCAS). pp. 1–4. IEEE (2017)
6. Esmaili, A., Kachuee, M., Shabany, M.: Nonlinear cuffless blood pressure estima-
tion of healthy subjects using pulse transit time and arrival time. IEEE Transac-
tions on Instrumentation and Measurement 66(12), 3299–3308 (2017)
7. Fazeli, S.: ECG Heartbeat Categorization Dataset. https://www.kaggle.com/
shayanfazeli/heartbeat, accessed: 05-2019
8. Goldberger AL, Amaral LAN, G.L.H.J.I.P.M.R.M.J.M.G.P.C.K.S.H.: Physiobank,
physiotoolkit, and physionet: Components of a new research resource for complex
physiologic signals. IEEE Engineering in Medicine and Biology Magazine 101(23),
215–220 (2003)
9. Jadhav, S.M., Nalbalwar, S.L., Ghatol, A.A.: Modular neural network based ar-
rhythmia classification system using ecg signal data. International Journal of In-
formation Technology and Knowledge Management 4(1), 205–209 (2011)
10. Kachuee, M., Fazeli, S., Sarrafzadeh, M.: Ecg heartbeat classification: A deep trans-
ferable representation. In: 2018 IEEE International Conference on Healthcare In-
formatics (ICHI). pp. 443–444. IEEE (2018)
11. Kim, J., Shin, H.S., Shin, K., Lee, M.: Robust algorithm for arrhythmia classifi-
cation in ecg using extreme learning machine. Biomedical engineering online 8(1),
31 (2009)
12. Martis, R.J., Acharya, U.R., Lim, C.M., Mandana, K., Ray, A.K., Chakraborty,
C.: Application of higher order cumulant features for cardiac health diagnosis using
ecg signals. International journal of neural systems 23(04), 1350014 (2013)
13. for the Advancement of Medical Instrumentation, A., et al.: Testing and reporting
performance results of cardiac rhythm and st segment measurement algorithms.
ANSI/AAMI EC38 1998 (1998)
14. Moody, G.B., Mark, R.G.: The impact of the mit-bih arrhythmia database. IEEE
Engineering in Medicine and Biology Magazine 20(3), 45–50 (2001)
15. Roopa, C., Harish, B.: A survey on various machine learning approaches for ecg
analysis. International Journal of Computer Applications 163(9), 25–33 (2017)
16. Society, H.R.: Heart diseases and disorders. https://www.hrsonline.org/
Patient-Resources/Heart-Diseases-Disorders, accessed: 05-2019
17. Vishwa, A., Lal, M.K., Dixit, S., Vardwaj, P.: Clasification of arrhythmic ecg data
using machine learning techniques. IJIMAI 1(4), 67–70 (2011)