A SURVEY ON ECG SIGNAL ANALYSIS BASED ON FEATURE EXTRACTION AND CLASSIFICATION
B.Srimathi, Me-Applied Electronics, Department of Electronics and Communication, IFET College of Engineering, Gangarampalayam, [email protected]
Electrocardiogram (ECG) is a periodic signal mainly used for the detection and diagnosis of abnormalities in cardiac cycle. It is important to classify the electrocardiogram signals for the diagnosis of heart disease. The fluctuation in the heart rate noted by ECG due to cardiac demand, also influenced by the occurrence of arrhythmias, diabetes and other cardiac disorders. In this work, I survey the different techniques of features extraction for ECG signal extraction from Arrhythmia database, and extracted features used to classify the ECG signals.
KEYWORDS: Electrocardiogram; features extraction; MIT-BIH Arrhythmia database.
Electrocardiography (ECG or EKG) is the process of recording the electrical activity of the heart using electrodes placed on the skin over a short period of time. The recorded ECG waveform is the electrocardiograph and the instrument is termed as the electrocardiogram. The electrodes which placed on the skin detect the electrical changes from the heart muscle’s electrophysiologic pattern of depolarizing and repolarizing during each heartbeat. This is a commonly performed cardiology test provides valuable information about cardiac disorders.
ECG is a technique that uses ultra sound waves to image the interior of the heart The test is usually done by using an ultrasound transducer (probe) placed on the skin of the chest over the heart. In some cases, a small probe is passed down the oesophagus. The frequency range of electrocardiogram signal varies from 0 HZ-300 Hz. The most information present in the range of 0.5 Hz-150 Hz. It is necessary to remove the higher frequencies which eliminate the unwanted signals, which result in the reduction of information less than 1% of the useful information.
Classification of ECG signals is a challenging problem because of the issues involved in the classification are lack of ECG signal features standardisation, variability in ECG features, individuality in ECG patterns, non existence of optimal classification rules variability in ECG waveforms of patients. Classifier in classifying arrhythmia on real time is also an issue in ECG classification. ECG classification has the steps namely preprocessing, feature extraction, feature normalization, and classification.
II. ECG ELECTRODES AND LEAD SYSTEMS
Electrode is a conductive pad which makes the body in contact with an electrical circuit with the electrocardiograph. A lead is not same as the electrode which is a connector to an electrode.. The ECG lead system is of two types namely Bipolar lead system and Unipolar lead system. The placement of electrodes areV1 – 4th Intercostal space to the right of the sternum
V2 – 4th Intercostal space to the left of the sternum
V3 – midway between V2 and V4
V4 – 5th Intercostal space at the midclavicular line
V5 – Same level as V4 on anterior auxiliary line
V6 – Same level as V4 on mid-auxiliary line
Fig1: Electrodes and Leads
BIPOLAR LEAD SYSTEM:
It is also known as “Einthoven leads”. In this lead system, ECG is recorded by using two electrodes as the difference of electrical potential between these two electrodes. In bipolar lead system, electrodes are placed in four different places in left arm (LA), left leg (LL), right arm (RA), right leg (RL). Usually the electrode placed in right leg act as a ground reference electrode. The positions to fix the electrode are
Lead I: Voltage between LA and RA is measured.
V1 = 0.53 mV
Lead II: Voltage between LL and RA is measured.
V2 = 0.71 mV
Lead III: Voltage between LL and LA is measured.
V3 = 0.38 mV
AUGUMENTED UNIPOLAR LEAD SYSTEM:
This system is introduced by Wilson, here voltage taken between single exploratory electrode and the central terminal. In this system, two equal resistors are connected to a pair of limb electrodes and the center point act as one terminal to measure the voltage
Lead aVR: aVR = – V1 – V3/2
Lead aVL: aVL = V1 – V2/2
Lead aVF: aVF = V2 – V1/2
Fig2: ECG waveform
The ECG waveform has the components which indicate the electrical activity of the heart includes
FEARURE DESCRIPTION DURATION
P wave Represents atrial depolarization spread from SA towards AV and from right atrium to the left atrium ;80ms
PR interval Measured from the beginning of the P wave to the beginning of the QRS complex. This interval reflects the time the electrical impulse take travel from sinus node through AV 120 to 200ms
QRS complex Represents the rapid depolarization of the right and left ventricles. The QRS complex has larger amplitude than P wave 80 to 100 ms
J point It is point at which the QRS complex finishes and the ST segment begins. ST segment The ST segment connects the QRS complex and the T wave; it represents the period when the ventricles are depolarized. T wave The T wave represents the repolarization of the ventricles. 160 ms
Corrected QT interval The QT interval is measured from the beginning of the QRS complex to the end of the T wave ;440 ms
U wave The U wave is hypothesized to be caused by the repolarization of the interventricular septum. It has low amplitude and often completely absent. III. LITERATURE SURVEY
Many researchers have been worked on classification of ECG signals to detect the abnormalities in the heart. The main intention of this literature survey is to study the techniques used for the classification and feature extraction of ECG signal and analyse the effective technique over another which provides the accurate results.
S.NO YEAR JOURNAL TITLE CONCEPT PERFORMANCE
1 2017 IEEE An Efficient wavelet based feature extraction scheme for ECG Signal Novel ECG feature extraction method based on wavelet Sensitivity and positive predictivity on MIT-BIH Arrhythmia database
2 2017 IEEE Design and Analysis of feature extraction algorithm for ECG signal using Adaptive threshold method Feature extraction method in time domain for ECG using adaptive threshold method to detect peak and heart condition Extraction of P,T wave, QRS complex, PR,QT,RR,ST interval and ST segment deviation
3 2017 Bio-medical engineering online ECG signal performance denoising assessment based on threshold timing of dual tree wavelet transform Wavelet decomposition on to evaluate ECG signal denoising performance
Result better than ordinary dual wavelet transform Robust for all noises used in real time ECG monitoring
4 2017 IEEE Prediction of heart disease using Hybrid technique for selecting feature Random forest and Native bayes
Select features before classification Improved accuracy and reduced computational time
5 2016 Computing in cardiology Abnormal heart sound detection based on sealed time-frequency representation and feature selection Feature selection method based on similarity to remove noise Overall score -85.4%
Sensitivity – 95.76%
Performance – 84%
6 2016 Bio-medical engineering international conference Feature extraction of ECG signal using Discrete sinc transform Extracting features based on completely modifying Mel-frequency ceptral coefficient Accuracy – 95.45%
Accuracy for DCT – 90.9%
7 2016 International conference on systems medicine and biology Kernel based feature extraction for patient adaptive ECG Beat classification Discriminate patients specific ECG beat using Kernel based feature extraction-Kernal Canonical Correlation Analysis KCCA features more effective on normal patients than arrhythmia patients
8 2015 Journal of Applied Research and Technology Feature extraction of ECG signal by applying adaptive threshold and principal component analysis Novel approach for QRS complex detection and extraction of ECG signal for different types of arrhythmia Hilbert transform and adaptive threshold technique approach Sensitivity – 92.86%
Predictivity – 99.7%
9 2012 IEEE A time domain morphology and gradient based algorithm for ECG feature extraction Combination of extrema detection and slope information with adaptive thresholding TDMG algorithm perform accurately irrespective of lead choosen, different disease categories, sampling frequency of ECG signal
10 2010 IEEE international conference Arrhythmia detection and classification using Morphological and dynamic features of ECG signal Approach based on combination of morphological and dynamic features Yield accuracy of 99.66% of 85945 heart
11 2006 IEEE International Conference Unsupervised segmentation of cardiac PET transmission images for automatic heart volume extraction Automatic method to extract heart volume from cardiac PET transmission images
Automatic 3D segmentation using MRF MRF segmentation results were of good quality in all cases and able to extract the heart volume from all the images
IV. TECHNIQUES USED FOR EXTRACTION AND CLASSIFICATION
Wavelet transform performs a signal analysis when frequency changes over time. It provide more precise information about signal data than other techniques.
Wavelet decomposition decompose an image into various sub images. It is called reconstruction of images using decomposed or analysis of applied noised image.
Adaptive threshold is unlike fixed threshold, where threshold value is depends on each pixel on neighouring pixel intensities.
Random forest/ Random Decision forests is a learning algorithm for classification, regression and other tasks which operate by constructing a multimedia of decision trees at training time.
Naïve bayes is a classifier of the family of simple probabilities classifier based on applying Bayes theorem with strong independence assumption between features.
Mel-frequency cepstrum is a representation of short term power spectrum of a sound, based on a linear cosine transform of a log power spectrum on a non linear Mel scale of frequency.
Kernel canonical correlation analysis is a method extends the classical linear canonical correlation analysis to a general non linear setting via a kernelization procedure.
This study has a detailed survey of different techniques used for ECG feature extraction and classification to efficiently analyse the signal information. Each technique has the different method to extract the features and classifiers to classify the ECG signal based on the tracted features. All techniques extract the features to analyse the ECG signal. Each technique provide different and better accuracy, predictivity and sensitivity of the signal from Arrhythmia database.
1 Mohammad Sarfraz, Ateeq Ahmed Khan, Francis F. Li, “Using Independent Component Analysis to Obtain Feature Space for Reliable ECG Arrhythmia Classification,” IEEE International Conference on Bioinformatics and Biomedicine. 2014, pp. 62-67.
2″A new QRS detection algorithm based on empirical mode decomposition,” in Bioinformatics and Biomedical Engineering, 2008. ICBBE 2008. The 2nd International Conference on, 2008, pp. 693–696.
3 X. Jiang et al., “ECG Arrhythmias Recognition System Based on Independent Component Analysis Feature Extraction,” IEEE TENCON 2006.
4 c. Li, C. Zheng, and C. Tai, “Detection of ecg characteristic points using wavelet transforms,” IEEE Trans Biomed Eng, vol. 42, no. I, pp. 21 -28, jan. 1995.
5 Liang H.Y., Sakari L., Iiro H.:” A heart sound segmentation algorithm using wavelet decomposition and reconstruction”. IEEE EMBC, Chicago, IL,USA, October 1997.
6 M. Suchetha, N. Kumaravel, B. Benisha, “De-noising and Arrhythmia Classification using EMD based features and Neural Network”,Proceedings of IEEE International conference on Communication and SignalProcessing, India, April 3-5, 2013, pp 883-887.
7 R. Harikumar and S.N. Shivappriya, “A Novel Approach for Different Morphological Characterization of ECG Signal”, Proceedings of the Fourth International Conference on Signal and Image Processing 2012 (ICSIP 2012), Lecture Notes in Electrical Engineering 221, Springer India 2013, pp 13-23, DOI: 10.1007/978-81-322-0997-3_2.
8 Ivaylo I Christov, Real time electrocardiogram QRS detection using combined adaptive threshold, Biomedical Engineering On Line 2004, 3:28 .9 Feature extraction of electrocardiogram signals by applying adaptive threshold and principal component analysis(R. Rodrígueza, A. Mexicanob, J. Bilac, S. Cervantesd, R. Ponceb) .
10 J. Sahambi, S. Tandon, and R. Bhatt, “Using wavelet transforms for ecg characterization. an on-line digital signal processing system,” IEEE Eng Med Bioi Mag, vol. 16, no. I, pp. 77 -83, jan.-feb. 1997.