The document discusses ECG signal analysis and abnormality detection using artificial neural networks. It defines normal and abnormal ECG signals, describing abnormalities like bradycardia and tachycardia. Two algorithms are described for detecting abnormalities: one analyzes heart rate and the other detects general heart diseases. An ANN system is used for ECG analysis and classification, taking spectral entropy, Poincare plot geometry, and largest Lyapunov exponent as inputs to classify eight cardiac conditions.
CCS355 Neural Network & Deep Learning UNIT III notes and Question bank .pdf
ECG Abnormality Detection & ANN Classification
1. Presentation On
Detection of ECG abnormalities and ANN – based ECG analysis
System.
Course Name: Biomedical Signal Processing
Course Code: ECE 5315
Submitted To:
Dr. Md. Maniruzzaman
Professor
ELECTRONICS AND COMMUNICATION ENGINEERING DISCIPLINE
Submitted By:
Abdul Kader
Student ID: M.Sc. 170909
1 st Year , Term- I
KHULNA UNIVERSITY, KHULNA -9208
3. ECG signal
• ECG-ElectroCardioGram is one of the famous diagnosing tools which
measure the electrical activities of the heart and record the details.
• The ECG works mostly by detecting and amplifying tiny electrical
changes on the skin that are caused when the heart muscle
depolarizes each heart beat.
4. Normal ECG Signal
• Frequency : 0.05 Hz to 100 Hz
• Duty range : 1 mV to 10mV
• The number of peaks and valleys in the ECG is represented by the alphabets as P,
Q, R, S and T.
Table : Values of the normal heart rate.
5. Abnormal ECG Signal
• The normal value of the heart beat : 60 to 100 beats/ minutes.
• Bradycardia = heart beat < 60 beats/min (slow heart beat)
• Tachycardia = heart beat > 100 beats/min ( high heart beat)
• If the cycle space is not even then it indicates an arrhythmia.
Arrhythmia is indicated by verifying the cycles.
6. Abnormalities Detection Algorithms
Generally two detection algorithms were built. These are :
i) heart rate related abnormalities algorithm and
ii) General heart diseases detection algorithm
Heart rate - related abnormalities detection algorithm:
The algorithm tests three consecutive ECG samples and compares a given sample to
its adjacent counterparts and to a threshold value of 0.25 volts. The fixed threshold
of 0.25 volts was chosen based on the following:
1. The selected ECG waveforms have a common range of amplitudes (0 to 3 volts)
of the samples making the idea of a fixed threshold practical as shown in Fig.
2. The detection mechanism does not rely on the amplitude; however it depends on
the number of the “R” peaks of the QRS complexes in the whole waveforms.
7. Fig. common range of samples amplitudes (0 to 3 volts) for different ECG waveform
8. General heart diseases detection algorithm
For the detection general heart diseases , a mechanism was developed to identify
any unusual drop in the voltage between the P wave and the QRS complex in
addition to the peak detection.
i) AV block detection
ii) Ventricular Fibrillation detection
iii) Sudden Cardiac Death Detection
9. AV Block Detection
The AV block is mainly characterized by a drop in the voltage between the P wave
and the QRS complex due to the latency in the signal propagation between the atria
and the ventricles through the AV bundle. The algorithm tests three consecutive
ECG samples and compares a given sample to its adjacent counterparts and to a
threshold value of -2.5 volts. The threshold of -2.5 volts was chosen to detect the
third degree AV block, because the drop in voltage is the main characteristic of such
conditions as clearly shown in Fig. A voltage–drop count that exceeds 60 is an
indication of a third degree AV Block.
Fig. voltage drop in third degree AV block ECG waveform
10. Ventricular Fibrillation Condition
The VF is mainly characterized by the absence of peaks and drops with large magnitudes, the
samples magnitude ranges from .3 volts to -.15 volts only. The algorithm tests if any given sample
lies outside the interval [-0.5, 0.5]. The samples of the VF ECG waveform have narrow range of
magnitudes as shown in Fig., therefore the threshold of ± .5 volts was chosen. If the number of peaks
counted is equal to 0, it is concluded that the patient suffers from Ventricular Fibrillation.
Fig. VF Ventricular Fibrillation ECG waveform with a narrow range of amplitudes
11. Sudden Cardiac Death Detection
The SCD is mainly characterized by the absence of peaks and drops along with a semi-linear
behavior of the signal. The algorithm tests if any given sample lies outside the range [- 0.15,0.15].
The threshold of ± .15 is very adequate to the SCD condition, as the absence of the electrical activity
of the heart on the second half of the ECG waveform has dramatically narrowed the range of the
amplitudes of the samples as shown in Fig.
Fig. Absence of the electrical activity of the heart on the second half of the ECG
waveform in the SCD Condition
12. ANN based ECG analysis system
• Artificial neural networks (ANNs) are biologically inspired networks
that are useful in application areas such as pattern recognition,
classification etc. The decision making process of the ANN is holistic,
based on the features of input patterns and is suitable for classification
of biomedical data.
• The ANN used for classification is shown in Fig. The input layer
consisted of nodes and in the subsequent hidden layers, process
neurons with the standard sigmoid activation function were used. The
output layer had three neurons, to divide the output domain into eight
classes (000 to 111).
14. Disease Classification using ANN
The cardiac disorders were classified into eight categories, namely
• left bundle branch block (LBBB)
• normal sinus rhythm (NSR)
• periventricular contraction (PVC)
• atrial fibrillation (AF)
• ventricular fibrillation(VF)
• complete heart block (CHB)
• ischemic/dilated cardiomyopathy
• sick sinus syndrome (SSS).
The ANN classifier was fed by three parameters derived from the heart rate signals.
They were :
spectral entropy
Poincare plot geometry and
largest Lyapunov exponent (LLE).
15. Spectral Entropy
Spectral entropy quantifies the spectral complexity of the time series. A variety of
spectral transformations exist. Of these, the Fourier transformation (FT) is the most
commonly used technique from which the power spectral density (PSD) can be
obtained. The PSD represents the distribution of power as a function of frequency.
Normalization of the PSD with respect to the total spectral power yields the
probability density function (PDF). Application of Shannon's channel entropy gives
an estimate of the spectral entropy of the process, where entropy is given by
where pf is the PDF value at frequency f.
The entropy is interpreted as a measure of uncertainty about the event at f. Thus
entropy can be used as a measure of system complexity. The spectral entropy
H(0<=H<=1) describes the complexity of the HRV signal. This spectral entropy H
was computed for the various types of cardiac signal.
16. Poincare Plot Geometry
Poincare plot geometry, a technique taken from non-linear dynamics, explains the
nature of R-R interval fluctuations, it is a graph of each R-R interval plotted against
the next interval. Poincare plot analysis is an emerging quantitative-visual technique
whereby the shape of the plot is categorized into functional classes that indicate the
degree of heart failure in a subject. Using plot we can obtain the summary
information as well as detailed beat-to-beat information on the behavior of the heart.
The Poincare plot can be analyzed quantitatively by calculating the standard
deviations of the distances of the R-R(i) to the lines y = x and y = -x ÷ 2*R-Rm,
where R-Rm is the mean of all R-R(i).
SD1 and SD2 are referred to as standard deviations. SD1 related to the fast beat-to-
beat variability in the data, and SD2 described the longer-term variability of R-R(i) .
The ratio SD1/SD2 can also be computed to describe the relationship between these
components.
18. Largest Lyapunov exponent
The Lyapunov exponent λ is a measure of the rate at which the trajectories separate
one from another. A negative exponent implies that the orbits approach a common
fixed point. A zero exponent means the orbits maintain their relative positions; they
are on a stable attractor. Finally, a positive exponent implies that the orbits are on a
chaotic attractor. For two points in a space X0 and X0 +Δ X0, that are function of
time and each of which will generate an orbit in that space using some equations or
system of equations, then the separation between the two orbits Δx will also be a
function of time. For a chaotic data set, the function Δx (X0,t) will behave
unpredictably. The mean exponential rate of divergence of two initially close orbits
is characterized by the form Δx (X0,t).
19. The Lyapunov exponent λ is useful for distinguishing various orbits. The
LLE is estimated using a least squares fit to an average line defined by
where di(n) is the distance between the ith phase-space point and its
nearest neighbor at the nth time step. This last averaging step is the
main feature that allows an accurate evaluation of the LLE, even when
we have short and noisy data.
20. Conclusion
The ECG is mainly used for diagnosis of heart disease. Various
supervised and unsupervised Artificial Neural Network model have
been proposed in the literature for ECG signals feature extraction and
classification. The ANN classifier was fed by three parameters namely
spectral entropy, Poincare plot geometry and largest Lyapunov exponent
(LLE) derived from the heart rate signals are discussed.
22. Sample Question
• What is normal and abnormal ECG signal ?
• Describe the two type of abnormal detection algorithm ?
( heart rate related abnormalities algorithm and general heart disease
detection algorithm)
• What is ANN based ECG system?
• Describe ANN classifier form heart rate signal .