ANN Based Classification for Heart Defibrillators 
M. Jabri, S. Pickard, P. Leong, Z. Chi, B. Flower, and Y. Xie 
Sydney University Electrical Engineering 
NSW 2006 Australia 
Abstract 
Current Inlxa-Cardia defibnllators make use of simple classification algo- 
rithms to determine patient conditions and subsequently to enable proper 
therapy. The simplicity is primarily due to the constraints on power dissipa- 
tion and area available for implememation. Sub-threshold implementation 
of artificial neural networks offer potential classifiers with higher perfor- 
mance than commercially available defibrillators. In this paper we explore 
several classifier architectures and discms micro-electromc implementation 
issues. 
1.0 INTRODUCTION 
Inlxa-Cardia Defibrillators (ICDs) represent an important therapy for people with heart dis- 
ease. These devices are implanted and perform three types of actiota: 
1.monitor the heart 
2.to pace the heart 
3.to apply high energy/high voltage electric shock 
They sense the electrical activity of the heart through leads attached to the heart tissue. Two 
types of sensing are commonly used: 
Single Chamber: Lead attached to the Right Ventricular Apex CRVA) 
Dual Chamber: An additional lead is attached to the High Right Atrium (HRA). 
The actions performed by defibrillators are based on the outcome of a classification procedure 
based on the heart rhythms of different heart diseases (abnormal rhythms or "arrhythmias"). 
637 
638 Jabri, Pickard, Leong, Chi, Flower, and Xie 
There are tens of different arrhythmias of interest to cardiologists. They are clustered into 
three groups according to the three therapies (actions) that ICEIs perform. 
Figure 1 shows an example of a Normal Sinus Rhythm. Note the regularity in the beats. Of 
interest to us is what is called the QRS complex which represents the electrical activity in the 
ventricle during a beat The R point represents the peak, and the distance between two heart 
beats is usually referred to as the RR interval. 
FIGURE 1. A Normal Sinus Rhhm (NSR) waveform 
Figure 2 shows an example of a Ventricular Tachycardia (more precisely a Ventricular Tachy- 
cardia Slow or VTS). Note that the beats are faster in comparison with an NSR. 
Ventricular Fibrillation (VF) is shown in Figure 3. Note the chaotic behavior and the absence 
of well defined heart beats. 
FIGURE 2. A Ventricular Tachycardia (VT) waveform 
The three waveforms discussed above are examples of Intra-Cardia Electro-Gmm.q (ICEG). 
NSR, VT and VF are representative of the type of action a defibfillator has to takes. For an 
NSR, an action of "continue momtoring" is used. For a VT an action of "pacing" is per- 
formed, whereas for VF a high energyPaigh voltage shock is issued. Because they are near- 
field signals, ICEGs are differera from external Eltro-Cardio-Grams (ECG). As a result, classi- 
fication algorithms developed for ECG pattems may not be necessarily valuable for ICEG rec- 
ognition. 
The difficulties in ICEG classification lie in that many arrhytlmas share similar features and 
fiy situations often need to be dealt with. For instance, many ICDs make use of the heart 
rate as a fundamental feature in the arrhythmia classification process. But several arrhythmias 
that require different type of therapeutic actions have similar heart rates. For example, a Sinus 
Tachycardia (ST) is an arrhythmia characterized with a heart rate that is higher than that of an 
NSR and in the vicinity of a VT. Many classifier would classify an ST as VT leading to a ther- 
apy of pacing, whereas an ST is supposed to be grouped under an NSR type of therapy. 
Another example is a fast VT which may be associated with heart rates that are indicative of 
ANN Based Classification for Heart Defibrillators 639 
VF. In this case the defibrillator would apply a VF type of therapy when only a VT type ther- 
apy is required (pacing). 
FIGURE 3. A Ventricular Fibrillation (VF) waveform. 
The overlap of the classes when only heart rate is used as the main classification feature high- 
lights the necessity of the comiderafion of further features with higher discrimination capabil- 
ities. Features that are commonly used in addition to the heart rate are: 
1.average heart rate (over a period of time) 
2.arrhythmia onset 
3.arrhythmia probability density functions 
Because of the limited power budget and area, arrhythmia classifiers in ICDs are kept 
extremely simple with respect to what could be achieved with more relaxed implememation 
constraints. As a result false positive (pacing or defibrillation when none is required) may be 
high and error rates may reach 13%. 
Artificial mural network techniques offer the potential of higher classification performance. In 
order to maintain as lower power consumption as possible, VLSI micro-power implementation 
techniques need to be considered. 
In this paper, we discms several classifier architectures and sub-threshold implememation 
techniques. Both single and dual chamber based classifications are considered. 
2.0 DATA 
Data used in our experiments were collected from Electro-Physiological Studies (EPS) at hos- 
pitals in Australia and the U'K. Altogether, and depending on whether single or dual chamber 
is considered, data from over 70 patients is available. Cardiologists from our commercial col- 
laborator have labelled this data. All tests were performed on a testing set that was not used in 
classifier building. Arrhythmias recorded dunng EPS are produced by stimulation. As a result, 
no natural transitions are captured. 
3.0 SINGLE CHAMBER CLASSIFICATION 
We have evaluated several approaches for single chamber classification. It is important to note 
here that in the case of single chamber, not all arrhythmias could be correctly classified (not 
even by human experts). This is because data from the RVA lead represents mainly the ventric- 
ular elecmcal activity, and many areal arrhythmias require atrial information for proper diag- 
nosis. 
640 Jabri, Pickard, Leong, Chi, Flower, and Xie 
3.1 MULTI-LAYER PERCEPTRONS 
Table 1 shows the performance of multi-layer percepttom 0MLP) trained using vanilla back- 
propagation, conjugate gradleto and a specialized limited precision training algorithm that we 
call Combined Search Algorithm (Xie and Jabri, 91). The input to the MI.P are 21 features 
extracted from the time domain. There are three outputs representing three main groupings: 
NSR, VT and VF. We do not have the space here to elaborate on the choice of the input fea- 
tures. Interested readers are referenced to (Chi and Jabri, 91; Leong and Jabri, 91). 
TABLE 1. Performance of Multi-layer Perceptron Based Classifiers 
Network Training Precision Average 
Algorithm Pfformance 
21-5-3 backprop. unlimited 96% 
21-5-3 conj.-grad unlimited 95.5% 
21-5-3 CSA 6 bits 94.8% 
The summary here indicates that a high performance single chamber based classification can 
be achieved for ventricular arrhythmias. It also indicates that limited precision training does 
not significantly degrade this performance. In the case of limited precision MLP, 6 bits plus a 
sign bit are used to represent network activities and weights. 
3.2 INDUCTION OF DECISION TREES 
The same training data used to tram the MLP was used to create a decision tree using the C4.5 
program developed by Ross Qumlan (a derivative of the 1I)3 algorithm). The resultant tree 
was then tested, and the performance was 95% correct classification. In order to achieve this 
high performance, the whole training data had to be used m the reduction process (windowing 
disabled). TNs has a negative side effect in that the trees generated tend to be large. 
The implementation of decision trees in VLSI is not a difficult procedure. The problem how- 
ever is that because of the binary decision process, the branching thresholds are difficult to be 
implemented m digital (for large trees) and even more difficult to be implemented m micro- 
power analog. The latter implementation technique would be possible if the induction process 
can take advantage of the hardware characteristics m a similar way that "m-loop" training of 
suNthreshold VLSI MLP achieves the same objective. 
4.0 DUAL CHAMBER BASED CLASSIFIERS 
Two architectures for dual chamber based classifiers have been investigated: Multi-Module- 
Neural Networks and a hybrid Decision Tree/MLP. The difference between the classifier archi- 
tectures is a function of which arrhythmia group is being targetted for classification. 
ANN Based Classification for Heart Defibrillators 641 
4.1 MULTI-MODULE NEURAL NETWORK 
The multi-module neural network architecture aims at improving the performance with respect 
to the classification of Supra-Ventricular Tachycardia (SVT). The architecture is shown in 
Figure 4 with the classifier being the right most block. 
FIGURE 4. Multi-Module Neural Network Classifier 
The idea behind this architecture is to divide the classification problem into that of discrimi- 
nating between NSR and SVT on one hand and VF and VT on the other. The details of the 
operation and the training of this structured classifier can be found in (Chi and Jabri, 91). 
In order to evaluate the performance of the MMNN classifier, a single large MlJP was also 
developed. The single large MLP makes use of the same input features as the MMNN and tar- 
gets the same classes. The performance comparison is shown in Table 2 which clearly shows 
that a higher performance is achieved using the MMNN. 
4.2 HYBRID DECISION TREE/MLP 
The hybrid decision tree/multi-layer perceptron "mimics" the classification process as per- 
formed by cardiologists. The architecture of the classifier is shown in Figure 5. The decision 
tree is used to produce a judgement on: 
1.The rate aspects of the ventricular and atrial channels, 
2.The relative timing between the atrial and ventricular beats. 
In parallel with the decision tree, a morphology based classifier is used to perform template 
matching. The morphology classifier is a simple MLP with input that are signal samples (sam- 
pled at half the speed of the normal sampling rate of the signal). 
The output of the timing and morphology classifiers are fed into an arbitrator which produces 
the class of the arrhythmia being observed. An "X out of Y" classifier is used to smooth out the 
642 Jabri, Pickard, Leong, Chi, Flower, and Xie 
TABLE 2. Performance of Multi-Module Neural Network 
Classifier and comparison with that of a single large MLP. 
Rhythm. MMNN MMNN Single 
Best % Worst % MLP % 
NSR 95.3 93.8 93.4 
ST 98.6 98.6 97.5 
SVT 96.4 93.3 95.4 
AT 95.9 93.2 71.2 
AF 86.7 85.4 77.5 
VT 99.4 99.4 100 
VTF 100 100 80.3 
VF 97 97 99.4 
Average 96.2 95.1 89.3 
SD 4.18 4.8 11.31 
classification output by the arbitrator and to produce an "averaged" final output class. Further 
details on the implementation and the operation of the hybrid classifier can be found in (Leong 
and Jabri, 91). 
This classifier achieves a high performance classification over several types of arrhythmia. 
Table 3 shows the performance on a multi-patient database and indicate a performance of over 
99% correct classification. 
fzom 
RVA -- 
pzob 
HIC . 
pzob 
--- Ncuaal Nttwozk gla$$ifi 
&zbis. zatt 
out of Y  FIHAL 
CLASS 
FIGURE 5. Architecture of the hybrid decision tree/neural network classifier. 
5.0 MICROELECTRONIC IMPLEMENTATIONS 
In all our classifier architecture investigations, micro-electromc implementation consider- 
atiom were a constant constraint. Many other architectures that can achieve competitive per- 
formance were not discussed in this paper because of their unsuitabihty for low power/small 
area implementation. The main challenge in a low power/small area VLSI implementation of 
classifiers similar to those discussed above, is how to implement in very low power a MLP 
architecture that can rehably learn and achieve a performance comparable to that of the func- 
ANN Based Classification for Heart Defibrillators 643 
tional similarions. Several design strategies can achieve the low power and small area objec- 
fives. 
TABLE 3. Performs. nee of the hybrid decision trdMLP chifier for d,ml 
chamber classification. 
SubClass Class NSR SVT VT VF 
NSR NSR 5247 4 2 0 
ST NSR 1535 24 2 1 
SVT SVT 0 1022 0 0 
AT SVT 0 52 0 0 
AF SVT 0 165 0 0 
VT VT 0 0 322 0 
VT 1:1 VT 2 0 555 0 
VF VF 0 0 2 196 
VTF VF 0 2 0 116 
Both digital and analog implementation techniques are being investigated and we report here 
on our analog implementation efforts only. Our analog implementations make use of the sub- 
threshold operation mode of MOS transistors in order to maintain a very low power dissipa- 
tion. 
5.1 MASTER PERTURBATOR CHIP 
The architecture of this chip is shown in Figure 6(a). Weights are implemented using a differ- 
ential capacitor scheme refreshed from digital RAM. Synapses are implememed as four quacl- 
rant Gilbert multipliers (Pickard et al, 92). The chip has been fabricated and is currently being 
tested. The building blocks have so far been successfully tested. Two networks are imple- 
mented a 7-5-3 (total of 50 synapses) and a small single layer network. The single layer net- 
work has been successfully trained to perform simple logic operatiom using the Weight 
Perturbation algorithm (Jabri and Flower, 91). 
5.2 THE BOURKE CHIP 
The BOURKE chip 0..eong and Jsbri, 92) makes use of Multiplying Digital to Analog Con- 
veners to implement the synapses. Weights are stored in digital registers. All neurom were 
implemented as external resistors for the sake of evaluation. Figure (fib) shows the schematics 
of a synapse. The BOURKE chip has a small network 3-3-1 and has been successfully tested 
(it was successfully trained to perform an XOR). A larger version of this chip with a 10-64 
network is being fabricated. 
6.0 Conclusions 
We have presented in this paW$ several architectures for single and dual chamber arrhythmia 
644 Jabri, Pickard, Leong, Chi, Flower, and )fie 
IIIIIIIIIII IIIIlIIIT 
Illllllllll 
IIIIIIIIIII 
11111111111 
IIIIII11111 
IIIIIIIIIII 
IIIIIIIIIII 
IIIIIIIIIII 
IIIIIIIIIII 
IIIIIIIIIII 
IIIIII11111 
LLILIlllIJL 
''1''1''1'' 
IIIIIIIll 
IIIIII11] 
IIIIIIIIT 
IIIIIIII1 
IIIIIII J' 
I  
IJllllllJ 
Illlllll,[ 
I11111111 
IIIIIIII! 
LI. I lJ IJJ] 
''l''l''l 
-I.I 
FIGURE 6. (a) Architecture of the Master Perturbator chip. (b) Schematics of the 
BOURKE chip synapse implementation. 
classifiers. In both cases a good classification performance was achieved. In particular, for the 
case of dual chamber classification, the complexity of the problem calls on more structured 
classifier architectures. Two microelectromc low power implementation were briefly pre- 
sented. Progress so far indicates that micro-power VLSI ANNs offer a technology that will 
enable the use of powerful classification strategies in implantable defibfillators. 
Acknowledgment 
Work presented in this paper was supported by the Australian Department of Industry Tech- 
nology & Commerce, Telectromcs Pacing Systems, and the Australian Research Council. 
References 
Z. Cl'fi and M. Jabri (1991), "Identification of Supraventricular and Ventricular Arrhythmias 
Using a Combination of Three Neural Networks". Proceedings of the Computers in Cardiol- 
ogy Conference, Vemce, Italy, September 1991. 
M. Jabri and B. Flower (1991), "Weight Perturbations: An optimal architecture and learning 
technique for analog VLSI feed-forward and recurrent multi-layer networks", Neural Compu- 
tation, Vol. 3, No. 4, MIT Press. 
P.H.W Leong and M. Jabri (1991), "Arrhythmia Classification Using Two Intracardiac 
Leads", Proceedings of the Computers in Cardiology Conference, Venice, Italy. 
P.H.W. Long and M. Jabri (1992), "An Analogue Low Power VLSI Neural Network", Pro-. 
ceedings of the Third Australian Conference on Neural Networks, pp. 147-150, Canberra, 
Australia. 
S. Pickard, M. Jabfi, P.H.W. Leong, B.G. Flower and P. Henderson (1992), "Low Power Ana- 
logue VLSI Implementation of A Feed-Forward Neural Network", Proceedings of the Third 
Australian Conference on Neural Networks, pp. 88-91, Canberra, Australia 
Y. Xie and M. Jabfi (1991), "Training Algorithms for Limited Precision Feed-forward Neural 
Networks", submitted to IEEE Transactions on Neural Networks and Neural Computation. 
