Iman Tavassoly


<H3}Fractal Analysis Of Heart Rate Variability</H3>

Objectives: The heart rate variability has been shown to have 
fractal behaviour during the time. Fractals are self similar shapes with 
fractional dimension and we can use them to find order in the biologic 
systems which seem to have non predictive and irregular behaviour. The 
exact relationship between fractal behaviour of heart rate variability and 
the condition of the cardiovascular system and its changes with age and 
diseases have not well studied.
We used an artificial neural network for screening the patient with 
arrhythmia as a pathologic state based on fractal dimension of heart rate 
variability.
<P>	
Methods: We used 115 digital rhythmograms from orto science software 
database library which included 75 healthy subjects and 40 patients with 
arrhythmia. We calculated the fractal dimension of heart rate variability 
using Nlyser software (version 3.5). We designed an artificial neural 
network by Neurosollutions software (version 4) and we used the age, gender 
and fractal dimension of heart rate variability of 90 rhythmograms (60 
healthy rhythmograms and 30 ones with arrhythmia) as input data and the 
condition of being healthy or having arrhythmia as target for training the 
artificial neural network. We used the data of 10 rhythmograms (5 healthy 
rhythmograms and 5 ones with arrhythmia) as validation set. Finally after 
training the network it was tested using the data of 15 rhythmograms.
<P>
Results: The designed artificial neural network could screen the healthy 
persons from arrhythmia patients based on the fractal dimension of heart 
rate variability with 0.2 mean squared errors. The network was 100% 
successful in finding healthy persons and it was 90.14% successful in 
finding arrhythmia patients.
<P>
Conclusions: Fractal dimension which is used as a factor showing fractal 
characteristics of heart rate variability can be used as a diagnostic tool 
to screen the patients with arrhythmia by artificial neural networks. The 
results will depend on the size of our training set for artificial neural 
network. If we develop the training set, the network will be more 
successful in screening the patients.
}