Researchers have reported that an electronic stethoscope and a personal computer have been used to distinguish innocent heart murmurs from those that may indicate a serious problem.
Studies in the US estimate that a heart murmur can be heard in 77 percent to 95 percent of children at some time during childhood. A heart murmur is an extra heart sound heard with each heartbeat. Less than one percent of heart murmurs are a sign of problems such as those caused by defective heart valves or a malformed heart.
Listening to the heart is the primary tool for distinguishing heart murmurs. But the human ear cannot appreciate many of the subtleties of heart sounds and the interpretation of the sounds is prone to error, said principal researcher Curt G. DeGroff, M.D., a practising paediatric cardiologist and assistant professor in the department of paediatrics at the University of Colorado Health Sciences Centre.
An artificial neural network (ANN) – a computer program that can recognise complex patterns – was developed by co-researcher Roop L. Mahajan, Ph.D., a professor of mechanical engineering at the University of Colorado, Boulder, and other colleagues.
ANNs can learn complex interactions and identify subtle relationships that may not be apparent to humans. Studies of ANNs in cardiology have been mainly concerned with the evaluation of electrocardiogram (ECG) signals. Their use on heart sounds has been examined in a few studies with limited results and applicability.
Here, researchers used heart sound recordings from 69 patients – 37 with abnormal heart murmurs and 32 with innocent murmurs – to train the network.
Using a unique mathematical model, they converted the sound recordings into the energy-per-unit of frequency interval to take advantage of the computer’s pattern-recognition capabilities.
‘The mathematical signature for each child and the patterns for innocent and abnormal murmurs are different,’ Mahajan explained.
The researchers fed samples of the heart recordings back to the ANN model and adjusted the frequency range and sensitivity of the signals to improve the computer’s ability to differentiate between the abnormal and innocent murmurs.
They also re-entered the data to mathematically mimic consultations with multiple experts. By doing so, they reached 100 percent sensitivity (the ability to identify an abnormal heart murmur) and 100 percent specificity (the ability to identify an innocent heart murmur).
Training the artificial neural network is akin to the way parents teach children life lessons, Mahajan noted. ‘As children, we are given some examples by our parents, and from those our brain neurons learn to make distinctions.’
Although the results are encouraging, DeGroff and Mahajan stress that this is a preliminary study and more development is needed before ANNs can be used for mass screening.
‘Such a device is not seen as replacing the need for a clinician’s assessment in a child found to have a heart murmur, but it may help the doctor render a better opinion,’ DeGroff concluded.