No more unnecessary heart surgeries
Photo: Královské Vinohrady University Hospital in Prague
In a healthy heart, ventricles are activated simultaneously. When the heart is damaged, though, for example after a major heart attack, one of the ventricles may get delayed, which reduces the performance of the heart. This is usually resolved by a biventricular cardiac pacemaker; however, this medical device is of no help to approximately one third of patients, and it may even harm them.
Regular ECG shows the electrical dyssynchrony the pacemaker takes care of, along with some other heart defects. In order to find out the patients whom the cardiac pacemaker really helps, doctors can now use high-frequency ECG technology developed by the team of Pavel Jurák from the Institute of Scientific Instruments at the Czech Academy of Sciences together with the Biomedical Engineering team at FNUSA-ICRC led by Pavel Leinveber and Karol Čurila with his colleagues from the Královské Vinohrady University Hospital in Prague. And Brno-based company Cardion helped make this cooperation possible.
An electrocardiogram (ECG) is a graph of electrical activity of the heart – a technology that has remained unchanged for a century now. Usually, it is analysed in frequencies up to 100 Hz, but this doesn’t clearly show when the individual heart segments are activated. Only since the 1980s did some scientists begin dealing with higher frequencies.
Researchers from #brnoregion have found a way to monitor high frequencies during ECG as well, and they even patented this method in the USA in 2018. “We’ve managed to show that a higher frequency in ECG shows the very moment when the given segment in the heart under the electrode gets electrically activated. This way, we can see potential delays of one or the other ventricle in milliseconds,” adds Leinveber.
This method should make it easy for doctors to evaluate the signals and determine precisely whether a cardiac pacemaker would help the patient. This diagnosis is easy to see thanks to depolarization maps that even less-experienced doctors can read, so the method is easy to use in everyday practice.
Currently, the team is working on a device that shows which position of pacemaker will benefit the patient most – directly in the operating room. They are also exploring ways to optimize the settings of already-implanted pacemakers to help the heart function even better.
Artificial intelligence can’t do it all
Among the members of the Medical Signals team at the Czech Academy of Sciences working on the development of the high-frequency ECG is Radovan Smíšek. He is focused on the analyses of signals sent by the heart in his dissertation paper. “In my paper, I present various pathologies detected using a low-frequency ECG. I monitor their behaviour in a high-frequency ECG, trying to gain new information,” explains Smíšek.
He also made use of this knowledge when he participated in and won a prestigious international competition organized by the International Society for Computerized Electrocardiology (ISCE).
Teams from all over the world worked there on a program capable of detecting a heart pathology called left bundle branch block, which is a disease in which one of the ventricles is delayed and can be mitigated using a pacemaker. Based on the results of the regular low-frequency ECG alone, it’s easy to confuse this disease with any other heart diseases where a cardiac pacemaker is of no help.
Photo: Královské Vinohrady University Hospital in Prague
“Recently, a scientific article has been published in the USA describing the criteria the ECG signal has to meet to enable us to more precisely predict which patients would benefit from the surgery,” explains Smíšek. The aim of the competition was to create a computer program which can find patients meeting the conditions set by the given article as well as automatically evaluate the suitability of the patients to undergo the surgery.
The competitors were provided data of three hundred patients for evaluation, and while most of them used AI, Smíšek created an algorithm without machine learning. “We had relatively little data available, which is a problem for AI. In the training set, there were only three hundred signals; in order to teach and create a reliable neuron network, though, a lot more would be needed,” he explained.
He decided to analyse the signals based on tiny deviations that were key to sorting the patients into individual groups, and his method turned out to be the most precise.