Automated ECG Interpretation

For more than 150 years, ECG recording technology has been developing and improving. Doctors now have access to multi-channel recorders, long-term (tens of days) ECG recording devices, and wearable recorders integrated into smartphones, watches, bracelets, etc. Data processing algorithms are also improving, including those using artificial intelligence, which is developing very rapidly in medicine. This mainly concerns areas of working with images, recorded signals, big data processing, and managing complex processes in large populations.

How can AI help in Interpreting a Patient’s Condition?

There are already scientific studies showing that AI can not only handle routine processes better than a doctor but also make disease predictions that specialists cannot. Since, reflects the electromechanical processes occurring in the heart, obviously, any deviation from the norm or even peculiarities in heart function will be reflected in the ECG. Currently, analyzing peculiarities and deviations from the norm is entirely the doctor’s responsibility: they can detect heart attacks, myocardial hypertrophy, cardiac conduction system blockages, etc. on the recording. However, ECG contains a huge amount of other information that cannot be seen by eye, even by the most experienced doctor. Now imagine that a patient’s blood pressure has just started to rise; or if a person is tired, overworked, or has exceeded their physical training regime – can ECG show such changes? It turns out that yes, it can. Machine learning-based algorithms have been developed for ECG signal analysis with frequency spectral analysis – wavelet analysis. The system evaluates not only the shape of ECG waves, their duration and amplitude, but also the frequency of signal oscillation at each recording point. Such analysis has been found to detect minor changes in cardiac electrical activity that cannot be seen with the naked eye. Based on the developed system, it became possible to detect decreased heart muscle function even before the onset of severe irreversible cardiac changes. With high accuracy, the program can detect decreased myocardial diastolic function, which indicates conditions such as cardiac overload due to increased pressure, initial stage of ischemia threatening to develop into infarction, onset of heart failure, and physical or emotional overload.

How does it work?

The algorithm requires only one ECG channel. Hence, ECG measurements have proven reliable and require minimal training for beginners. The electrical signal’s waveform propagates from one end of the heart to the other. This signal can be measured using conductive electrodes placed at appropriate locations on the skin. The electrical current flowing through the heart can be detected on the skin non-invasively since the human body contains fluids with ions that provide electrical conductivity. The heart’s electrical signal can have small amplitude values in the microvolt range. The waveform of the ECG graphical representation is created by calculating the difference in electrical potentials between two electrodes. Thus, electrode placement is an important factor in ECG measurement, where different locations can produce different forms of the same signal. For example, a 3-lead system produces 3 different signal waveforms, identified as channels, representing the same electrical activity of the heart. Each lead produces a different signal waveform, containing variable electrophysiological information in each channel. Leads II and III provide information related to the inferior surface of the heart. Meanwhile, Lead I provides lateral information. Single-lead ECG systems use two electrodes to detect one ECG signal. Electrode placement remains important in determining the type of information that will be obtained from the system. For example, a single-lead sensor, Lead-I, positioned horizontally between the patient’s right and left shoulders, will show a different ECG signal waveform compared to the ECG signal from Lead II. Lead II shows a higher R peak than Lead I, making it more suitable for determining the R-R interval and calculating heart rate. But both Lead I and Lead II offer a vertical P wave, which can be used to detect certain heart conditions. On the other hand, signal amplitude can help evaluate other heart functions as it relates to myocardial mass and various heart tissue properties. A single-lead system uses fewer hardware components than a 12-lead system. This makes it more comfortable for the patient to move, as a single-lead system does not require many bulky attachments to the body. This opens up enormous prospects for remote diagnosis and monitoring of cardiovascular diseases using mobile phones and wearable gadgets without visiting medical facilities. At any time, any of us can determine if everything is okay with our heart function. Furthermore, integrating the developed algorithm into medical systems will allow doctors to immediately react to deterioration in a patient’s condition, conduct necessary additional examinations, and provide preventive and therapeutic recommendations. In Norav Medical we believe that integration of AI automated interpretation in mobile devices will provide excellent patient experiences, enriching, thus, the process of monitoring and prevention of potentially hazardous conditions. We urge you to become our partner in this breathtaking journey to new horizons of ECG technology.  

References

1. Georgiou, K., Larentzakis, A. V., Khamis, N. N., Alsuhaibani, G. I., Alaska, Y. A., & Giallafos, E. J. (2018). Can wearable devices accurately measure heart rate variability? A systematic review. Folia medica, 60(1), 7-20.
2. Huang, P. S., Tseng, Y. H., Tsai, C. F., Chen, J. J., Yang, S. C., Chiu, F. C., … & Tsai, C. T. (2022). An artificial intelligence-enabled ECG algorithm for the prediction and localization of angiography-proven coronary artery disease. Biomedicines, 10(2), 394.
3. Marinucci, D., Sbrollini, A., Marcantoni, I., Morettini, M., Swenne, C. A., & Burattini, L. (2020). Artificial neural network for atrial fibrillation identification in portable devices. Sensors, 20(12), 3570.