ECG Results Analysis by Artificial Intelligence | SBERMED AI
About company
Our solutions
Resource center

Contact us to find out more and start working with us

    * Required field

    SberMedAI LLC uses Cookies (files with information about past visits to the website) for service personalization and user convenience. To continue browsing the website, you need to enable cookies and accept our Cookies policy. You may disable cookies, but the website will become unavailable. Cookies policy


    ECG Results Analysis by Artificial Intelligence

    April 25 2022


    Reading time 10 minutes


    Electrocardiography (ECG) was introduced into clinical practice more than 100 years ago and has not lost its relevance at the moment. An affordable and inexpensive examination makes it possible to assess heart function in a wide range of diseases, without resorting to invasive procedures1.

    Cutting-edge technologies can change our conventional ideas of diagnostic capabilities of this method. Artificial intelligence- (AI-) based digital algorithms helps to extract more information from the cardiogram than is possible with human decoding. Due to computing methods, a clinician gets valuable data about the state of the body based on the electric potentials of heart.

    Which Purposes the ECG is Used for

    Artificial intelligence facilitates enables more accurate solutions to various diagnostic tasks:

    • Arrhythmias identification and classification. The AI analyzes ECG recording and classifies them identifying characteristic features of irregular rhythm2.

    • Help in case of acute pathologies. Outcomes of a number of diseases (atrial fibrillation, myocardial infarction) are often determined by the time spent to make diagnosis and start treatment. AI helps to reduce the interval between the appearance of first clinical symptoms and provision of medical assistance3.

    • Enhanced priority of non-invasive diagnosis. Identification of heart diseases is often connected with blood sampling: detection of cardiac markers in myocardial infarction4 or content of electrolytes in the serum in metabolic disorders5. AI raises the importance of electrocardiography as an alternative and safe method.

    • Low-quality recordings processing. The algorithms work efficiently with signals that include noises and other artifacts. This reduces clinician’s dependence on the quality of the equipment and possible mistakes made when getting the ECG. The physician does not have to repeat the procedure, which saves valuable time6.

    • Monitoring of heart function. AI can be integrated into cardiac services to facilitate accurate processing of recordings and timely identification of pathologies7.

    • Forecasting in cardiology. Neural networks make it possible to look forward and determine the probability of a heart disease. The analysis of electrophysiological signals helps to assess cardiovascular risks, for instance, in case of atherosclerosis8. The use of the technologies is especially relevant in case of recordings that seem normal in usual assessment. It is important to identify signs of ‘hidden’ pathologies, for example, atrial fibrillation9.

    Recognition of ECG Rhythms

    Atrial fibrillation (AF) is irregular and rapid heart rate. Possible consequences of AF include stroke or heart failure10.

    Scientists from China carry out work to improve monitoring of this rhythm disturbance. AI provides wide opportunities to recognize pathologies: the neural network managed to recognize the signs of AF on the electrocardiogram with accuracy, sensitivity and specificity exceeding 99%11.

    Training dataset included 16,557 12-lead ECG recordings provided by several hospitals. Deep learning-based method was tested on 3,312 cardiograms11.

    Arrhythmias Classification

    Scientists from the Stanford University and University of California in San Francisco provided a deep neural network for automated arrhythmias detection and classification. The AI is able to recognize 12 rhythm classes even for raw electrocardiography data. The training was carried out using 91,232 ECG in one abduction for 53,549 patients12.

    The algorithm can forecast changes in rhythm about once in a second. The test set included 328 unique electrocardiograms described by cardiologists. Sensitivity and specificity of the algorithms exceeded 90% almost in all cases12.

    Researchers evaluated how accurately the developed AI detects rhythm disturbances from the ECG. The assessment used the F1 indicator, a harmonic means of the accuracy and completeness of the classification. The average F1 indicator for the neural network was 0.83712.

    The development of Taiwanese scientists classifies an additional diagnostic pattern. Along with 12 classes of rhythm, the AI identifies signs of ST elevation myocardial infarction. The accuracy of life-threatening condition recognition on ECG was assessed by a special statistical indicator, the area under the AUC curve. During a testing, the artificial intelligence showed the following results3:

    • in case of a single marking of acute myocardial infarction, when pathology in the cardiogram appear separately – 0.9962;

    • in case of multiple marking, when the signs of myocardial infarction were combined with other rhythm disturbances – 0.9375.

    The authors point out that the method will help prioritize care for patients with chest pain3

    Practice of the AI Use in ECG Decoding

    Врач слушает сердечный ритм пациентки

    Left ventricular hypertrophy (LHV) means an increase in the mass of left ventricular (LV) of heart. This increases the size of cardiac cells, cardiomyocytes. Progressive LHV can lead to LV dysfunction and heart failure13. Timely identification and treatment of the pathology improves forecast for patients from the risk group14.

    Scientists from the Republic of Korea have worked out algorithms to identify LHV from the ECG data. The testing results are published in the scientific journal Nature Communications. The efficiency of the deep learning-based AI algorithm was compared with the capabilities of the machine interpretation of electrocardiograms and the expert opinions of a cardiologist14:

    • Electrocardiograph algorithm: assessed the probability of heart pathology by four levels: no LHV, consider LHV, probable LHV and LHV.

    • Cardiologist: made diagnosis based on clinical data and ECG results: the Sokolow-Lyon index, Romhilt-Estes system and the Cornell voltage criteria. Additionally, demographic information was considered: gender, age, height and weight.

    • AI-based algorithm: studied both demographic characteristics and ECG characteristics, such as heart rate, waves and intervals. Raw electrocardiac signals were also included in the analysis.

    The study was conducted at two hospitals for more than 21,000 patients. The quality characteristic of the AUC classification for neural network was 0.880 and 0.868 for checks in two hospitals. With the same specificity, the Sokolow-Lyon index sensitivity of the AI exceeded that of a machine interpretation by 177.7% and 143.8% respectively14.

    American scientists from the Wake Forest School of Medicine and Loyola University Chicago have developed a model to predict heart failure (HF) based on the ECG data15.

    For this purpose, data from the ARIC, a prospective epidemiologic study of atherosclerosis, was used. The AI was tested with the help of selected data on the initial visits of 14,613 patients from 1987 to 1989. The obtained forecast was verified by confirmed HF episodes with 5.5% of patients within 10 years15.

    Reliability of the AI predictions was compared with ARIC and Framingham Heart Study (FHS) heart failure risk calculators. ARIC and FHS models predict the probability of a disease based on many clinical factors, including15:

    • gender;

    • age;

    • body mass index;

    • systolic blood pressure and heart rate;

    • presence of diabetes mellitus;

    • presence of cardiovascular diseases: left ventricular hypertrophy or valvular heart disease.

    The developed AI algorithm initially analyzed 12-lead ECG only15.

    AUC indicator for ARIC and FHS HF risk calculators was 0.802 and 0.780. The neural network, which used the data from electrocardiogram only, showed a comparable result – 0.756. The scientists managed to improve it to 0.818 by including clinical factors in the analysis15.

    The authors note the value of electrocardiography in forecasting HF as an inexpensive and affordable method. Based on its data, the neural network can predict new episodes of a disease in feature with a high degree of probability15.

    Hyperkalemia, an elevated level of potassium in the blood, is often asymptomatic and can lead to life-threatening rhythm disturbances. A joint project of the Mayo Clinic and AliveCor company, an advanced cardiology technology company, is devoted to the issue of efficient hyperkalemia monitoring16.

    The neural network was learnt using more than 1.6 million ECGs of about 450,000 patients with chronic kidney disease (CKD), who have elevated level of potassium. Related cardiotoxic effects can be seen on a cardiogram. A deep learning-based model classified such changes with high accuracy: the AUC indicator was from 0.853 to 0.90116.

    The researches pointed out a number of advantages of the developed AI in clinical setting16:

    • the algorithm can be applied in hyperkalemia screening diagnosis;

    • the method reliably determines the level of potassium without blood sampling;

    • prompt monitoring of blood electrolyte composition helps to correct the therapy in time and prevent fatal complications.

    ECG analysis with the use of AI, due to accelerated processing of entries, is in demand in the intensive care setting, when it is necessary to quickly help patients in grave condition. The short ECG by SberMedAI company is aimed at optimizing the work of the ambulance or urgent care center team and identifying cardiovascular pathologies.

    Service algorithm:

    1. The physician upload the data from electrocardiography.

    2. The AI carries out analysis and identifies possible rhythm disturbances and other signs of a heart disease.

    3. The results can be seen by health specialists, ambulance or urgent care center teams who have access to the service.

    4. A connected specialist takes the call and conducts an online consultation via video communication. An electronic digital signature is used when sending medical documents.

    As new data arrives, the neural network automatically raises the accuracy of decisions. You can get quality medical care regardless of your location: all you need is access to the online service through your work or home computer.

    New machine learning-based approaches improve the classical method of diagnosing cardiovascular pathologies. The AI does not only automate and accelerate the processing of cardiograms, but also eliminates possible discrepancy when interpreting results.


    1. Noel G Boyle and Jitendra K Vohra, The Enduring Role of the Electrocardiogram as a Diagnostic Tool in Cardiology. J Am Coll Cardiol. 4 April 2017; 69 (13): 1704-1706. doi: 10.1016/j.jacc.2017.01.036.

    2. Venkat D Nagarajan, Su-Lin Lee, Jan-Lukas Robertus, Christoph A Nienaber, Natalia A Trayanova and Sabine Ernst, Artificial Intelligence in the Diagnosis and Management of Arrhythmias. Eur Heart J. 7 October 2021; 42 (38): 3904-3916. doi: 10.1093/eurheartj/ehab544.

    3. Kuan-Cheng Chang, Po-Hsin Hsieh, Mei-Yao Wu, et al. Usefulness of Multi-Labelling Artificial Intelligence in Detecting Rhythm Disorders and Acute ST-Elevation Myocardial Infarction on 12-Lead Electrocardiogram, European Heart Journal — Digital Health. June 2021; 2(2):299–310. doi: 10.1093/ehjdh/ztab029.

    4. Yifan Zhaoa, Jing Xionga, Yang Hou, et al. Early Detection of ST-Segment Elevated Myocardial Infarction by Artificial Intelligence with 12-Lead Electrocardiogram Int J Cardiol. 2020 15 October;317:223-230. doi: 10.1016/j.ijcard.2020.04.089.

    5. Joon-Myoung Kwon MD, MS, Min-Seung Jung BS, Kyung-Hee Kim MD, et al. Artificial Intelligence for Detecting Electrolyte Imbalance Using Electrocardiography. Ann Noninvasive Electrocardiol. May 2021; 26(3):e12839. doi: 10.1111/anec.12839.

    6. Lishen Qiu, Wenqiang Cai, Miao Zhang, et al. Two-Stage ECG Signal Denoising Based on Deep Convolutional Network. Physiol Meas. 9 December 2021; 42 (11). doi: 10.1088/1361-6579/ac34ea.

    7. Zachi I Attia, MS, Peter A Noseworthy, MD, Prof. Francisco Lopez-Jimenez, MD, et al.
      An Artificial Intelligence-Enabled ECG Algorithm for the Identification of Patients with Arial Fibrillation During Sinus Rhythm: a Retrospective Analysis of Outcome Prediction. Lancet. 7 September 2019; 394 (10201): 861-867. doi: 10.1016/S0140-6736(19)31721-0.

    8. Bharath Ambale-Venkatesh, Xiaoying Yang, Colin O Wu, et al. Cardiovascular Event Prediction by Machine Learning: The Multi-Ethnic Study of Atherosclerosis. Circ Res. 13 October 2017;121(9):1092-1101. doi: 10.1161/CIRCRESAHA.117.311312.

    9. Georgios Christopoulos, Jonathan Graff-Radford, Camden L. Lopez, et al. Artificial Intelligence–Electrocardiography to Predict Incident Atrial Fibrillation: A Population-Based Study. Circ Arrhythm Electrophysiol. 2020 December;13(12):e009355. doi: 10.1161/CIRCEP.120.009355.

    10. Jayson R. Baman, MD, and Rod S. Passman, MD, MSCE, Atrial Fibrillation JAMA. 1 June 2021; 325(21):2218. doi: 10.1001/jama.2020.23700.

    11. Wenjuan Cai, Yundai Chen, Jun Guo, et al. Accurate Detection of Atrial Fibrillation from 12-Lead ECG Using Deep Neural Network. Comput Biol Med. 2020 January;116:103378. doi: 10.1016/j.compbiomed.2019.103378.

    12. Awni Y. Hannun, Pranav Rajpurkar, Masoumeh Haghpanahi, et al. Cardiologist-Level Arrhythmia Detection and Classification in Ambulatory Electrocardiograms Using a Deep Neural Network. Nat Med. January 2019; 25(1):65-69. doi: 10.1038/s41591-018-0268-3.

    13. Davide Lazzeroni, Ornella Rimoldi and Paolo G Camici, From Left Ventricular Hypertrophy to Dysfunction and Failure Circ J. 2016;80(3):555-64. doi: 10.1253/circj.CJ-16-0062.

    14. Joon-Myoung Kwon, Ki-Hyun Jeon, Hyue Mee Kim, Min Jeong Kim, Sung Min Lim, Kyung-Hee Kim, Pil Sang Song, Jinsik Park, Rak Kyeong Choi and Byung-Hee Oh, Comparing the Performance of Artificial Intelligence and Conventional Diagnosis Criteria for Detecting Left Ventricular Hypertrophy Using Electrocardiography. EP Europace. 2020 March, 22(3): 412–419,

    15. Oguz Akbilgic, Liam Butler, Ibrahim Karabayir, et al. ECG-AI: Electrocardiographic Artificial Intelligence Model for Prediction of Heart Failure. Eur Heart J Digit Health. 9 October 2021; 2 (4): 626-634. doi: 10.1093/ehjdh/ztab080.

    16. Conner D. Galloway, MSc, Alexander V. Valys, BS, Jacqueline B. Shreibati, MD, et al Development and Validation of a Deep-Learning Model to Screen for Hyperkalemia From the Electrocardiogram. JAMA Cardiol. 1 January 2019; 4(5):428-436. doi: 10.1001/jamacardio.2019.0640.

    Похожие статьи