Improving automatic T wave annotation in the ECG with robustness tests facilitated by a novel ECG data augmentation technique

• Applicant: Professor Dr. Konstantinos Rizas
• Subject Area: Medical Informatics and Medical Bioinformatics
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 549333517

The electrocardiogram (ECG) reflects the electrical activity of the heart. It is measured from electrodes placed on the body as voltage changes over time. The most important component of the ECG is the repolarization, which accounts for the most vulnerable phase of the cell cycle and is defined by the so called "T-wave". Annotations of the T wave fiducial points in the ECG (onset and offset) are needed for the diagnosis of diseases and risk assessment of patients. Our group has developed a novel risk stratification tool called periodic repolarization dynamics (PRD), which identifies patients suffering from various cardiovascular disorders showing a high-risk for sudden cardiac death. PRD reflects the effect of low-frequency (<= 0.1Hz) phasic sympathetic activation on myocardial cells. Increased PRD has been shown to be a strong and independent predictor of mortality, cardiovascular mortality, sudden-cardiac death, and arrhythmic events in patients with ischemic and non-ischemic cardiomyopathy. Moreover, there is evidence that PRD can correctly identify patients that benefit from prophylactic implantation of a defibrillator (ICD). A main limitation of the method is that it requires the manual annotation of the boundaries (onset and offset) of at least 2,000 T-waves. This manual T-wave annotation is not only resource demanding, but is also a source of bias, as it is highly dependent on the experience of the clinician performing this analysis. Current methods for automatic T-wave annotation are not robust and require manual control and correction of errors. This project's main objective is to significantly improve automated T wave annotation using artificial intelligence (AI). However, in order to achieve this goal a huge amount of ECG recordings obtained from thousands of patients at different conditions (heart rate, respiratory rate) and various levels of noise are required. This huge amount of data is currently not available. In this project we aim to firstly apply an innovative augmentation method, suitable for artificially enlarging ECG datasets and subsequently use machine-learning techniques in order to create a free, open-source and robust method for automatic and error-free T-wave annotation technology. This technology can be subsequently applied in everyday clinical practice to improve ECG interpretation, diagnosis of life-threatening diseases and improve ECG-based risk-stratification.

• DFG Programme: Research Grants
• International Connection: Austria
• Co-Investigator: Privatdozent Dr.-Ing. Axel Loewe
• Cooperation Partner: Professor Dr. Axel Bauer