LearnTWAGold

According to the World Health Organization, cardiovascular diseases (CVDs) are the leading cause of mortality worldwide compared with any other cause, of which half of them are from sudden cardiac death (SCD). Patients classified into a high-risk group to suffer from SCD are eligible to therapy, whose ultimate treatment consists of implantable cardioverter defibrillator, a small device embodied in the chest under the skin. This therapy is prescribed after Electrophysiological Study, which allows stratifying patients in risk groups. So, the development of noninvasive markers for risk stratification of sudden cardiac death is a subject of great attention. The main interest is driven by the design of indexes able to manage with ambulatory recordings, resulting in a low-cost effective technology, minimally cumbersome to patients, and accessible to a wide population segment. Although T- wave alternans (TWA) has been proven to be a good computational indicator, the bioelectrical indexes developed in the last decades to detect this phenomenon have been scarcely transferred to clinical routine. The reason is that no gold standard for the methodological validation of the algorithms has been developed yet. Since TWA are manifested as small fluctuations of the ST-T complex in the surface electrocardiogram (ECG), they are invisible to the naked eye and very difficult to identify by straight visualization of the ECG. Furthermore, the lack of ground truth has also prevented the application of Machine Learning methods to the detection of TWA. Among the two most accepted computerized reported methods, the Spectral Method (SM) and the Modified Moving Average (MMA), none of them have been shown to be suitable for ambulatory recordings. The SM requires clinical supervision to stabilize the ECG in a middle-term, and the MMA has been shown to be less robust as it may report false positive TWA due to the inherent T-wave variability. The current project aims to report contributions in the open issues referred above, namely, the development of Machine and Deep Learning algorithms to characterize the TWA phenomenon on ambulatory recordings, and the contribution with a set of annotated data for methodological validation. The algorithms proposed in this project will help in the identification of cardiac patients at risk, and can be implemented in existing monitoring devices at a low cost. So the development of the diagnostic support systems of this project will have a significant impact in the National Health System through the improvement of decision making, both in patient stratification and surgery application, and also through a cost reduction due to the application of this diagnostic tool to ambulatory outpatients./3

Team

Publications and Achievements

Journal Papers

Pascual-Sánchez, L., Goya-Esteban, R., Cruz-Roldán, F., Hernández-Madrid, A., & Blanco-Velasco, M. (2024). Machine learning based detection of T–wave alternans in real ambulatory conditions. Computer Methods and Programs in Biomedicine, 249, 108157.

Conference Papers

Pascual-Sánchez, L., Goya-Esteban, R., Cruz-Roldán, F., Hernández-Madrid, A., & Blanco-Velasco, M. (2023, October). Analysis of the window size effect for T-Wave Alternans detection through Machine Learning methods. In 2023 Computing in Cardiology (CinC) (Vol. 50, pp. 1-4). IEEE.

Sosa García, A.S., Goya-Esteban, R., Pascual-Sánchez, L., Blanco-Velasco, M. (2024, November).T-wave alternans analysis: a machine learning and deep learning approach in Python. In 2024 Congreso Anual de la Sociedad Española de Ingeniería Biomédica (CASEIB).

PhD Thesis

Desarrollo de métodos basados en aprendizaje para la detección de alternancias en la onda T sobre registros ambulatorios en señales electrocardiográficas. Pascual-Sánchez, L. Advisors: Blanco-Velasco, M., Goya-Esteban, R. Ongoing.