Design and implementation of an annotated ambulatory ECG database for T-Wave alternans detection

Abstract Background T-wave alternans (TWA) reflects subtle beat-to-beat fluctuations in ventricular repolarization that indicate electrical instability and increased arrhythmic vulnerability. Despite strong scientific evidence, its clinical integration remains limited, largely due to the absence of a validated gold standard and the lack of annotated ambulatory ECG databases. Classical detection methods were designed for controlled heart-rate settings and translate poorly to real-world recordings, where autonomic, behavioral and circadian factors shape repolarization dynamics. Purpose To develop an evidence-based methodological framework and acquisition protocol for a high-quality annotated ambulatory ECG database specific to TWA, enabling rigorous qualitative and quantitative analysis and supporting the development of explainable learning-based algorithms. Methods A comprehensive review of ambulatory ECG literature informed the identification of clinical profiles with a high likelihood of exhibiting TWA. Evidence from acute coronary settings, post-infarction cohorts, ventricular electrical instability and chronic cardiac conditions guided the definition of inclusion and exclusion criteria. Technical specifications—lead configuration, recording duration, sampling, annotation strategy and metadata (including sex)—were established to ensure reliable alternans capture. All data handling procedures were defined within a national project focused on explainable machine learning. Results To date, 88 ambulatory ECG recordings meeting the evidence-based clinical criteria have been successfully acquired, representing the initial population of the database. These cases come from conditions with the highest documented TWA prevalence, ensuring meaningful alternans episodes for annotation. The protocol integrates validated thresholds for alternans magnitude and guarantees the capture of physiologically relevant variability inherent to ambulatory environments. This structure enables robust benchmarking of classical methods (Spectral Method, Modified Moving Average) and provides suitable ground for data-driven models. Conclusions We present the first comprehensive, evidence-driven protocol for constructing an annotated ambulatory ECG database dedicated to TWA. The initial cohort of 88 high-quality recordings demonstrates the feasibility and clinical relevance of the framework, which supports objective performance evaluation of detection techniques, development of explainable learning-based methods, and future sex-specific analyses in cardiac risk stratification.