arXiv subjects

Axel Bauer

Publications and source records attributed to Axel Bauer.

4 recordsLinked to original sources

arXiv2025

BenchECG and xECG: a benchmark and baseline for ECG foundation models

Electrocardiograms (ECGs) are inexpensive, widely used, and well-suited to deep learning. Recently, interest has grown in developing foundation models for ECGs - models that generalise across diverse downstream tasks. However, consistent evaluation has been lacking: prior work often uses narrow task selections and inconsistent datasets, hindering fair comparison. Here, we introduce BenchECG, a standardised benchmark comprising a comprehensive suite of publicly available ECG datasets and versatile tasks. We also propose xECG, an xLSTM-based recurrent model trained with SimDINOv2 self-supervised learning, which achieves the best BenchECG score compared to publicly available state-of-the-art models. In particular, xECG is the only publicly available model to perform strongly on all datasets and tasks. By standardising evaluation, BenchECG enables rigorous comparison and aims to accelerate progress in ECG representation learning. xECG achieves superior performance over earlier approaches, defining a new baseline for future ECG foundation models.

cs.LG↗

arXiv2024

EchoDFKD: Data-Free Knowledge Distillation for Cardiac Ultrasound Segmentation using Synthetic Data

The application of machine learning to medical ultrasound videos of the heart, i.e., echocardiography, has recently gained traction with the availability of large public datasets. Traditional supervised tasks, such as ejection fraction regression, are now making way for approaches focusing more on the latent structure of data distributions, as well as generative methods. We propose a model trained exclusively by knowledge distillation, either on real or synthetical data, involving retrieving masks suggested by a teacher model. We achieve state-of-the-art (SOTA) values on the task of identifying end-diastolic and end-systolic frames. By training the model only on synthetic data, it reaches segmentation capabilities close to the performance when trained on real data with a significantly reduced number of weights. A comparison with the 5 main existing methods shows that our method outperforms the others in most cases. We also present a new evaluation method that does not require human annotation and instead relies on a large auxiliary model. We show that this method produces scores consistent with those obtained from human annotations. Relying on the integrated knowledge from a vast amount of records, this method overcomes certain inherent limitations of human annotator labeling. Code: https://github.com/GregoirePetit/EchoDFKD

cs.CV↗

arXiv2008

Bivariate phase-rectified signal averaging

Phase-Rectified Signal Averaging (PRSA) was shown to be a powerful tool for the study of quasi-periodic oscillations and nonlinear effects in non-stationary signals. Here we present a bivariate PRSA technique for the study of the inter-relationship between two simultaneous data recordings. Its performance is compared with traditional cross-correlation analysis, which, however, does not work well for non-stationary data and cannot distinguish the coupling directions in complex nonlinear situations. We show that bivariate PRSA allows the analysis of events in one signal at times where the other signal is in a certain phase or state; it is stable in the presence of noise and impassible to non-stationarities.

physics.data-an↗

arXiv2006

Demonstration of Circadian Rhythm in Heart Rate Turbulence using Novel Application of Correlator Functions

Background: It has been difficult to demonstrate circadian rhythm in the two parameters of heart rate turbulence, turbulence onset (TO) and turbulence slope (TS). Objective: To devise a new method for detecting circadian rhythm in noisy data, and apply it to selected Holter recordings from two post-myocardial infarction databases, Cardiac Arrhythmia Suppression Trial (CAST, n=684) and Innovative Stratification of Arrhythmic Risk (ISAR, n=327). Methods: For each patient, TS and TO were calculated for each hour with >4 VPCs. An autocorrelation function Corr(Delta t) = was then calculated, and averaged over all patients. Positive Corr(Delta t) indicates that TS at a given hour and Delta t hours later are similar. TO was treated likewise. Simulations and mathematical analysis showed that circadian rhythm required Corr(Delta t) to have a U-shape consisting of positive values near Delta t=0 and 23, and negative values for intermediate Delta t. Significant deviation of Corr(Delta t) from the correlator function of pure noise was evaluated as a chi-squared value. Results: Circadian patterns were not apparent in hourly averages of TS and TO plotted against clock time, which had large error bars. Their correlator functions, however, produced chi-squared values of ~10 in CAST (both p<0.0001) and ~3 in ISAR (both p<0.0001), indicating presence of circadian rhythmicity. Conclusion: Correlator functions may be a powerful tool for detecting presence of circadian rhythms in noisy data, even with recordings limited to 24 hours.

physics.med-ph↗