Year: 2025
Type: preprint
Abstract: The detection of meaningful waves in an electrocardiogram signal is a critical step in any computerassisted decision system. Neural network algorithms have achieved state-of-the-art results in this domain. However, prior studies have reported unrealistic state sequence predictions, including unlikely state transitions and excessively short or long sojourn times within states. These issues often arise from a suboptimal training process or training data that do not represent the testing data. This work aims to overcome such limitations by proposing an integer program problem with novel state transition constraints and a new objective function with a penalty term that mitigates unrealistic state sequences. Methods: The Fourier synchrosqueezed transform is applied to each electrocardiogram signal, from where a long-short term memory network computes conditional state output probability distribution per sample; these are used as input to an integer program problem to infer the most likely state sequence. Results: Our approach increases the system's ability to reconstruct the true state sequence of events, with the F1 score increasing on average from 60. 2% to 69. 5%, compared to the standard algorithm that uses an argmax function for the reconstruction of a state sequence. Conclusions: Incorporating domain-specific constraints and a tailored objective function, all inspired by established principles of cardiac physiology, improves the segmentation of an electrocardiogram signal. Significance: This study advances not only the reliability but also the transparency of computer-assisted biomedical decision systems.
Authors: Jorge Oliveira, Rui Camacho, Marcelo Nogueira
Language: English
Topic: ECG Monitoring and Analysis
Subfield: Cardiology and Cardiovascular Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W4407019641