EUVIMEDEUVIMED entdecken
Crossref

PCSA-UNet: A U-Net With Physiologically Constrained Sparse Attention for Fetal R-Peak Detection

Durgesh Ameta, Yashovardhan Thopte, Sanjay Somasundaram, Megha Ameta, Amit Shukla

IEEE Transactions on Instrumentation and Measurement · 2026 · 75 · 2515714-2515714

DOI: 10.1109/tim.2026.3714607 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2026. Verfasst von Durgesh Ameta und 4 weiteren Autor:innen. Die Arbeit nennt 56 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Fetal heart rate (FHR) monitoring is essential for assessing fetal well-being during pregnancy and labor; however, reliable detection of fetal R-peaks from abdominal electrocardiography (AECG) remains challenging due to strong maternal ECG (MECG) interference, low signal-to-noise ratio (SNR), and substantial intersubject variability. To address these challenges, we propose a U-Net-based framework with physiologically constrained sparse attention (PCSA-UNet) for end-to-end, sample-wise fetal R-peak detection from multilead AECG signals. The architecture integrates a 1-D encoder–decoder backbone for precise preservation of local ECG morphology with a novel sparse attention mechanism, physiologically constrained sparse attention (PCSA) that explicitly incorporates prior knowledge of FHR periodicity. By restricting attention computation to physiologically plausible temporal offsets, PCSA effectively captures long-range rhythmic dependencies while avoiding the quadratic complexity and noninformative interactions associated with vanilla self-attention. MECG interference is further addressed through a physiologically motivated loss function that penalizes false fetal detections at maternal-only peaks, reducing the need for explicit MECG separation for the task of fetal R-peak detection. The proposed method is extensively evaluated on two public benchmark datasets, the abdominal and direct fetal ECG database (ADFECGDB) and the bio-signal data database (BDDB) (labor and pregnancy subsets). Compared with reported literature results, PCSA-UNet achieves competitive performance, attaining $F1$ -scores of 99.91 % on ADFECGDB, 98.79 % on BDDB-P, and 98.35 % on BDDB-L. Furthermore, inference latency measurements suggest the potential for real-time deployment on an edge GPU platform such as the NVIDIA Jetson Orin Nano. Overall, PCSA-UNet demonstrates that embedding physiological priors into sparse attention mechanisms within a U-Net framework enables competitive fetal R-peak detection with strong accuracy and stability directly from AECG. By utilizing maternal annotations exclusively during training, the framework eliminates the need for explicit MECG separation during inference, representing a potential approach for noninvasive fetal monitoring. The source codes and trained models are available at this GitHub repository.

Zitieren

APA

Ameta, D., Thopte, Y., Somasundaram, S., Ameta, M., & Shukla, A. (2026). PCSA-UNet: A U-Net With Physiologically Constrained Sparse Attention for Fetal R-Peak Detection. IEEE Transactions on Instrumentation and Measurement, 75, 2515714-2515714. https://doi.org/10.1109/tim.2026.3714607

Vancouver

Ameta D, Thopte Y, Somasundaram S, Ameta M, Shukla A. PCSA-UNet: A U-Net With Physiologically Constrained Sparse Attention for Fetal R-Peak Detection. IEEE Transactions on Instrumentation and Measurement. 2026;75:2515714-2515714. doi:10.1109/tim.2026.3714607

BibTeX
@article{Ameta2026,
  author = {Ameta, Durgesh and Thopte, Yashovardhan and Somasundaram, Sanjay and Ameta, Megha and Shukla, Amit},
  title = {PCSA-UNet: A U-Net With Physiologically Constrained Sparse Attention for Fetal R-Peak Detection},
  journal = {IEEE Transactions on Instrumentation and Measurement},
  year = {2026},
  volume = {75},
  pages = {2515714--2515714},
  doi = {10.1109/tim.2026.3714607}
}
Kostenlos starten →