PERFECG-Net: A Lightweight Periodicity-Focused Fetal ECG R-Peak Detection Network

Reliable detection of fetal R-peaks from abdominal electrocardiography (AECG) is essential for accurate and continuous fetal heart rate (FHR) monitoring, particularly in point-of-care and wearable healthcare applications. Conventional signal processing approaches often operate offline and degrade under low signal-to-noise conditions, while existing deep learning-based solutions typically rely on large, computationally intensive architectures, limiting their suitability for real-time deployment. To address these challenges, we propose PERFECG-Net (PERiodicity-Focused Fetal ECG Network), a lightweight end-to-end framework for fetal R-peak detection that explicitly embeds physiological cardiac periodicity into its architecture. The core component, a Physiologically-Informed Convolutional Bank (PICB), consists of multiple convolutional kernels aligned with physiologically plausible fetal heart rate ranges, enabling efficient extraction of rhythmic fetal ECG patterns. Despite its compact design with fewer than 150K parameters, PERFECG-Net achieves competitive detection performance across benchmark fetal ECG datasets while maintaining low computational complexity and real-time inference capability. To further evaluate edge deployment feasibility, the proposed model was deployed on an ESP32 microcontroller, demonstrating stable detection performance with real time inference. These results demonstrate that incorporating physiological priors into an ultra-lightweight architecture enables accurate, robust, and deployable fetal heart monitoring suitable for continuous clinical and home-based use. The source codes and trained models are available at: https://github.com/SSanjay0614/PERFECG-Net.