Related Experiment Video
Updated: Aug 29, 2025
03:57
Anesthesia-free Heartbeat Measurements in Freely Moving Zebrafish
Published on: April 18, 2025
579
A U - Net Deep Learning Model for Infant Heart Rate Estimation from Ballistography
Summary
Ballistography (BSG) offers a low-cost infant heart rate monitoring alternative to ECG. A deep learning U-Net model achieved 80% predictive performance for heartbeat detection from BSG signals, though an enhanced layer did not improve heart rate estimation.
Area of Science:
- Biomedical Engineering
- Signal Processing
- Machine Learning
Background:
- Ballistography (BSG) is a non-intrusive, cost-effective method for infant heart rate (HR) monitoring, serving as an alternative to electrocardiography (ECG).
- Detecting heartbeats in BSG signals is challenging due to patient variability and noise susceptibility.
Purpose of the Study:
- To estimate infant heart rate (HR) from bed-based pressure mat BSG signals using a deep learning approach.
- To evaluate the effectiveness of a modified U-Net deep neural network for heartbeat detection and HR estimation.
Main Methods:
- A U-Net deep neural network was trained using supervised learning, with ground truth heartbeats derived from simultaneously recorded ECG signals.
- The U-Net model was modified with an IC-layer to enhance generalization capabilities for heartbeat detection.
Main Results:
- The U-Net model without the IC-layer achieved a predictive performance of 80% for heartbeat detection.
- The inclusion of the IC-layer improved the model's generalization for detecting heartbeats but did not enhance the accuracy of HR estimation.
Conclusions:
- Deep learning, specifically the U-Net architecture, shows promise for estimating heart rate from BSG signals in infants.
- While model enhancements can improve generalization for heartbeat detection, further research is needed to optimize HR estimation accuracy from BSG data.
