Related Experiment Video
Updated: May 6, 2026
06:56
Human Fetal Blood Flow Quantification with Magnetic Resonance Imaging and Motion Compensation
Published on: January 7, 2021
2.4K
Automatic classification of fetal heart rate based on a multi-scale LSTM network
Lin Rao1,2, Jia Lu1,2, Hai-Rong Wu3
1International Peace Maternity and Child Health Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China.
Frontiers in Physiology
|June 27, 2024
Summary
A novel multi-scale LSTM artificial intelligence model accurately classifies fetal heart rate patterns during labor, improving upon existing methods. This AI tool supports obstetricians in diagnosing abnormal fetal heart rates, enhancing patient care.
Area of Science:
- Artificial Intelligence in Healthcare
- Medical Signal Processing
- Obstetrics and Gynecology
Background:
- Fetal heart rate monitoring during labor is crucial for identifying distress.
- Interpretation of fetal heart rate patterns faces challenges due to guideline discrepancies and human error.
- Artificial intelligence (AI) offers a potential solution to aid in the accurate diagnosis of abnormal fetal heart rates.
Purpose of the Study:
- To develop and evaluate a multi-scale long short-term memory (LSTM) neural network for automated fetal heart rate classification.
- To compare the performance of multi-scale LSTM models against traditional LSTM models.
- To investigate the impact of different sampling rates on classification accuracy.
Main Methods:
- Preprocessing techniques were used to handle missing signals and artifacts.
- Data augmentation was employed to address class imbalance issues.
- A multi-scale LSTM neural network was trained on time-series fetal heart rate data for classification.
Main Results:
- Multi-scale LSTM models demonstrated superior performance compared to regular LSTM models.
- The model with a 10 Hz sampling rate achieved the highest accuracy (85.73%), specificity (85.32%), and precision (85.53%) on the CTU-UHB dataset.
- The model achieved an Area Under the Receiver Operating Curve (AUC) of 0.918, indicating high diagnostic credibility.
Conclusions:
- The developed multi-scale LSTM model shows significant potential for accurate automated fetal heart rate classification.
- Incorporating varied sampling rates improved overall performance metrics, including accuracy, specificity, and AUC.
- Further research should consider clinical characteristics and gestational age for enhanced model generalizability.
