[A fetal electrocardiogram signal extraction method based on long short term memory network optimized by genetic...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Fetal electrocardiogram (FECG) signal extraction is crucial for perinatal monitoring.
• Existing methods face challenges in accurately isolating FECG from maternal signals.
• Improving FECG extraction accuracy enhances diagnostic capabilities for fetal health.

Purpose of the Study:

• To propose and validate a novel GA-LSTM method for accurate FECG signal extraction.
• To optimize LSTM network parameters using a Genetic Algorithm (GA) for improved performance.
• To enhance the clarity and reliability of FECG signals for perinatal monitoring.

Main Methods:

• Utilized a Genetic Algorithm (GA) to optimize Long Short-Term Memory (LSTM) network parameters (hidden neurons, learning rate, training epochs).
• Developed a GA-LSTM model to estimate the nonlinear transformation of maternal ECG to abdominal signals.
• Estimated and subtracted maternal ECG from abdominal signals to isolate pure FECG signals.

Main Results:

• The GA-LSTM method demonstrated superior performance in extracting clearer FECG signals compared to NLMS, GA-SVM, and standard LSTM.
• Significant improvements were observed in accuracy, sensitivity, precision, and overall probability.
• Clinical data from two databases validated the effectiveness of the proposed method.

Conclusions:

• The GA-LSTM method provides a robust approach for extracting relatively pure fetal electrocardiogram signals.
• This technique holds significant application value for improving perinatal fetal health monitoring.
• Optimized deep learning models offer a promising direction for advanced biomedical signal processing.