GRU Neural Network Improved Bioimpedance Based Stroke Volume Estimation during Ergometry Stress Test

Area of Science:

• Biomedical Engineering
• Cardiology
• Artificial Intelligence

Background:

• Cardiovascular diseases (CVDs) are a leading cause of non-communicable diseases, necessitating early diagnosis for effective treatment.
• Exercise stress tests are common for CVD identification, but hemodynamic monitoring accuracy can be limited.
• Thoracic electrical bioimpedance offers beat-to-beat stroke volume calculation but is sensitive to motion artifacts.

Purpose of the Study:

• To develop an improved method for measuring bioimpedance signals during exercise stress tests.
• To reduce signal artifacts and accurately calculate hemodynamic parameters.
• To enhance the diagnostic capabilities for cardiovascular diseases.

Main Methods:

• A novel approach utilizing a gated recurrent unit (GRU) neural network combined with ECG signals.
• Investigating hemodynamic status and parameters during exercise.
• Comparing the GRU network's performance against ensemble averaging, adaptive filters, and shallow neural networks.

Main Results:

• The GRU neural network demonstrated superior performance in handling single artifact events compared to shallow neural networks.
• Achieved a mean error of -0.0244 and mean square error of 0.0181 for normalized stroke volume.
• The GRU network proved more effective than other algorithms in processing time-correlated data during exercise stress tests.

Conclusions:

• The proposed GRU neural network method significantly improves the accuracy of bioimpedance measurements during exercise stress tests.
• This technique offers a promising advancement for non-invasive hemodynamic monitoring and cardiovascular disease diagnostics.
• The GRU network's ability to reduce artifacts enhances the reliability of stroke volume calculations.