Time-frequency time-space LSTM for robust classification of physiological signals

Area of Science:

• Physiology
• Computer Science
• Biomedical Engineering

Background:

• Automated analysis of physiological time series is crucial for clinical applications.
• Long Short-Term Memory (LSTM) networks are effective for time-series data classification.

Purpose of the Study:

• To introduce time-frequency and time-space properties as robust tools for LSTM processing of long physiological time-series data.
• To improve the efficiency and accuracy of automated physiological data analysis.

Main Methods:

• Utilized time-frequency and time-space properties of time series.
• Applied Long Short-Term Memory (LSTM) deep recurrent neural network architecture.
• Classified sensor-induced physiological signals from two databases.

Main Results:

• Achieved very high classification accuracy for physiological time-series data.
• Demonstrated significant time savings in data learning processes.
• Showcased potential for cost-effectiveness and user comfort in clinical trials.

Conclusions:

• The proposed approach using time-frequency and time-space properties enhances LSTM performance for physiological data.
• This method offers a promising, efficient, and user-friendly solution for clinical data analysis.
• Potential to reduce the need for multiple wearable sensors in clinical trials.