Towards Predicting Future Impedance Distributions from Temporal Sequences of EIT Measurements Using a Recurrent...

Area of Science:

• Medical Imaging
• Data Science
• Electrical Engineering

Background:

• Electrical Impedance Tomography (EIT) is increasingly using data-driven techniques.
• Machine learning (ML) effectively addresses nonlinear, inverse, and ill-posed problems in EIT image reconstruction.
• Reliable imaging is critical for EIT applications, such as monitoring intensive care patients.

Purpose of the Study:

• To explore the potential of recurrent EIT measurements combined with Long Short-Term Memory (LSTM) and Variational Autoencoder (VAE).
• To predict the next time-instance impedance distribution for improved EIT image reconstruction.
• To enable early detection of deviations from established physiological cycles in patients.

Main Methods:

• Application of recurrent sequences of EIT measurements.
• Integration of Long Short-Term Memory (LSTM) cells.
• Utilisation of a Variational Autoencoder (VAE) framework.

Main Results:

• Demonstrated potential for predicting future impedance distributions in EIT.
• Showcased a novel combination of LSTM and VAE for time-series EIT data.
• Indicated possibilities for enhanced image accuracy and early anomaly detection.

Conclusions:

• The proposed LSTM-VAE model shows promise for advancing EIT applications.
• This data-driven approach can improve the reliability and predictive capabilities of EIT systems.
• Future work could focus on clinical validation for intensive care monitoring.