Prediction of future EIT respiratory cycle images using an LSTM approach

This repository describes the experimental data acquisition of EIT data. For this purpose, a balloon in a phantom tank is ventilated by a medical ventilator. Thirty-two equidistantly distributed electrodes on the tank capture the EIT data using a Sciospec EIT device. A data-driven reconstruction approach, composed of a Variational Autoencoder (VAE) and an LSTM mapper, is used to reconstruct the conductivity distribution in the electrode plane.

Contents

• Data acquisition
• VAE training
• LSTM Mapper training

Reconstruction network architecture

First, the VAE is used to learn a latent representation of the conductivity distribution in the cross-sectional electrode plane. Second, a mapper with an LSTM is trained on a sequence of EIT measurements to predict the next EIT image.

Fig. 1: Overview of the temporal data arrangement for predicting a subsequent construction based on previous measurements. The blue boxes represent the EIT measurements, and the corresponding cross-sectional conductivity distribution in the electrode plane is depicted in the bottom boxes.

Results

Fig. 2: Randomly chosen EIT image predictions from the test data set with the corresponding ground truth

Please see or cite:

@INPROCEEDINGS{11253129,
 author={Thönes, Jacob P. and Spors, Sascha},
 booktitle={2025 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)},
 title={Towards Predicting Future Impedance Distributions from Temporal Sequences of EIT Measurements Using a Recurrent Network},
 year={2025},
 volume={},
 number={},
 pages={1-5},
 keywords={Training;Electrical impedance tomography;Impedance measurement;Training data;Ventilation;Time measurement;Impedance;Reliability;Image reconstruction;Long short term memory},
 doi={10.1109/EMBC58623.2025.11253129}
}