Robust Compensation of Contact Impedance Change With Fully Connected Neural Network in Brain EIT

Electrical impedance tomography (EIT) is a promising candidate for brain imaging. Contact impedance change between the scalp and electrodes has a large impact on boundary measurement and causes deterioration or failure of image reconstruction. To correct the measured voltage when contact impedance varies in brain EIT, a robust compensation method based on a fully connected neural network (FCNN) is proposed. With the proposed method, the relationship between measured invalid voltage and predicted compensation voltage is established. To validate the effectiveness of the proposed method, image reconstruction of simulated hemorrhage in a three-layer head model is conducted. Images reconstructed with the compensated voltage are compared with that using the directly measured data. The performance of the proposed method is tested when the contact impedance of a single electrode and multiple electrodes varies. In addition, the impact of the increase of inclusion size and interruption of noise which may occur simultaneously with contact impedance change is respectively studied. Conductivity variation of the scalp layer, skull layer, and brain tissue layer is also considered. Qualitative evaluation and quantitative comparison show that images reconstructed with voltage compensated by the proposed method are much better than those with directly measured data. In other words, the invalid voltage measured under contact impedance change is well corrected with the proposed compensation method. There is little difference between the results under different testing which indicates the robustness of the proposed method. To conclude, it is possible for the proposed method to improve reconstruction quality when there is contact impedance change in brain EIT.