An Effective Voltage Data Restoration Approach for High-Quality Imaging in Electrical Impedance Tomography

During the long-term monitoring with electrical impedance tomography (EIT), complete electrode detachment causes abnormal measurements, degrading imaging quality or causing reconstruction failure. Aiming at the problem, an effective voltage data restoration approach is developed. Firstly, the detached electrode is identified by calculating average measurement variation. Then, based on the prior information of electrode detachment, a hybrid network model integrating temporal convolutional network (TCN) with bidirectional long short-term memory network (BiLSTM) is constructed. Numerical simulations are conducted to assess the effectiveness of the proposed approach. Compared with restoration strategies based on LSTM, BiLSTM, TCN and TCN-LSTM, the proposed TCN-BiLSTM method yields the highest average R-squared value and the lowest MAE value across different case. In terms of image reconstruction, the inclusion cannot be identified from reconstructed images when abnormal measurements are used. Nevertheless, reconstruction quality is largely improved when voltage data restored by the proposed method is applied for image reconstruction. It is also demonstrated that the proposed method is resilient to noise interference. Moreover, phantom experimental validation is carried out. This study offers a novel solution to enhance the reliability of long-term EIT monitoring.