Densely Connected Convolutional Neural Network-Based Invalid Data Compensation for Brain Electrical Impedance Tomography

Electrical impedance tomography (EIT) is a potential technique for the brain imaging. With this technique, pathology related conductivity variation inside the intracerebral domain can be visualized. However, electrode disconnection is a common phenomenon during the long-term monitoring with EIT. This will cause data loss and lead to image reconstruction failure. To address this problem, a novel method based on densely connected convolutional neural network is proposed to compensate the invalid data in the case of electrode disconnection occurring in the brain EIT. Based on this method, advanced features of voltage sequences can be extracted. Also, it alleviates the problems of information disappearance in the higher layer and gradient vanish in the lower layer. To evaluate the performance of the proposed compensation method, extensive simulation work is conducted. In addition, phantom experiments are carried out. Both of qualitative evaluation and quantitative analysis demonstrate that the proposed method is able to well compensate the invalid data under various conditions when there is electrode disconnection. Compared with the compensation methods based on fully connected neural network and shallow convolutional neural network, much better performance can be obtained when the proposed method is adopted. This work would offer an alternative for the improvement of reconstruction quality under electrode disconnection.