Removal of Vibration Interference Artifacts in Electrical Impedance Tomography Monitoring Using Residual Learning Strategy

Electrical impedance tomography (EIT) can be used for real-time bedside monitoring of pathological changes in human brain tissue and dynamic imaging. However, EIT measurement signals are easily disturbed by noise, tremor in stroke patients, and other factors in clinical practice, which result in significant artifacts in the reconstructed images. These artifacts degrade the image quality and affect the detection of focal targets. To address this problem, we propose a multiscale 1-D residual convolutional network (MS-1DResCNN) to remove chaotic noise and interference from the conductivity distribution data and retain the target feature data, thereby improving the robustness of the reconstruction algorithm to noise and interference. The structural similarity (SSIM) and normalized mean square error (NMSE) were used to evaluate the performance of the proposed method in suppressing image reconstruction artifacts under noise-free conditions, different noise simulation levels, and varying vibration interference frequencies. The results of physical experiments showed that, under 10-Hz vibration interference, the reconstructed images using the MS-1DResCNN method improved the SSIM by 40.4%, 22.0%, 17.4%, and 11.7% and reduced the NMSE by 74.3%, 72.4%, 65.2%, and 29.5%, compared to the damped least squares (DLSs), artificial neural network (ANN), error-constraint network (Ec-Net), and U-Net methods, respectively. This method effectively improves the anti-interference ability of conditional algorithms, significantly reduces artifacts in EIT images, and makes the perturbation target clearer and more precise, thereby providing stable and reliable algorithmic support for the clinical application of EIT.