Image reconstruction of Electrical Impedance Tomography (EIT) is a highly nonlinear ill-posed inverse problem, which is sensitive to the measurement noise and model errors. An improved Convolutional Neural Network (CNN) method is proposed for the EIT lung imaging. The proposed method is optimized based on the Visual Geometry Group (VGG) model, adding the batch normalization (BN) layer, ELU activation function, global average pooling (GAP) layer, and radial basis function (RBF) neural network. These optimizations help speed up network convergence, and improve reconstruction accuracy and robustness. Nearly 10 thousand EIT simulation models generated from chest CT images of 60 patients are used for the network training. The chest deformation, lung hyperdilation and atelectasis are randomly simulated during the model generation process. The proposed method after training is tested through a series of simulation data and experimental models. The reconstruction quality is quantitatively compared by calculating the root mean square error (RMSE) and image correlation coefficient (ICC). On average, the proposed method achieves 0.082 RMSE and 0.892 ICC through experimental results. The proposed method achieves high-resolution and robust shape reconstructions with multiphase conductivity for EIT lung imaging, especially in the presence of the measurement noise and interference. The proposed method is promising in providing quantitative images for potential clinical applications, such as human thorax imaging.

The poor spatial resolution of electrical impedance tomography (EIT) limits its use in medical devices because of its highly nonlinear and ill-posed natures. A novel hierarchical block sparse Bayesian learning (BSBL) method is designed for lung respiratory monitoring with EIT. It is its excellent modeling capability and noise robustness that allows BSBL to adaptively explore and exploit the internal conductivity distribution, e.g., block sparsity and intra-block correlation. Firstly, K-Nearest Neighbor (KNN) strategy is employed to automatically group each block based on the clustering property and incorporated to constrain the intra-block correlation matrix based on the spatial distance. Secondly, a three-layer hierarchical BSBL model using weighted Laplace (WL) prior is considered to enhance the recovery performance. Finally, an efficient bound optimization (BO) method is performed for Bayesian inference, avoiding the tedious parameter adjustment. This leads to the improvements in recovery performance, algorithm robustness and computational efficiency. Moreover, root mean square error (RMSE), image correlation coefficient (ICC), and relative size coverage ratio (RCR) are applied for quantitative comparisons of image quality. Numerical simulations and in vivo lung respiratory experiments showed that the proposed KNN-BSBL-WL method outperforms existing referenced methods in terms of reconstruction accuracy and computational time. On average, KNN-BSBL-WL obtains the RMSE = 0.07±0.02, ICC = 0.96±0.01, and RCR = 0.98±0.05, which are closest to the true values of the recorded snapshots obtained in the experiments. Meanwhile, the average time of KNN-BSBL-WL is 0.16 second, achieving an acceleration ratio close to ~3 compared with the referenced KNN-BSBL method. Therefore, the proposed method is an efficient and robust means of imaging reconstruction, which further improves the feasibility of EIT in respiratory clinical applications.

A new shape reconstruction framework that incorporates sparse Bayesian learning (SBL) and the B-spline level set (BLS) is proposed for difference electrical impedance tomography (EIT). To begin with, the conductivity distribution to be reconstructed is assumed to be piecewise constant under the BLS framework. Second, the boundaries of the inclusions are represented by the B-spline curves in a parametric and explicit way. Next, the optimal shapes are found by solving the unknown variables using the SBL method. Then, the proposed method takes full advantage of B-spline flexible representation and Bayesian learning capability, resulting in improvements in reconstruction performance, noise robustness, and computational efficiency. Finally, a series of numerical and experimental tests are performed to verify the feasibility of the proposed method. Furthermore, the robustness studies are also investigated considering the inhomogeneous background, different numbers of control points, and different initial guesses. The results show that the proposed method leads to reliable and robust shape reconstructions, as evaluated by several quantitative metrics. Specifically, the forced vital capacity (FVC) tests of ten healthy volunteers show that the mean error in the 1-s rate of the shape reconstructions is less than 5% compared with simultaneous spirometry measurements. Hence, the proposed method is promising in providing an effective assessment for lung imaging.

The growing prevalence of pulmonary diseases underscores the urgent need for early diagnosis to reduce mortality rates. Traditional pulmonary function tests often require active patient participation, making them difficult to administer in certain populations. This study presents a novel method for lung function evaluation during breathing using Spatiotemporal Residual Recurrent Neural Networks (STResRNN) based on Electrical Impedance Tomography (EIT). Our method integrates a residual network (ResNet) with a hybrid attention mechanism (HAM) and a bidirectional long short-term memory (BiLSTM) structure to effectively extract spatiotemporal features from EIT image sequences. This allows for accurate classification of lung ventilation function, distinguishing between normal, obstructive, and injured conditions without the need for forceful breathing maneuvers. In a clinical trial with 165 participants, the STResRNN achieved an overall accuracy of 97.4%. The F1 scores for Health, COPD, and Injury were 98.4%, 97.5%, and 96.3%, respectively. These results demonstrate the robustness of the model in identifying various lung conditions. Further parameter optimization and ablation studies validated the effectiveness of the network’s components. This research highlights the potential of EIT-based respiratory data for noninvasive and effortless lung function evaluation, paving the way for early pulmonary disease screening and expanding the clinical applications of EIT.

A multichannel bioimpedance spectroscopy (MC-BIS) method is proposed for regional identification of breast tumors. First, the sensor is partitioned to scan the breast across nine subregions, and an empty-field calibration algorithm is applied to normalize the impedance spectra. Next, numerical simulations are conducted to investigate the relationship between the electrical characteristics of unifocal tumors and their regional distribution. Subsequently, the performance of three classifiers—support vector machine (SVM), random forest (RF), and feedforward neural network (FNN)—is evaluated for bifocal tumor localization. The simulation results indicate significant differences in the impedance characteristics between tumor regions and normal tissue regions ( ${P} \lt 0.001$ ). When a tumor is present in a subregion, the corresponding imaginary part relaxation impedance ${Z} _{\text {imag-relax}}$ exceeds $2.004~\Omega $ . For bifocal breast tumor localization, the FNN classifier achieved the best performance, with a classification accuracy of 95.46% through fivefold cross-validation. To validate the simulation results, biological tissues with distinct electrical properties were selected to simulate tumor and normal tissue. The experimental accuracy reached 86.94%. The MC-BIS method enables rapid and accurate localization of tumor regions, providing a new technological approach for early screening and diagnosis of breast cancer.