Early animal studies laid essential groundwork for the use of EIT in pulmonary perfusion assessment, providing preliminary validation of its feasibility. Research conducted in porcine and ovine models demonstrated that EIT can reliably detect abnormalities in regional perfusion distribution, with good concordance between EIT-derived measurements and reference imaging modalities such as SPECT and PET (Frerichs et al., 2002; Stowe et al., 2019; Borges et al., 1985; Bluth et al., 2019). Subsequent investigations have progressively advanced toward optimizing imaging methodologies and contrast agent selection. These animal experiments not only facilitated the evolution of EIT from two-dimensional to three-dimensional imaging (Larrabee et al., 2023) but also explored safer contrast alternatives such as sodium bicarbonate (NaHCO₃) (Gao et al., 2025), establishing an important foundation for future clinical translation of this technology.

In the intensive care unit (ICU), the real-time dynamic monitoring capability of EIT is of particular importance. By integrating ECG gating technology, EIT can provide perfusion waveforms in multiple lung regions and relative V/Q matching status, offering an important basis for clinical assessment of V/Q mismatch. In a prospective study, researchers evaluated the reliability of ECG-gated EIT in measuring stroke volume in critically ill patients. The results showed good agreement with measurements obtained by transpulmonary thermodilution (TPTD), supporting the feasibility of using regional impedance changes synchronized with cardiac activity to estimate stroke volume variations (Braun et al., 2020). Marco Leali and colleagues further proposed a noninvasive ECG-gated EIT calibration method that integrates ECG signals with EIT data to generate absolute V/Q images (V/Q-abs). The calibrated V/Q-abs images closely approximated those obtained from invasive monitoring, demonstrating the potential of this approach to achieve completely noninvasive quantification of V/Q matching for dynamic bedside perfusion monitoring (Leali et al., 2024).

Acquisition of pulmonary perfusion images with EIT typically requires the integration of several technical methods:

Respiratory pause (RP): Images are obtained during brief apnea to minimize respiratory interference and improve perfusion signal contrast (He H. W. et al., 2021).

Contrast enhancement: Injection of 15–20 mL of cold saline as a conductive contrast agent temporarily alters impedance distribution, enhancing the contrast between perfused regions and surrounding tissues (He H. W. et al., 2021).

Electrocardiogram-gated EIT (ECG-gated EIT): By synchronously recording ECG signals, impedance changes across multiple cardiac cycles can be averaged to improve image stability (McArdle et al., 1988).

Frequency-domain filtering (FDF): Since ventilation and perfusion signals differ in frequency, filters are applied to separate these components in the frequency domain from the raw data (Zadehkoochak et al., 1992).

Principal component analysis (PCA): Variance-based analysis of EIT time-series data separates respiration-related impedance changes (e.g., tidal volume variations) from cardiac-related changes (e.g., pulmonary vascular filling/emptying) into distinct principal components (Jang et al., 2020).

Currently, most EIT devices employ 16- or 32-electrode systems, which limit image resolution. Nevertheless, their advantages lie in bedside applicability and the ability to perform serial measurements at multiple time points, making them particularly suitable for dynamically tracking changes in pulmonary perfusion.

In recent years, with the continuous breakthroughs in deep learning algorithms, the integration of AI with EIT has been constantly developing, and its application scope in medical imaging and the treatment of pulmonary diseases has gradually expanded. Researchers have proposed an enhanced EIT approach known as single-network reconstruction, in which a neural network directly reconstructs images from raw EIT data. By learning the complex relationship between surface electrical measurements and internal conductivity distributions, the network is able to generate high-resolution images that are critical for diagnostic purposes (Zheng et al., 2022).

AI also shows considerable potential in pulmonary feature extraction. In the future, AI-based algorithms may process EIT data to extract clinically relevant parameters such as global inhomogeneity (GI), center of ventilation (CoV), regional ventilation delay (RVD), tidal impedance variation (TIV), and end-expiratory lung impedance (EELI). These parameters are essential for assessing lung function and may contribute to optimizing patient management and improving clinical outcomes (Cappellini et al., 2024). Furthermore, these AI-enabled quantitative features may be integrated with the characteristic pattern of focal perfusion defects seen in PE, thereby enabling automated detection and risk alerting on top of AI-assisted interpretation. Such integration is expected to enhance the capability of EIT for early bedside identification of PE and for dynamic risk assessment throughout the clinical course.

In this review, we define studies with a sample size <20, whether prospective or retrospective, as small clinical studies to differentiate them from single-case reports. For clarity and transparency, the sample size of each study has been explicitly indicated in the table (Table 2).

Overall, current evidence from case reports and small-scale studies demonstrates that EIT can identify regional perfusion defects and V/Q mismatch in patients with PE, even when ventilation remains preserved. In critically ill or postoperative patients, EIT has provided valuable bedside information when conventional imaging was delayed or contraindicated. Moreover, EIT also shows potential for dynamic monitoring of therapeutic effects. For instance, after adjustment of anticoagulation, thrombolysis, or ventilatory parameters, the perfusion signal gradually improves, which is consistent with the imaging of embolus absorption.

Based on the above case reports and small clinical studies, a characteristic finding of acute PE on EIT is a focal, segmental pattern of ventilation–perfusion mismatch. Specifically, in EIT perfusion imaging, PE is often manifested as markedly weakened or absent perfusion signals in one lung or specific regions, while ventilation images remain relatively preserved, resulting in a characteristic V/Q mismatch. It should be emphasized that the V/Q mismatch identified by EIT reflects a spatial dissociation between ventilation and perfusion rather than an alteration in physiological V/Q ratios. Similar to conventional perfusion imaging, EIT shows significantly decreased impedance variation in perfusion-deficient areas, with well-defined regional boundaries, which can be valuable for bedside dynamic monitoring of disease progression or treatment response. Multiple animal studies have confirmed that hypertonic saline–enhanced EIT perfusion imaging correlates well with CTPA or SPECT perfusion imaging (Frerichs et al., 2002; Borges et al., 1985; Hentze et al., 2018). Preliminary clinical studies and case reports have also demonstrated good agreement between EIT perfusion imaging and CTPA findings (He et al., 2020b; Safaee et al., 2020; He et al., 2020c; Yuan et al., 2021).

In addition, several reports have shown that perfusion defects observed on EIT images gradually improved following thrombolytic or anticoagulant therapy, suggesting its potential utility in treatment response evaluation (Yuan et al., 2021; Wang et al., 2021; Prins et al., 2023; Ding et al., 2024). The figure below illustrates the typical manifestations of EIT perfusion imaging in different types of PE (Figure 1) (Wang et al., 2021).

By reviewing relevant literature, we have initially collated the main diagnostic parameters and calculation methods of EIT currently used for the assessment of PE (Table 3).

However, these parameters have not been widely adopted in clinical practice, primarily due to significant technical and methodological heterogeneity, which has made it difficult to establish uniform diagnostic thresholds. Firstly, the processing workflow of EIT involves multiple critical steps, such as image reconstruction, filtering parameter settings, and ECG gating strategies. However, there is currently a lack of internationally standardized operating protocols for these steps, leading to variations in the fundamental characteristics of images generated by different devices or research teams. Secondly, differences also exist in the definition of the region of ROI and its reference baseline. Whether using geometric division based on the entire image or physiological division based on functional lung contours, the specific implementation involves a certain degree of subjectivity or algorithm dependency. This reduces comparability across different studies and affects the reproducibility of research findings. Furthermore, diversity exists in the core algorithms and parameter definitions themselves, such as the separation of ventilation and perfusion signals, calculation of V/Q correlation, and models for estimating dead space fraction. The use of different calculation methods in various experimental studies makes data from different sources difficult to compare or integrate directly.

In order to make the textual description more concrete, we have included representative figures (Figures 2, 3), derived from unpublished data courtesy of Dr. Huiting Li (Department of Pulmonary Circulation, Shanghai Pulmonary Hospital, Tongji University; Infivision ET1000) and used with permission, hoping that readers can gain a clear understanding of EIT imaging and segmentation.

This review further compares the numerical values of EIT-derived parameters reported across different clinical studies (Table 4). Notably, substantial variability exists in the absolute values of the same diagnostic parameter among studies. For example, the reported dead space% in patients with PE ranges broadly from approximately 30%–50%, and similar variability is observed in left–right perfusion ratios and V/Q mismatch proportions. Such inter-study discrepancies primarily stem from methodological heterogeneity—including differences in contrast agent concentration, breath-holding strategies, ROI segmentation, image reconstruction algorithms, and analytical thresholds. These sources of variation explain why a universal quantitative diagnostic cutoff for EIT has not yet been established. Nevertheless, these parameters retain significant value for trend monitoring and dynamic bedside assessment.

Although EIT cannot yet replace standard imaging modalities such as CTPA, it offers several unique advantages: it does not require patient transfer, making it particularly suitable for critically ill patients in the ICU; it provides strong dynamic monitoring capability, enabling multi-timepoint evaluation of perfusion changes and real-time observation during thrombolysis; it allows integration with ventilation images to facilitate identification of V/Q mismatch; and it is radiation-free and contrast-free, making it applicable to patients with contraindications.

Nevertheless, EIT still faces several limitations. According to GRADE evidence-based criteria, the current level of clinical evidence supporting the use of EIT for PE diagnosis remains low. Most available studies consist of case reports and small sample analyses, with only one prospective study among the 14 studies summarized. In addition, substantial methodological heterogeneity exists across studies, including variability in hypertonic saline concentrations and differences in image-processing algorithms, making it difficult to draw robust conclusions. Furthermore, the limited spatial resolution of EIT constrains its ability to detect subsegmental or more distal pulmonary emboli, and its diagnostic sensitivity remains markedly inferior to that of CTPA (He et al., 2020b). Most studies also failed to report the time interval between EIT and CTPA examinations, limiting the ability to evaluate the true timeliness of EIT as an early screening tool.

At present, EIT has been preliminarily applied in the clinical management of PE. Overall, its role in diagnosis appears relatively more established, while its use in therapeutic monitoring remains limited, for example, comparing perfusion before and after thrombolysis. Furthermore, its diagnostic sensitivity is still inferior to that of CTPA (She et al., 2023). An ideal clinical strategy would be to integrate these modalities based on patient conditions: using EIT for early bedside imaging and continuous monitoring, confirming the diagnosis with CTPA in stable situations, and employing echocardiography for cardiac function assessment, thereby enabling a more precise and safer individualized treatment plan.

Based on the above case studies, we further refined this integrated pathway. For ICU patients with high-risk factors or clinical instability, EIT should first be utilized for bedside monitoring. When unexplained desaturation or hemodynamic fluctuations occur, an immediate EIT examination is warranted. If the images demonstrate a clear regional V/Q mismatch, this can serve as an important basis for initiating empirical anticoagulation, and such patients may be classified as “highly suspicious.” For patients with positive EIT findings and relatively stable vital signs, confirmatory imaging and risk stratification should be performed using conventional modalities. Simultaneously, bedside echocardiography should be used to assess right ventricular function and identify acute cor pulmonale, thereby providing hemodynamic evidence to guide subsequent therapeutic decisions.

Once treatment has begun, the utility of EIT can be extended further. EIT is capable of continuously tracking dynamic changes in perfusion defects, allowing clinicians to objectively assess the actual response to anticoagulation or thrombolytic therapy and to adjust treatment strategies in a timely manner. Of course, the clinical effectiveness and workflow feasibility of this integrated approach require validation through future prospective studies.

Current mainstream two-dimensional EIT systems are susceptible to artifacts caused by physiological activity occurring outside the imaging plane, such as diaphragmatic motion or changes in body position. These artifacts may compromise image quality and reduce the accuracy of regional localization. The development of three-dimensional EIT offers a promising solution to this challenge. By deploying dual electrode belts for simultaneous data acquisition and applying three-dimensional reconstruction algorithms, full-volume lung imaging can be achieved. This approach not only helps reduce motion-induced artifacts but also improves the anatomical accuracy of lesion localization (Grychtol et al., 2019; Gao et al., 2024). Efforts to optimize signal quality are also ongoing. Hyun et al. developed a novel automated signal quality index (SQI) method using discriminant models and manifold learning to detect abnormal CVS induced by motion artifacts, representing the first attempt to enhance EIT cardiopulmonary monitoring by assessing CVS signal quality (Min Hyun et al., 2023).

EIT is inherently limited by its low spatial resolution and modest conductivity contrast, which can make the resulting images difficult to interpret. Achieving high-quality three-dimensional imaging places additional demands on hardware, requiring more sophisticated electrode arrays and substantially more complex reconstruction algorithms. To overcome these bottlenecks, researchers have increasingly integrated advanced computational methods into EIT reconstruction.

Dong Liu et al. applied convolutional neural network (CNN)-induced regularization with deep image prior (DIP) to EIT reconstruction, offering a novel approach to regularization in EIT inverse problems (Liu et al., 2023). Similarly, Junwu Wang et al. proposed using image priors to guide neural network initialization, thereby improving EIT image reconstruction quality (Wang et al., 2025). Together, these studies highlight the impact of integrating modern computational techniques and neural network architectures on advancing EIT technology, providing more accurate, efficient, and versatile imaging solutions for medical and scientific applications.

Collectively, these studies illustrate the impact of integrating advanced computational methods and neural network architectures on advancing EIT technology, providing more accurate, efficient, and versatile imaging solutions for medical and scientific applications.

Given the current technological limitations of EIT, its clinical application should adhere to prudent ethical standards. EIT should be regarded as part of an integrated diagnostic pathway rather than an independent diagnostic tool. For patients in whom EIT screening suggests a high suspicion of PE or presents complex findings, confirmation with higher-precision imaging modalities such as CTPA is essential. Moreover, prior to clinical use, patients and their families should be adequately informed of the benefits, risks, and limitations of the technique.

With continuous improvements in digital image quality and data processing algorithms, the technical bottlenecks of EIT in pulmonary perfusion imaging are gradually being overcome. Advances in EIT technologies, including the integration of AI and novel sensors, are opening a new era of EIT research.