Electrical Impedance Tomography: Advances and Clinical Insights

Electrical impedance tomography, or EIT, creates images of the inside of the body by passing tiny electrical currents between electrodes placed on the skin and measuring the resulting voltage patterns. Unlike CT scans or MRI, it uses no radiation, requires no trips to the imaging suite, and can run continuously at the bedside. That combination has made it especially valuable in intensive care, where clinicians need moment-to-moment feedback on how a patient’s lungs are ventilating. But advances in hardware, reconstruction algorithms, and machine learning are pushing EIT well beyond its original niche, into stroke detection, cardiac monitoring, and even gastrointestinal assessment.

How the Technology Works in Practice

A typical clinical EIT setup involves a belt of 16 or 32 small electrodes wrapped around the patient’s chest. The system injects an alternating current through one pair of electrodes while the remaining electrodes measure voltages. It then cycles through different injection pairs, collecting a full set of measurements in a fraction of a second. Because different tissues resist electrical current differently, the voltage patterns change depending on what lies beneath each electrode. Air-filled lung tissue has high impedance; fluid-filled or collapsed tissue has low impedance. A computer reconstructs these measurements into a cross-sectional image, updated many times per second.

The mathematical challenge behind those images is significant. Turning boundary voltage data into a map of internal conductivity is what researchers call an “ill-posed inverse problem,” meaning small measurement errors can produce large image errors. Traditional approaches relied on regularization techniques to stabilize solutions, but deep learning methods have recently gained traction. Neural networks trained on simulated or clinical data can now reconstruct EIT images with improved stability, and researchers have proposed architectures that estimate whether each point in the imaging domain belongs to background tissue or an anomaly, working directly in continuous space rather than on a fixed pixel grid.¹ These advances are increasingly seen as the foundation for future integrated EIT diagnostic systems.²

Guiding Ventilator Settings in ARDS

The most mature clinical application of EIT is helping clinicians set the right level of positive end-expiratory pressure, or PEEP, for patients with acute respiratory distress syndrome. PEEP keeps airways open between breaths, but setting it too low leaves parts of the lung collapsed, and setting it too high overinflates healthy regions. Both extremes can worsen lung injury. Traditionally, clinicians pick a PEEP level based on oxygenation tables or pressure-volume curves, which tell you about the lungs as a whole but nothing about what is happening region by region.

EIT changes that equation. By showing ventilation distribution in real time, it lets clinicians see exactly where collapse and overdistension are occurring. The most widely studied approach involves performing a stepwise PEEP trial while EIT tracks regional compliance: compliance losses at higher PEEP indicate overdistension, while losses at lower PEEP indicate collapse. The optimal PEEP sits at the crossover point where both are minimized.³

A systematic review and meta-analysis pooling data from hundreds of ARDS patients found that EIT-guided PEEP improved lung compliance compared to conventional methods, with a mean difference of about 4 mL/cmH₂O. Driving pressure and mechanical power also dropped. In a subset of roughly 260 patients across three studies, EIT-guided PEEP was associated with a lower mortality rate, with a relative risk of about 0.64.⁴ A separate meta-analysis reached a similar conclusion, noting that real-time bedside assessment of regional ventilation may limit ventilator-induced lung injury and enhance survival.⁵

A randomized controlled trial focused on moderate-to-severe ARDS found that the EIT group showed significantly better static respiratory compliance during the first two days after the intervention, particularly among patients with the most severe disease. The authors attributed this to EIT’s ability to find a PEEP level that balances recruitment against overdistension, maintaining more uniform ventilation across lung regions.⁶ A pilot feasibility study showed that adjusting both PEEP and tidal volume using an EIT-based protocol improved oxygenation and reduced alveolar cycling without promoting global overdistension.⁷

Real-Time Pneumothorax Detection

One of EIT’s most striking capabilities is catching pneumothorax, the accumulation of air between the lung and the chest wall, almost as it happens. In an experimental study, pneumothoraces as small as 20 mL were detected with 100% sensitivity and 95% specificity. Their location was correctly identified in every case, with a total delay of only three breathing cycles.⁸ The system could also distinguish a pneumothorax from the overdistension that occurs during recruitment maneuvers, a distinction that matters because the two look very different from a treatment standpoint even though both involve excess air.

EIT’s value for pneumothorax detection extends to other contexts as well. The consensus statement from the TRanslational EIT developmeNt stuDy group highlighted the ability to identify adverse events during mechanical ventilation, including pneumothorax and derecruitment after procedures like suctioning, enabling earlier therapeutic intervention.⁹ A review of EIT during mechanical ventilation confirmed its usefulness for pneumothorax detection, quantifying pulmonary edema, and comparing ventilation distribution across different ventilation modes.¹⁰

Lung Perfusion and Ventilation-Perfusion Matching

Ventilation is only half the picture. Oxygen exchange depends on blood flow reaching the same lung regions that are receiving air. EIT can estimate regional lung perfusion by tracking the impedance changes caused by a bolus of hypertonic saline injected intravenously. This creates a brief, measurable contrast effect as the saline passes through the pulmonary vasculature, allowing clinicians to build a perfusion map alongside the ventilation map.

The standard technique required a brief breath hold during the saline injection. A recent randomized crossover trial in 20 mechanically ventilated ICU patients tested whether the breath hold could be skipped. The correlation between perfusion maps obtained with and without apnea was high, with a median correlation coefficient of 0.94, and global ventilation-perfusion match, dead space, and shunt measurements were comparable between the two approaches.¹¹ Eliminating the breath hold matters because it makes the technique feasible in patients who cannot tolerate even brief pauses in ventilation. Earlier work had established the methodology of using hypertonic saline injection during a respiratory pause to calculate dead space, shunt, and ventilation-perfusion match percentages from EIT-derived images.¹²

How EIT Compares to CT and Ultrasound

A natural question is how much you can trust EIT images when the gold standard remains CT. A study comparing EIT to CT in 32 non-intubated subjects found that regional ventilation distribution measured by EIT closely matched CT, with biases below about 1.3% and limits of agreement within roughly 10% across posterior and right-left lung regions. Body mass index, sex, fat percentage, and chest circumference did not meaningfully affect the agreement.¹³

EIT has also been compared with lung ultrasound in a clinically important scenario: predicting extubation failure. In patients being weaned from a ventilator, those who failed extubation had less lung surface available for ventilation on EIT and higher scores on lung ultrasound before the breathing tube was removed. Both tools picked up differences between success and failure groups, suggesting they capture complementary information about lung aeration.¹⁴

Stroke Detection With EIT

Stroke is a time-critical emergency where the treatment for a blood clot is the opposite of the treatment for a bleed. Getting the wrong one can be fatal. CT or MRI scanning is needed to tell the difference, but that requires reaching a hospital with the right equipment. Researchers have been exploring whether EIT could provide a faster, portable alternative for early stroke classification.

A multi-frequency symmetry-difference EIT approach combined with machine learning achieved an average accuracy of about 85% when distinguishing bleeding from clot in human data, and roughly 77% accuracy when distinguishing stroke from no stroke.¹⁵ A simulation study using realistic head models tested classifier accuracy at various noise levels and found roughly 80% accuracy for lesion differentiation and about 75% for lesion detection under realistic signal-to-noise conditions.¹⁶ These numbers are nowhere near replacing a CT scan, but the goal is not diagnostic certainty. The goal is triage: giving paramedics a tool that can suggest whether a patient is bleeding or clotting so treatment can begin during transport.

Cardiac Output Monitoring

Because the heart is surrounded by lungs, and blood volume changes with each heartbeat cause measurable impedance shifts, researchers have been investigating whether EIT can track stroke volume and cardiac output non-invasively. A narrative review confirmed that EIT can provide continuous non-invasive stroke volume measurements.¹⁷

The results so far are mixed. In a swine model, EIT-derived stroke volume showed good trending ability, with 94.2% concordance against an invasive reference method across 104 data pairs, though the overall percentage error was about 25%.¹⁸ In healthy human volunteers, however, the results were less encouraging. ECG-gated 3D EIT produced a mean error of 0.0 ± 15.2 mL for absolute stroke volume estimation and a concordance rate of only about 81% for trending relative changes, which the authors described as insufficient for most clinical uses.¹⁹ Cardiac EIT remains a work in progress. The technology tracks the direction of change reasonably well but lacks the precision needed for absolute measurements in humans.

Neonatal and Pediatric Care

Babies in neonatal intensive care units are some of the most challenging patients to image. They are tiny, fragile, and cannot be moved easily. EIT belts sized for neonates offer continuous bedside monitoring without radiation exposure. In one case, a premature infant born at 34 weeks with respiratory distress syndrome was being monitored with EIT as part of an observational study. Retrospective analysis of the EIT data revealed a progressive pattern suggesting pneumothorax roughly three hours before a chest X-ray confirmed a left-sided pneumothorax. The EIT signals showed increased air content on the affected side, reduced ventilation there, and decreased impedance on the opposite side consistent with mediastinal shift.²⁰

A separate study in preterm infants on high-frequency ventilation showed that the cross-sectional lung volume changes measured by EIT were representative of whole-lung volume changes, validating EIT as a monitoring tool even in very small patients whose lungs extend well above and below the electrode belt.²¹ For clinicians managing newborns on ventilators, this means EIT is not just measuring a thin slice of lung; it reflects what is happening throughout.

Wearable Hardware and Dry Electrodes

Traditional EIT systems use wet gel electrodes, which dry out over hours, irritate skin during prolonged monitoring, and require trained staff to apply. Two separate lines of research are working to change that. One group designed a wearable wireless EIT belt with dry electrodes that adapts to different body sizes, eliminating the need for gel and simplifying application.²² Another team developed textile electrodes integrated into a clothing belt, combining them with a custom portable EIT system for lung imaging.²³ Both approaches point toward a future where EIT can move out of the ICU and into ambulances, recovery wards, or even home monitoring for patients with chronic lung disease.

Practical Limitations

EIT is not without headaches. The most common practical problem is motion artifacts. When a patient shifts position, coughs, or is repositioned by staff, the baseline impedance signal can jump abruptly and not return to its previous level. Spike-like and drifting artifacts also occur. A study developing artifact-removal algorithms showed that image errors from these disturbances could be reduced dramatically, by roughly 89% for baseline drifting, 88% for step-like artifacts, and 98% for spike-like artifacts, while preserving the underlying signal.²⁴

Electrode contact quality is another persistent issue. If an electrode loosens or fails, the resulting bad data can corrupt the entire image. Researchers have developed diagnostic schemes that detect faulty electrodes and adaptively correct the erroneous measurements, aiming to maintain image quality even when one or more electrodes degrade during use.²⁵ These engineering problems may sound mundane compared to the clinical applications, but they are the main barrier between EIT as a research tool and EIT as a routine bedside standard.

There are also inherent physical limitations. EIT produces two-dimensional cross-sectional images at the electrode plane, and spatial resolution is far coarser than CT. The images are best understood as maps of regional function rather than anatomical detail. Clinicians trained on high-resolution CT scans sometimes find EIT images hard to interpret, which is why standardization efforts have focused on developing common terminology, processing pipelines, and display conventions.

Beyond the Chest

While lung monitoring dominates EIT research, the technology is being explored in several other body regions. Electrical impedance spectroscopy, a closely related technique, has been investigated for cancer detection, exploiting the fact that malignant tissue has different electrical properties than normal or benign tissue. A clinical study used impedance measurements at multiple frequencies to differentiate malignant from nonmalignant anomalies based on the Cole model, which characterizes how tissue impedance varies with frequency.²⁶

Gastrointestinal applications have also been explored. EIT has been used to track gastric emptying of liquids in real time, with results that correlated well with radiotelemetry capsule measurements. One study found that acidic liquid (pH 3) took significantly longer to empty from the stomach than neutral liquid (pH 7), with mean gastric residence times of roughly 35 to 47 minutes versus 14 to 23 minutes.²⁷ While gastric EIT has not reached routine clinical use, it offers a radiation-free alternative for studying gut motility.

Veterinary Applications

EIT has found a practical foothold in veterinary medicine, particularly in equine anesthesia. Horses are prone to severe lung collapse under general anesthesia due to their body weight compressing dependent lung regions. EIT allows monitoring of ventilation distribution during surgery, and one study found a strong linear correlation between EIT impedance changes and tidal volume measured by spirometry in individual horses, with correlation coefficients averaging 0.99.²⁸ The correlations were highly individual, however, meaning calibration had to be done for each horse rather than using a universal conversion factor.

EIT has also been used to assess the effectiveness of alveolar recruitment maneuvers in anesthetized horses, showing that the proportion of tidal volume reaching the dependent (lower) lung regions increased during recruitment, presumably as collapsed tissue reopened. The researchers suggested that monitoring dependent-lung compliance with EIT could substitute for repeated blood gas measurements when optimizing PEEP settings in horses.²⁹ The veterinary work is interesting in its own right, but it also serves as a testing ground for techniques that may eventually translate back to human critical care, where the same physiological principles apply but with fewer opportunities for controlled experimentation.