Regional ventilation dynamics of electrical impedance tomography validated with four-dimensional computed tomography: single-center, prospective, observational study

Abstract

Background

The dynamic regional accuracy of electrical impedance tomography has not yet been validated. We aimed to compare the regional accuracy of electrical impedance tomography with that of four-dimensional computed tomography during dynamic ventilation.

Methods

This single-center, prospective, observational study conducted in a general intensive care unit included adult patients receiving mechanical ventilation from July 2021 to February 2024. The patients were mechanically ventilated passively and underwent electrical impedance tomography and four-dimensional computed tomography on the same day.

Results

Overall, 45 patients were analyzed. The correlation coefficients in regional dynamic ventilation between four-dimensional computed tomography and electrical impedance tomography in each region were 0.963, 0.963, 0.835 (ventral, central, and dorsal, respectively) in the right lung and 0.947, 0.927, 0.823 (ventral, central, and dorsal, respectively) in the left lung. The correlation coefficient was low when the regional ventilation distribution detected by the electrical impedance tomography was < 2%. After excluding nine patients with a regional ventilation distribution of < 2%, the ventral, central, and dorsal correlation coefficients were 0.963, 0.963, and 0.946 in the right lung and 0.942, 0.924, and 0.951, respectively, in the left lung.

Conclusions

Regional ventilation using electrical impedance tomography during dynamic ventilation was highly accurate and consistent with the time phase compared to four-dimensional computed tomography. Given the high correlation between these modalities, they can contribute significantly to further studies on regional ventilation dynamics.

Trial registration number ClinicalTrials.gov (No. UMIN00044386).

Supplementary Information

The online version contains supplementary material available at 10.1186/s13054-024-05130-8.

Background

Electrical impedance tomography (EIT) provides real-time evaluation of respiratory dynamics by monitoring impedance changes. EIT is effective in evaluating ventilatory patterns [1] and determining the optimal positive end-expiratory pressure (PEEP) settings [2–4]. However, while EIT makes it possible to monitor impedance changes and estimate ventilation distribution, EIT generates images from relative impedance values and not absolute impedance values, making it impossible to assess ventilation volume directly. Consequently, studies have been directed toward validating the accuracy of EIT for lung volume estimation.

Typically, lung volume is measured using helium dilution, nitrogen washout, or computed tomography (CT). A comparative study of the end-expiratory lung volumes estimated using EIT and helium dilution demonstrated high concordance and low variability [5]. However, discrepancies exist in the correlation between lung volumes assessed by nitrogen washout and EIT; while one study has reported a strong correlation [6], another has indicated poor alignment [7]. Static [8] and dynamic [1, 9, 10] CT results show a strong correlation with EIT results. However, these techniques acquire data with patients adopting a breath-holding position, making it impossible to assess the dynamic volume changes during inspiration and expiration. Moreover, dynamic CT scans focus on a limited section of the lung, providing insufficient data for whole-lung volume comparisons. EIT generally yields precise measurements of the lung regions directly beneath the electrode belt; however, its accuracy in measuring dynamic ventilatory changes at the lung apex and base remains uncertain [11, 12].

For a detailed examination of ventilatory dynamics, an imaging modality that synchronizes with the respiratory cycles is required. Dynamic ventilation CT (four-dimensional CT [4D-CT]) integrates respiratory data into dynamic CT scans, yielding three-dimensional images over time. Although 4D-CT has been primarily used in radiotherapy [13–15], pathological analyses of chronic obstructive pulmonary disease [16–19], and animal studies [20–22], its application in mechanically ventilated patients is not widespread [23–25]. Using this technique, the entire lung can be scanned during uninterrupted mechanical ventilation (MV). We previously demonstrated a robust correlation among ventilator volume, EIT, and 4D-CT in the evaluation of whole-lung ventilatory dynamics [26], and showed that EIT information reflects the ventilatory dynamics of the entire lung, rather than just the area covered by the belt. However, it remains unclear how accurately the regional ventilatory dynamics of lung volume with complex lung structures can be assessed. We, therefore, aimed to validate the accuracy of regional impedance changes during dynamic ventilation by comparing them with air volume changes measured using 4D-CT.

Methods

Study design

This prospective, observational study was conducted in the general intensive care unit (ICU) of a university hospital in Tochigi, Japan. Patients aged > 18 years under MV with 4D-CT and EIT data were analyzed from July 2021 to February 2024 (UMIN00044386, registration date: June 1, 2021). This study was approved by the Institutional Research Ethics Committee of Jichi Medical University Hospital (approval number: A20-103). Written informed consent was obtained from each participant or their nearest relative. Patients younger than 18 years and those with a pneumothorax, mediastinal emphysema, or esophageal hiatal hernia were excluded. The ventilation modes, including pressure- or volume-controlled ventilation and ventilator settings, were selected by the same physicians during the EIT and 4D-CT measurements. All patients were deeply sedated and breathed passively during measurements. Patients with spontaneous breathing were not included in the study. The presence or absence of spontaneous breathing was confirmed by examining the ventilator graphics and the patient's breathing pattern. Some of the cases in this study have been published as research letters [26].

Participants

Data on the baseline characteristics of the patients, including age, sex, height, body weight on admission to the ICU, day of four-dimensional computed tomography (4D-CT), predicted body weight (PBW), body mass index, and type of ICU admission, were collected from patient electronic medical records. The PaO₂/F_IO₂ (P/F) ratio on days 1 and 4D-CT, the use of continuous renal replacement therapy, and prone positioning were also determined. Medical history, including chronic obstructive pulmonary disease, was evaluated. Regarding ventilatory parameters, the ventilation mode, peak inspiratory pressure, PEEP, plateau pressure (P_plat), auto-PEEP, respiratory rate, tidal volume (V_T), and inspiratory oxygen concentration (F_IO₂) were evaluated. The Acute Physiology and Chronic Health Evaluation II score [27] was used to assess the severity of each patient’s illness. Mortality rates during hospital stays were also evaluated.

MV protocol and parameters

HAMILTON-C6® (Hamilton Medical AG, Switzerland) was used to ventilate the patients during EIT and 4D-CT. A MemoryBox (Hamilton Medical AG, Switzerland) was used as an external recording device to record ventilatory parameters. Ventilation modes, including pressure-controlled ventilation and volume-controlled ventilation, were selected by the physicians. The ventilator settings were as follows: V_T = 6–8 mL/PBW. The respiratory rate, inspiratory time, F_IO₂, and minute volume were set as appropriate. The average V_T for 10 breaths was calculated, excluding abnormal values (outside the interquartile range (IQR) × 1.5, based on the waveform amplitude and wavelength).

EIT protocol and parameters

EIT was performed using a BB² EIT device (Sentec AG, Landquart, Switzerland) or a PulmoVista 500 (Dräger, Lubeck, Germany). The EIT belt was positioned around the patient's chest at the fourth and fifth intercostal space. The ventilator settings were kept consistent for both EIT and 4D-CT. The scan rate was 20 Hz for the PulmoVista 500 and 47 ± 2 Hz for the BB² EIT device. Regarding the image reconstruction algorithms of both devices, the PulmoVista 500 used the Newton–Raphson reconstruction algorithm, while the BB² EIT device used the Graz Consensus Reconstruction algorithm for EIT (GREIT). In this study, both reconstruction algorithms and data processing did not rely on thoracic models derived from individual CT scans. Specifically, the BB² EIT device comes equipped with multiple presets for standard thoracic shapes and lung contours based on CT data, which are automatically selected based on patient information such as height and weight. However, since the lung contours were defined in two dimensions, ventilated areas near the heart could potentially be excluded when viewed in three dimensions. Therefore, in this study, the analysis was performed without setting the lung contours. On the other hand, the PulmoVista 500 used an elliptical geometry. EIT data were recorded for several minutes during stable passive breathing before or after 4D-CT. Ten stable continuous breaths without outliers were selected and averaged for analysis using ibeX software (Sentec AG) or PV500 Data Analysis (Dräger). The following two points were considered outliers when analyzing stable respiration using the EIT:

1. Breaths for which the signal quality was poor or unacceptable.

2. Breaths for which the global impedance change and respiration times were beyond the IQR × 1.5.

Breaths before and after the outliers were also excluded to avoid missing waveforms around the outliers. Averaged breath was classified into 20 phases, from the end of inspiration to the end of the next inspiration, to correspond with the 4D-CT data.

In this study, EIT did not specifically define lung regions for data analysis; instead, it divided the entire area from the body surface to the thoracic cavity into three equal parts. This approach was adopted because EIT does not provide anatomical assessment, meaning that the lung region settings may not always be accurate. To avoid ambiguity associated with lung region settings, this validation study defined the region up to the chest wall for EIT and ensured consistency by analyzing all the obtained data. The ROI setting in EIT was based on the two software packages. The ibeX software automatically generated six ROIs—left and right, ventral, central, and dorsal—based on the data. In the Dräger software, the ROIs were manually set by dividing the displayed screen into two equal parts from left to right and three equal parts from ventral to dorsal (Figure S1). The ventilation distribution for each ROI, as determined by EIT, was calculated as a proportion of the entire lung region.

4D-CT protocol

4D-CT was performed using an 80-row multi-detector scanner (Aquilion Prime SP; Canon Medical Systems, Otawara, Japan). An AZ-733VI (Anzai Medical, Tokyo, Japan), with a laser that sweeps across the body surface and a sensor that detects the reflection of the laser light, was used as the respiratory gating system. The sensor presents the movement of the body surface (displacement of the distance to the sensor) caused by the respiratory motion in the form of a respiratory waveform. By capturing respiratory information, a 4D-CT scan synchronized with the patient’s breathing was acquired. 4D-CT was performed in helical scan mode to cover the entire thorax. The following scanning and reconstruction parameters were set: tube current = 40 or 60 mA, depending on body weight; tube voltage = 120 kV; rotation time = 0.35 s; total scanning time based on the respiratory rate; imaging field of view = 320–500 mm, depending on body size; collimation = 0.5 mm; slice thickness = 0.5 mm; reconstruction interval = 20 frames/breath; deep learning reconstruction = Advanced Intelligent Clear-IW Engine Body Sharp Standard [28] and half-reconstruction [29]; and slice thickness/interval = 0.5 mm/0.5 mm. Radiation exposure was adjusted to be below the recommended dose according to the National Diagnostic Reference Levels in Japan, 2020 edition. The 4D-CT images were constructed under mediastinal conditions and analyzed using SYNAPSE VINCENT (FUJIFILM, Tokyo, Japan).

Analysis of 4D-CT data using SYNAPSE VINCENT

One breath was defined as the point from the end of inhalation to the next end of inhalation and was categorized into 20 phases. Inspiratory and expiratory endpoints were set to coincide by checking the volume waveform of the 4D-CT and the ventilator waveforms. SK and KT made the final decision based on waveform information. Trimming of the lung area was partially performed automatically using the VINCENT application and manually by SK and KT.

The CT number was extracted from -1,000 Hounsfield units (HUs) to + 100 HUs for each voxel, as previously described [30]. Air volume (vol %) was calculated using the formula:

$$\text{Air volume}=\text{SUM}\left(-\text{CT number}/1,000\times \text{voxel}\right)$$

Negative values were regarded as zero. Six regions of interest (ROIs) were categorized into the right and left lungs, which were separated anatomically. The ventral, mid, and dorsal separation was set up by classifying the most ventral and dorsal lung regions equally into three sections. Each ROI was manually set by the researcher. The reference point was set at the carina position in the axial images during the end-expiratory phase (Figure S1).

Primary endpoints

The primary endpoint was the agreement between 4D-CT and EIT, with normalization across the 20 phases of each ROI.

Statistical analysis

All analyses were conducted using JMP 17 Pro (SAS Institute Inc., Cary, NC) and Microsoft Excel. Data are presented as medians and interquartile ranges (25th–75th percentiles) or percentages. The relationship between EIT and 4D-CT was evaluated using a linear correlation coefficient and Bland–Altman plot. The twenty phases of the EIT and 4D-CT images were aligned at the end-inspiratory phase. The absolute value of the impedance can vary from case to case depending on the amount of water in the body and contact with the belt. Therefore, the air content obtained using the air volume change in 4D-CT and the impedance change in EIT were normalized (end inspiration [x-axis = 1, y-axis = 1] and end expiration [0, 0]) and compared. As defined earlier, the phase identified as end-inspiration does not always represent the maximum value in each region in all cases. Consequently, in these instances, the normalized values may exceed 1. If the EIT’s ventilation distribution of any ROI was < 2%, the cases were excluded due to small tidal signals and reanalyzed in a sensitivity analysis.

Results

Enrollment and baseline characteristics

Forty-five patients were included in the study. Table 1 summarizes the patient characteristics, while Table 2 presents the ventilator parameters for 4D-CT. The V_T/PBW and compliance were 7.05 mL/kg (6.63–7.38 mL/kg) and 40.6 mL/cmH₂O (31.0–59.2 mL/cmH₂O), respectively.

Table 1.

Patient characteristics

Characteristics	All patients
	(n = 45)
Age, years	64 (50–69)
Sex, male	84.4%
Height, cm	165 (159–171)
Weight, kg (admission to ICU)	71.3 (58.0–83.3)
BMI, kg/m2	26.6 (23.0–29.2)
APACHE II	18 (15–22)
Prone position	35.6%
Reasons for admission
Intra-pulmonary diseases
Coronavirus disease 2019 (n)	8
Interstitial pneumonitis (n)	4
Aspiration pneumonia (n)	3
Bronchial asthma (n)	1
Tuberculosis (n)	1
Pneumocystis pneumonia (n)	3
Eosinophilic pneumonitis (n)	1
Inhaled pneumonitis (n)	1
Legionnaires' disease (n)	2
Other pneumonia (n)	3
Alveolar hemorrhage (n)	2
Extra-pulmonary diseases
Acute heart failure (n)	1
Cerebral vascular disorder (n)	1
Sepsis (n)	7
Hematologic malignancies (n)	4
Other (n)	3
Medical history
COPD	4.4%
P/F ratio (admission to ICU)	156 (96–205)
P/F ratio classification
≥ 300 mmHg	8.9%
≤ 200 to < 300 mmHg	17.8%
≤ 100 to < 200 mmHg	48.9%
< 100 mmHg	24.4%
P/F ratio (day of 4D-CT)	260 (199–337)
P/F ratio classification
≥ 300 mmHg	33.3%
≤ 200 to < 300 mmHg	42.2%
≤ 100 to < 200 mmHg	22.2%
< 100 mmHg	2.2%
MV duration, days	19 (9–36)
Hospital mortality	26.7%

APACHE II, Acute Physiology and Chronic Health Evaluation II; BMI, body mass index; COPD, chronic obstructive pulmonary disease; CRRT; Continuous renal replacement therapy; ICU, intensive care unit; MV, mechanical ventilation; P/F, PaO₂/F_IO₂ ratio; 4D-CT, four-dimensional computed tomography

Table 2.

Ventilator parameters

	All patients
	(n = 45)
Mode of ventilation
PCV	35.6%
VCV	64.4%
PIP, cmH2O	24 (21–30)
PEEP, cmH2O	8 (5–12)
Pplat, cmH2O	19 (17–23)
Auto PEEP, cmH2O	0.3 (0.0–0.8)
Respiratory rate, /min	18 (16–20)
Tidal volume, mL	417 (364–500)
Tidal volume/PBW, mL/kg	7.05 (6.63–7.38)
FIO2	0.4 (0.3–0.5)
Compliance, mL/cmH2O	40.6 (31.0–59.2)

F_IO₂, fraction of inspired oxygen; PCV, pressure-controlled ventilation; PEEP, positive end-expiratory pressure; PIP, peak inspiratory pressure; Pplat, plateau pressure; VCV, volume-controlled ventilation; PBW, predicted body weight.

Regional dynamic ventilation in ROIs

Overall, 900 images (one image at each time point, in total, 20 images per patient) were acquired. The correlation coefficients of 4D-CT and EIT were 0.963 (ventral), 0.963 (central), and 0.835 (dorsal) in the right lung, and 0.947 (ventral), 0.927 (central), and 0.823 (dorsal) in the left lung. Additionally, a Bland–Altman plot was prepared to evaluate bias and precision with mean ± standard deviation values of 0.02 ± 0.09, 0.03 ± 0.09, and 0.03 ± 0.21 for the right lung and 0.02 ± 0.11, 0.02 ± 0.13, and -0.01 ± 0.23 for the left lung, respectively (Figs. 1, 2).

Fig. 1.

Correlation coefficients of normalized values of four-dimensional computed tomography (4D-CT) and electrical impedance tomography (EIT). The horizontal axis denotes the ventilated volume on 4D-CT, and the vertical axis denotes the normalized impedance values of EIT. The correlation coefficients were slightly low on the bilateral dorsal side but very high on the ventral side and in the center. RD and LD are more variable; therefore, the scale varies compared with the other ROIs. LV, left ventral; LC, left central; LD, left dorsal; RV, right ventral; RC, right central; RD, right dorsal

Fig. 2.

Bland–Altman plot to evaluate the accuracy of regional dynamic ventilation of EIT with 4D-CT. The horizontal axis denotes the mean between 4D-CT and EIT, and the vertical axis denotes the difference between EIT and 4D-CT (EIT − 4D-CT). The middle solid line presents the mean difference between EIT and 4D-CT. The upper and lower dotted lines represent the limits of agreement. RD and LD are more variable; therefore, the scale is varied compared to other ROIs. The limits of agreement ranged from −0.16 to 0.20 (ventral), −0.15 to 0.20 (central), and −0.39 to 0.45 (dorsal) in the right lung and from −0.20 to 0.23 (ventral), −0.22 to 0.27 (central), and −0.46 to 0.43 (dorsal) in the left lung. LV, left ventral; LC, left central; LD, left dorsal; RV, right ventral; RC, right central; RD, right dorsal

Sensitive analysis excluding ventilation distribution < 2%

The ventilation distribution is shown in Fig. 3. In ROIs with a lower ventilation distribution, the waveform of the impedance changes had a clearly flattened or reversed shape in some cases (Figure S2). Extremely small ventilation was found particularly in the dorsal regions. Descriptive statistics showed that the 25th percentile of regional ventilation distribution in the right and left dorsal regions was 4.0% and 3.0%, respectively (Fig. 3). Based on these findings, the subsequent sensitivity analysis excluded nine patients who had a ventilation distribution of less than 2% in either the right or left dorsal region. The number of ROIs corresponding to ventilation distribution below 2% was 14 (5.2%). In a sub-analysis of this population, the correlation coefficients between 4D-CT and EIT were 0.963 (ventral), 0.963 (central), and 0.946 (dorsal) in the right lung, and 0.942 (ventral), 0.924 (central), and 0.951 (dorsal) in the left lung. A Bland–Altman plot was constructed to evaluate bias and precision, with mean ± standard deviation values of 0.02 ± 0.09, 0.03 ± 0.09, and 0.01 ± 0.11 for the right lung and 0.02 ± 0.11, 0.03 ± 0.13, and 0.02 ± 0.10 for the left lung, respectively (Figs. 4, 5).

Fig. 3.

Ventilation distribution of six ROIs in EIT. The vertical axis shows the percentage of ventilation distribution. The ventilation distribution was 18% (14–22%), 31% (24–38%), 6% (4–9%) in RV, RC, and RD, respectively, and 17% (12–20%), 22% (18–30%), and 5% (3–9%) in LV, LC, and LD, respectively. LV, left ventral; LC, left central; LD, left dorsal; RV, right ventral; RC, right central; RD, right dorsal

Fig. 4.

Correlation coefficients of the normalized values of 4D-CT and EIT. Patients with very low ventilation distribution of EIT (< 2%) were excluded from this sensitive analysis. The horizontal axis denotes the ventilated volume on 4D-CT, and the vertical axis denotes the normalized impedance values of EIT. Correlation coefficients were very high in all areas. LV, left ventral; LC, left central; LD, left dorsal; RV, right ventral; RC, right central; RD, right dorsal; EIT, electrical impedance tomography; 4D-CT, four-dimensional computed tomography

Fig. 5.

Bland–Altman plot to evaluate the accuracy of regional dynamic ventilation of EIT with 4D-CT. This excluded the patients who presented very low ventilation distribution of EIT (< 2%) as a sensitive analysis. The horizontal axis denotes the mean between 4D-CT and EIT, and the vertical axis denotes the difference between EIT and 4D-CT (EIT − 4D-CT). The middle solid line presents the mean difference between EIT and 4D-CT. The upper and lower dotted lines mean the limits of agreement. The limits of agreement ranged from − 0.16 to 0.20 (ventral), − 0.15 to 0.20 (central), and − 0.21 to 0.22 (dorsal) in the right lung and from − 0.20 to 0.24 (ventral), − 0.23 to 0.28 (central), and − 0.19 to 0.22 (dorsal) in the left lung. LV, left ventral; LC, left central; LD, left dorsal; RV, right ventral; RC, right central; RD, right dorsal

Discussion

We observed a very high correlation between the data obtained from the EIT and 4D-CT during regional dynamic ventilation, which is consistent with the timing phases. Notably, when the EIT was categorized into six ROIs, the correlation coefficient was lower when the ventilation distribution of a particular ROI was < 2%. This is to be expected, as when the signal becomes very small, the measurement results approach the limits of measurement accuracy, leading to a substantial increase in relative error. Consequently, this inherently results in a poor correlation when comparing different measurement methods. A strong correlation was observed in regions where the ventilation signal is at least 2%, indicating the reliability of measurements in this range when compared with standard axial CT imaging. To our knowledge, this is the first study to investigate the accuracy of EIT in assessing regional ventilatory dynamics in the entire lung.

The advantage of EIT is that it allows non-invasive monitoring of ventilatory dynamics at the bedside. However, whether EIT can accurately measure dynamic regional ventilation changes was not validated owing to its limited spatial resolution. Our results suggest that EIT has robust utility and reliability as a bedside tool to assess regional ventilation dynamics. Furthermore, previous studies have validated the accuracy of EIT under both static [5–8] and dynamic conditions [1, 9, 10, 31, 32]. However, the consistency of EIT with dynamic changes in local lung capacity has been reported in only a few articles [1, 10, 22]. EIT generates images from relative values and not absolute values, making it difficult for physicians to predict lung volume and capacity accurately. Our findings indicate that regional impedance variations detected by EIT closely correlate with the regional tidal volume, suggesting their utility in reflecting regional ventilation dynamics, particularly in low-tidal volume ventilation.

In this study, dynamic regional impedance changes identified by EIT were congruent with alterations in the regional air volume measured by 4D-CT. Previous studies have validated the accuracy of dynamic CT using ROIs, but assessments were confined to fixed-width slices of the transverse section and did not encompass the whole lung [9, 10]. We discovered that EIT ROI information was highly correlated with whole-lung volume from the apex to the base. Although the sensitivity decreased when moving away from the EIT belt, given the elliptical model of impedance variation, we believe that our results reflect lung volume changes per ROI more accurately.

Our findings revealed that the ROI settings determined by EIT corresponded well with the ROIs defined by axial CT imaging. Although the EIT belt was placed according to the manufacturer’s recommendations, the cross-section encircled by the EIT belt did not necessarily match the axial plane of the CT. However, axial CT images were used in this study because it is generally easier to understand the assessment of ventilation distribution and evaluate the effect of gravity, which reflects pathologies such as acute respiratory distress syndrome and atelectasis. The results showed good agreement and are expected to be useful for future clinical applications.

The left and right lungs were anatomically categorized into an ROI setting using 4D-CT. Previous studies have been validated by matching axial CT images, and in setting the ROI, the left and right lungs were separated at the midline without distinguishing between them [10]. Even when using our method, evaluation by EIT reflected the left and right lungs with a high degree of accuracy.

The correlation coefficients for the dorsal ROI were slightly lower than those for the ventral and central ROIs. This correlation may be weakened by a small ventilation distribution, which may be affected by cardiac oscillations and other factors (Figure S2). Particularly in critically ill patients, including those with acute respiratory distress syndrome, dorsal ventilation distribution is likely to be reduced owing to dorsal atelectasis and pleural effusions, requiring careful attention. In our study, the lungs were categorized into six ROIs for analysis; therefore, the values of ventilation distribution in each pixel and other ROI settings, such as two ROIs and eight ROIs, were not validated. However, our results suggest that the regional ventilation dynamics with EIT are accurate if a certain ventilation distribution is obtained and is not extremely small. Poor correlations observed in ROIs with minimal ventilation are considered to have negligible clinical impact. Therefore, to assess regional ventilation dynamics using EIT, the ventilation distribution must be evaluated.

The applicability of 4D-CT has been corroborated in non-ventilated patients and animal studies; however, our study accurately assessed the ventilatory dynamics in patients. In one animal study, 4D-CT effectively distinguished between collapsed and reopened alveoli, shedding light on lung injury [33]. Despite the need for low-dose radiation in 4D-CT to minimize exposure, the image resolution is reduced. Nevertheless, the use of advanced techniques, such as deep learning reconstruction [28, 34–36] and half-reconstruction [29], has proven successful in preserving image quality. Based on our previous study [26], the margin of error in calculating air volume changes using 4D-CT was minimal, supporting the notion that this method could be instrumental in evaluating ventilator-induced lung injury [33]. Although 4D-CT was performed during passive breathing in this study, it can be performed during spontaneous breathing, which could allow us to analyze four-dimensional diaphragm movements [23–25] and the pathophysiology of lung injury caused by patient self-injury, which has not yet been clarified.

Our study has some limitations. First, two researchers manually delineated the lung fields, in addition to automatic segmentation using software, which could have introduced variability. However, the high accuracy of the EIT estimates, as reported in our previous study [26], suggests consistency in the manual delineation. Second, although 4D-CT breaks down a single breath into 20 phases, the imaging interval for 4D-CT is longer than that for EIT, which may result in slight discrepancies between the time phases. For instance, if a breath lasts for 3 s, each phase in 4D-CT spans 0.15 s, whereas EIT captures 20–50 images per second, with each image representing a 0.02–0.15 s interval. However, in this study, despite the interval difference between EIT and 4D-CT, the correlation coefficient was very high. The high dynamic accuracy of EIT across these phases suggests that a higher accuracy could be achieved regardless of the number of time points, assuming that the same matching method was used. Third, differentiating pleural effusion from atelectasis is challenging because of low radiation imaging, which potentially leads to inaccuracies in determining tidal recruitment. However, this study focused on the air volume assessment to mitigate the impact of this issue. Fourth, the limited number of patients eligible for 4D-CT because the scans were scheduled at specific times on weekdays, coupled with the limited number of researchers, may have introduced selection bias. Despite this, the diversity of the patient population, which included individuals with respiratory failure, suggests that our findings are applicable to critically ill patients. Fourth, 4D-CT and EIT data were not recorded concurrently. However, because the scans were performed in close succession under identical ventilation settings, we believe that the resultant errors are minimal. Fifth, in this study, the ROI settings for 4D-CT and EIT were not explicitly set in the same way. This may have contributed to the low correlation between 4D-CT and EIT in the dorsal region. By aligning the ROI settings for 4D-CT to those used for EIT, this issue could potentially be resolved. In the future, it may be beneficial to perform offline reconstruction using individual geometric data referenced from CT scans based on the current study data to improve EIT image reconstruction. Finally, ventilation distributions below 2% were excluded in this study. Fourteen ROIs corresponded to such regions, and a total of nine cases were excluded. In such pathological conditions, atelectasis and pleural effusion are frequently observed, making the accurate evaluation of ventilation distribution in these areas clinically relevant and valuable. However, caution is needed when assessing the ventilation dynamics in these regions. It is unclear whether time-phase ventilation delays (pendelluft phenomenon) or tidal recruitment can be accurately evaluated. Further detailed analysis of these areas in the future may lead to a more precise understanding of the underlying diseases.

Despite these limitations, EIT demonstrated high accuracy in assessing regional dynamic ventilation unless the ventilation distribution was exceedingly low. In particular, excellent agreement with lung volume changes from the apex to the base of the lung on regionalized axial CT images may improve physicians’ understanding of ventilatory dynamics.

Conclusions

This study confirmed that regional dynamic ventilation, as delineated by EIT, closely aligns with dynamic volume changes measured by 4D-CT across different lung regions. This significant correlation underpins the potential of these modalities in advancing research on regional ventilation dynamics.

Abbreviations

CT

Computed tomography

EIT

Electrical impedance tomography

ROIs

Regions of interest

MV

Mechanical ventilation

4D-CT

Four-dimensional computed tomography

ICU

Intensive care unit

PBW

Predicted body weight

PEEP

Positive end-expiratory pressure

Pplat

Plateau pressure

VT

Tidal volume

FIO2

Inspiratory oxygen concentration

IQR

Interquartile range

Author contributions

SK and SU designed the study; SK, KT, and AS performed the measurements; SK, KN, MT, and AS performed the statistical analysis; SK, AS, KT, and MS interpreted the data and drafted the manuscript. All authors read and approved the final manuscript.

Funding

This study was supported by a grant from the Japan Society for the Promotion of Science (grant JP21K16594), internal department funds, and commissioned research with NIHON KOHDEN CORPORATION.

Availability of data and materials

The datasets used and analyzed in the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the Institutional Research Ethics Committee of Jichi Medical University Hospital (approval number: A20-103). Written informed consent was obtained from each participant or their nearest relative. This study was conducted in compliance with the Declaration of Helsinki and the Ethical Guidelines for Medical Research Involving Human Subjects.

Consent for publication

Not applicable.

Competing interests

SK provided a consultation agreement with Hamilton Medical. The BB2 EIT device was rented through a research contract to conduct this study. The other authors declare that they have no competing interests.