Received 20 August 2022, accepted 21 September 2022, date of publication 26 September 2022, date of current version 4 October 2022. Digital Object Identifier 10.1109/ACCESS.2022.3209809 Effect of Calibration for Tissue Differentiation Between Healthy and Neoplasm Lung Using Minimally Invasive Electrical Impedance Spectroscopy GEORGINA COMPANY-SE 1 , LEXA NESCOLARDE 1 , VIRGINIA PAJARES2 , ALFONS TORREGO2 , PERE J. RIU 1 , (Senior Member, IEEE), JAVIER ROSELL 1 , (Senior Member, IEEE), AND RAMON BRAGÓS 1 1 Department of Electronic Engineering, Universitat Politècnica de Catalunya, 08034 Barcelona, Spain 2 Department of Respiratory Medicine, Hospital de la Santa Creu i Sant Pau, 08041 Barcelona, Spain Corresponding author: Lexa Nescolarde (lexa.nescolarde@upc.edu) This work was supported in part by the Spanish Ministry of Science and Innovation under Grant RTI2018-098116-B-C21; in part by the Secretariat of Universities and Research of the Generalitat de Catalunya and the European Social Fund developed in the Electronic and Biomedical Instrumentation Research Group of the Electronic Engineering Department, Universitat Politècnica de Catalunya; and in part by the Interventional Pulmonology Unit of the Respiratory Medicine Department of the Hospital de la Santa Creu i Sant Pau, Barcelona, Spain. This work involved human subjects or animals in its research. Approval of all ethical and experimental procedures and protocols was granted by the Ethics Committee on Clinical Investigation of the Hospital de la Santa Creu i Sant Pau, Barcelona, Spain, under Application No. CEIC-73/2010. ABSTRACT This study proposes a calibration method and analyses the effect of this calibration in lung measures, using minimally invasive electrical impedance spectroscopy with the 3-electrode method, for tissue differentiation between healthy and neoplasm lung tissue. Tissue measurements were performed in 99 patients [54 healthy tissue and 15 neoplastic tissue samples obtained] with an indicated bronchoscopy. Statistically significant difference (P < 0.001) were found between healthy lung tissue and neoplasm lung tissue in bioimpedance parameters. The calibration of the bioimpedance measures with respect to a measure performed in bronchi reduces the inter-patient dispersion, increasing the sensitivity, decreasing the specificity and increasing the area below the ROC curve for three out of four impedance-derived estimators. Results also show that there are no significant differences between healthy lung tissue among smoker, non-smoker and ex-smoker samples, which was initially stated as a possible cause of EIS measurement dispersion in lungs. INDEX TERMS Bronchi, bronchoscopy, calibration, electrical impedance spectroscopy (EIS), lung. I. INTRODUCTION lung cancer is the leading cause of cancer death in the world. Respiratory disorders have a big impact in the population Both are smoking-related conditions [1]. worldwide. According to the European Respiratory Society, In lung cancer, late detection in advance stages is common chronic obstructive pulmonary disease (COPD) is the third and is related to poor prognosis [2]. Diagnostic of lung global cause of death in more developed countries. Moreover, peripheral and central nodules is increasing because of num- ber of patients with indeterminate nodules are discovered in CT screening and verified with other diagnostic options such The associate editor coordinating the review of this manuscript and as minimally invasive bronchoscopic procedures to establish approving it for publication was Norbert Herencsar . final histological type. However, the diagnostic yield using This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. 103150 For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ VOLUME 10, 2022 G. Company-Se et al.: Effect of Calibration for Tissue Differentiation Between Healthy and Neoplasm Lung virtual bronchoscopy (VB), radial endobronchial ultrasound and the differences between inflated and deflated lung. Later (r-EBUS), electromagnetic navigation (EMN) and ultrathin Coll et al. [12] and Riu et al. [13] present studies demonstrat- bronchoscopes remains suboptimal [3], [4], and their high ing the potential for tissue differentiation through minimally economic cost makes them unavailable in most centers. invasive electrical impedance spectroscopy in lung using the We aim to use Electrical Impedance Spectroscopy (EIS) to 4-electrode method. complement the actual methods of diagnosis of lung diseases This manuscript (2nd phase) is the continuation of the pre- as it could allow the differentiation between healthy lung vious study (1st phase) entitled ‘‘Minimally invasive lung tis- tissue and neoplasm lung tissue and help in the choice of the sue differentiation using electrical impedance spectroscopy: specific sample location. a comparison of the 3- and 4- electrode methods’’ performed EIS technique is one of the existing methods of impedance by Company-Se et al. [14]. It compared the capacity of tissue analysis. Impedance is defined as the opposition to the flow of differentiation of the minimally invasive electrical impedance an alternating electrical current which is dependent on the fre- spectroscopy in lungs using the 4-electrode method and the quency of this current [5]. When the impedance is measured 3-electrode method. The results showed that both meth- in biological tissue is named as bioimpedance (Z). It measures ods were adequate for tissue differentiation but 3-electrode the passive electrical properties of the tissue after the intro- method was more feasible for its clinical use because of its duction of a low amplitude alternating current to the organism lower complexity, both in the catheter configuration (single [5], [6]. The bioimpedance is a complex number with a real electrode) and in the measurement system architecture. This part (the resistance, R) and an imaginary part (the reactance, previous study proposed for future works to increase the Xc), both parts are dependent of the geometry of the mea- sample size for the differentiation between healthy lung tissue sured region, the location of electrodes and the tissue elec- and neoplasm lung tissue using the 3-electrode method. trical passive properties [5]. The physiological fluids have In this 2nd phase the measures performed in healthy lung low resistance and dominates the measured resistance, while tissue and in neoplasm lung tissue showed high inter-patient cell membranes act as capacitors, having high impedance variability. This variability could hinder the tissue differenti- at low frequencies and low impedance at high frequencies ation in lungs. There are several causes for this variability: and contributes mainly to the reactive part. Due to these 1) The measured absolute values of the R and Xc spectra behaviors, the electrical current introduced in the biological are influenced by the tissue properties (the variable under tissue divides into resistive and capacitive pathways and it measurement) but also by the geometry of the measurement changes with the frequency [6]. An alternative representation (body shape of the patient and electrode positions). Geometri- of the Bioimpedance, as a complex number, is the use of the cal factors such as body mass index (BMI) has been reported modulus (Z) and the phase angle (PA). The PA represents as one significant factor for changes in lung metrics [15]; the relative time lag between the injected current and the 2) The breathing produces also impedance changes due to the generated voltage [7]. Bioimpedance data can be obtained considerable air volume change from inspiration to expiration using single or multiple frequencies. When the bioimpedance and the influence of the non-conductive air contents in the data is obtained using a broad band of frequencies is known lung tissue. This phenomenon could increase the inter-patient as bioimpedance spectroscopy [6]. The advantage of the EIS variability as depending on the patient, the breathing cycle method, to measure and analyze bioimpedance data, is based will be different; 3) In the 3-electrode method, the electrode on the fact that current at low frequency (lower than 10 kHz) impedance of the catheter tip is measured and could increase flows through the extracellular medium while current at high the intra- and inter-patient variability due to poor contact frequencies (over 100 kHz) flows through both, intracellular of the catheter tip against the lung tissue and the liquids and extracellular medium, giving more information about the accumulation in the airways; 4) Another potential cause for structure of the tissue. inter-patient variability is cigarette consumption. It could There are previous studies about lung bioimpedance mea- contribute to the increase of the inter-patient dispersion. surements. Toso et al. [8], through an impedance plethysmo- Smoking-induced epithelial abnormalities can serve both as graph emitting 50 kHz alternating current, reported different targets for abnormal inflammatory responses and as initia- impedance vector distribution in patients with lung cancer tors of deregulated inflammation. Cytokines, chemokines, as compared with healthy patients. A reduced Xc and a and growth factors released by alveolar macrophages, lym- smaller PA were found while R was preserved in patients phocytes, neutrophils, endothelial cells, and fibroblasts with lung cancer. Nierman et al. [9] performed transthoracic may act to promote epithelial dysfunction and malignant bioelectrical impedance analysis to quantify extravascular progression [16], [17]. lung water in animal models. Orschulik et al. [10] used While the ventilation-induced impedance modulation non-invasive bioimpedance spectroscopy for the diagnosis of effect can be reduced using averaging, the other potential acute respiratory distress syndrome in an animal model. causes of variability need a calibration method capable to Some previous studies have been carried out by our reduce this variability in order to perform tissue differ- research group. Sanchez et al. [11] performed minimally entiation with success. For example, electrical impedance invasive lung bioimpedance measurements to study the char- measures using the 3-electrode method in cardiology uses acteristics of lung bioimpedance (calibration and linearity) a floating measure within the heart (catheter completely VOLUME 10, 2022 103151 G. Company-Se et al.: Effect of Calibration for Tissue Differentiation Between Healthy and Neoplasm Lung FIGURE 1. Schematic representation of the bioimpedance acquisition system. surrounded by blood) to calibrate the geometrical factors 31.0 ± 4.3 kgm−2 ); (ex-smokers: n = 23, Age: 71 ± 12 yr; in the bioimpedance measures obtained in contact with the Weight: 79.3 ± 13.8 kg; BMI: 27.5 ± 4.8 kgm−2 ; years myocardial walls [18], and also, partially, the electrode without smoking = 22 ± 11 yr)] while the number of samples impedance effect. In lungs, a floating measure completely obtained from neoplasm lung tissue were 15 (Age: 70 ± 9 yr; surrounded by air to calibrate is not viable due to the Weight: 75.3 ± 11.2 kg; BMI: 26.3 ± 4.1 kgm−2 ). non-conductive property of the air and it is not feasible to Ethics approval was obtained from the Hospital de la Santa locate the catheter in a place where the tip electrode will be Creu i Sant Pau (CEIC-73/2020) according to principles surrounded by a well-known tissue and which will be affected of the Declaration of Helsinki for experiments with human by geometrical factors similar to the ones that will affect beings. All patients proved signed informed consent. the tissue impedance measurements for each patient. For this reason, we proposed to acquire a bioimpedance measure in principal bronchus and use it to calibrate the lung tissue B. MEASUREMENT SYSTEM bioimpedance measures. The acquisition of bioimpedance measures were performed The aim of this study, by using minimally invasive elec- using a tetrapolar catheter (Medtronic 5F RF Marinr), 115 cm trical impedance spectroscopy with the 3-electrode method, long with a diameter of 1.65 mm (5 F) and two skin electrodes is to propose a calibration method and to analyze the effect (Ambu BlueSensor VLC ref: VLC-00-s/10 and 3M Company of this calibration in measures performed in the bronchi for ref: 9160F) placed on the right side of the patients at the level tissue differentiation in different groups: healthy lung tissue of the ribs. Only the catheter tip electrode will be used in the (no radiological abnormalities in CT Thorax) and neoplasm measurements. lung tissue. Also, the possible differences in the impedance The measurement system is made up of 3 devices (Fig. 1): measurements in healthy tissue in smokers, non-smokers and 1) an optically insulated battery-powered patient interface ex-smokers will be verified to check if this factor would affect insulated front end (that includes the impedance front end); the ability to differentiate between healthy lung tissue and 2) a rugged PC platform based on a PXI system from National lung neoplasm. Instruments; and 3) an analog-optical interface front-end to connect the PXI with the insulated front end. An arbi- trary waveform generator generates a multisine excitation II. MATERIALS AND METHODS signal that is composed of 26 frequencies between 1 kHz A. PARTICIPANTS and 1 MHz. To ensure a current lower than the maximum Minimally invasive EIS measures were taken in 99 patients allowable patient auxiliary current stablished in the IEC (Age: 65 ± 16 yr; Weight: 76.8 ± 15.6 kg; BMI: 27.7 ± 60601-1:2005 (<1mA rms measured with the circuit pro- 5.5 kgm−2 ) with a bronchoscopy indicated during the period posed in the IEC 60601-1:2005) the front end includes an between November 2021 and February 2022 at the ‘‘Hos- AC-coupled current source that attenuates the low-frequency pital de la Santa Creu i Sant Pau’’. All of them underwent components accordingly with the current limit pattern speci- bioimpedance measurement. However, 30 of them had other fied by this standard. The system was verified including the characteristics than healthy lung tissue or neoplasm lung 26 frequency components (1 kHz – 1MHz) simultaneously. tissue such as emphysema or fibrosis. For this reason, out The voltage (V(t)) and current (I(t)) are simultaneously of the 99 patients measured by bioimpedance, only 69 were acquired. Then, with the optical-analog interface connected to considered for analysis (healthy: 54 and neoplastic: 15). the PXI, the excitation is converted into an optical signal. The The number of bioimpedance samples obtained in healthy optical signal is then converted back into an electrical signal lung tissue were 54 [(non-smokers: n = 22, Age: 59 ± 19 yr; inside the front end. The voltage and current signals, opti- Weight: 70.8 ± 16.6 kg; BMI: 26.7 ± 5.8 kgm−2 ); smokers: cally transmitted from the front end to the optical-electrical n = 9, Age: 66 ± 7 yr; Weight: 83.5 ± 11.9 kg; BMI: interface, are acquired with the digitizer card. The acquisition 103152 VOLUME 10, 2022 G. Company-Se et al.: Effect of Calibration for Tissue Differentiation Between Healthy and Neoplasm Lung FIGURE 2. Right figure: Schematic representation of the 3-electrode method. In the 3-electrode method the LC and LP electrodes are placed on the skin using skin surface electrodes. Left figure: dimensions of the catheter. Only the tip electrode of the catheter is used to perform the measurements. system takes simultaneous samples of voltage and current the ratio between the voltage and current coefficients of the at 20 MSamples/s. From the acquired signals, 60 impedance FFT corresponding to each injected frequency. spectra per second are obtained. The acquisition takes 12 s at 60 spectra per second. The 3- Bioimpedance measures were obtained using the electrode measurements were calibrated with a measurement 3-electrode method. To inject the current (HC) and detect the over a known resistor (600 Ohms) connected to the catheter potential (HP) the electrode located at the tip of the catheter tip and to the external electrode connectors. is used. The two skin electrodes are used as low current (LC) and loc potential (LP) electrodes (Fig. 2). E. CALIBRATION USING BRONCHUS C. MEASUREMENT PROTOCOL To remove the geometrical factors of the patients a multi- Bronchoscopy, a procedure used to inspect the airways, was plicative factor calibration of the bioimpedance of the lung performed to obtain the bioimpedance measures. As part of measures is proposed. The proposed method aims to calibrate the diagnostic process, radiological imaging technique (CT the lung measures with respect to a measure performed in or PET/CT) were performed in each patient before bron- the bronchial tissue (principal bronchus) for each respec- choscopy procedure. To obtain the bioimpedance measures, tive patient. A measurement in the bronchi is of no interest the catheter was inserted through a port of the bronchoscope. in clinical practice, therefore, impedance measurement in During the bronchoscopy, patients are placed in a supine bronchial tissue offers the advantage of calibration while position with the upper airways anaesthetized with topical 2% not losing relevant clinical information. Moreover, because lidocaine. Moreover, intravenous sedation is provided with of its low cell content, bronchial tissue should have a flat midazolam, fentanyl and propofol. During the process, mea- impedance spectrum, thus being suitable as calibration refer- sures in bronchial tissue, healthy lung tissue and neoplasm ence [14]. The obtained impedance modulus (|Z|) of the lung lung tissue, if applicable, were taken. The acquisition of the is divided by the mean value (mean value of impedance at measures had a duration of 12 seconds. each frequency, during a time interval) of |Z| of the bronchial tissue and then multiplied by a factor of 100 , which is the D. EIS MEASUREMENTS expected impedance magnitude value obtained in the bronchi To obtain the EIS measurements the system applies a mul- [12] (1). The PA calibrated of the lung measure is obtained by tisine current signal and acquires the voltage and current subtracting the original value of the PA of the lung measure signals. The Fast Fourier Transform (FFT) is used to obtain minus the mean value of the PA obtained in the bronchi tissue VOLUME 10, 2022 103153 G. Company-Se et al.: Effect of Calibration for Tissue Differentiation Between Healthy and Neoplasm Lung sample (2). III. RESULTS A. MULTI-FREQUENCY RESPONSE FOR MINIMALLY |Z (f , t)|calibrated INVASIVE HEALTHY LUNG TISSUE MEASUREMENTS = 100 ∗ |Z (f , t)|lung /mean(|Z (f , t)|bronchi ) (1) Fig. 3 shows the mean (continuous line) and SD (dashed PAcalibrated (f , t) lines) values of |Z|, PA, R and Xc plotted along the fre- = PAlung (f , t) − mean(PAbronchi (f , t)) (2) quency range (15 kHz – 307 kHz) used for the measures obtained in healthy lung tissue divided in smoker patients F. DATA ANALYSIS (red), non-smokers patients (green) and ex-smoker patients For tissue differentiation analysis among non-smoker, smoker (blue) for non-calibrated (left) bioimpedance measures and and ex-smoker samples in healthy lung tissue samples as well calibrated bioimpedance measures (right) showing an inter- as for tissue differentiation analysis between healthy lung sample reduction of the dispersion and increasing data tissue and neoplasm lung tissue the averaged spectra of the homogeneity. bioimpedance measurements, obtained using the 3-electrode method, throughout the acquisition time was used. B. TISSUE DIFFERENTIATION AMONG NON-SMOKERS, The frequency range chosen to visualize and analyze SMOKERS AND EX-SMOKERS PATIENTS IN CALIBRATED the data was 15 kHz – 307 kHz. The values from fre- AND NON-CALIBRATED DATA quencies higher and lower than this range were discarded Table 1 lists the descriptive parameters, specified as the mean due to electrode effects at low frequency and capacitive ± SD, 95% confidence interval for mean (lower bound and coupling errors at high frequency. For tissue differentia- upper bound) of |Z|, PA, R and Xc and the results of the tion analysis the frequency of 15 kHz for |Z| and R and one-way ANOVA including the Fisher coefficient (F) for the frequency of 307 kHz for PA and Xc were chosen. the minimally-invasive bioimpedance measures performed These frequencies were chosen based on the higher distance in healthy lung tissue (non-smokers: n = 22; smokers: between the means of the groups used to perform the tissue n = 9; ex-smokers: n = 23) for the measures calibrated differentiation. and non-calibrated. No statistically significant differences The normality of the distribution of the variables was (P > 0.05) related to the smoking condition are found determined by the Kolmogorv-Smirnov (healthy lung tissue among the three groups analyzed for both calibrated and non- samples) test and Shapiro-Wilk test (neoplasm lung tissue calibrated samples). The variables normally distributed are shown as the data. mean ± standard deviation (SD) and 95% confidence interval (CI) for the mean (lower bound and upper bound). Non- normally distributed variables are shown as statistic median C. MULTI-FREQUENCY RESPONSE FOR MINIMALLY (interquartile range, IQR) and minimum - maximum. One- INVASIVE HEALTHY LUNG TISSUE AND NEOPLASM LUNG way analysis of variance (ANOVA) was used to determine TISSUE MEASUREMENTS statistically significant differences in the |Z|, PA, R and Xc Fig. 4 shows the mean (continuous line) and SD (dashed values among smokers, non-smokers and ex-smoker samples lines) values of |Z|, PA, R and Xc plotted along the frequency in healthy lung tissue. Repeated measures t-test was used to range (15 kHz – 307 kHz) used for the measures obtained in determine statistically significant differences in the |Z|, PA, healthy lung tissue (green) and neoplasm lung tissue (black) R and Xc values between non-calibrated data and calibrated before (left) and after (right) calibration respectively. Results data among smokers, non-smokers and ex-smoker healthy show an increase in the separation between tissues in |Z|, lung samples. One-way analysis of variance (ANOVA, para- R and Xc, especially the first two. metric data) and Mann–Whitney U test (non-parametric data) was used to determine statistically significant differences in the |Z|, PA, R and Xc values between healthy lung tissue D. TISSUE DIFFERENTIATION BETWEEN HEALTHY LUNG and neoplasm lung tissue. In addition, the area under the TISSUE AND NEOPLASM LUNG TISSUE Receiver Operating Characteristic (ROC) curve was used to Table 2 lists the descriptive parameters, specified as the mean measure the discriminative capacity of the non-calibrated ± SD, 95% confidence interval for mean (lower bound and and calibrated measure of |Z|, PA, R and Xc according to upper bound) for normally distributed variables and specified tissue classification (1: healthy lung tissue; 2: neoplasm lung as statistic median (interquartile range, IQR) and minimum – tissue) by biopsy. Following the ROC analysis area under maximum for non-normally distributed variables of |Z|, PA, curve (AUC) above 0.9 is considered a very good model R and Xc and the results of the one-way ANOVA including and AUC above 0.97 it is considered as excellent. A value the Fisher coefficient (F) and the Mann–Whitney U test less than 0.5 indicates the model is no better than random results including the U statistic (U) for the minimally invasive prediction. bioimpedance measures performed in healthy lung tissue The statistical software IBM R SPSS R version 28.0 (IBM (n = 54) and in neoplasm lung tissue (n = 15) for Corp, Armonk, NY, United States) was used for data analysis. the measures calibrated and non-calibrated. Statistically The level of statistical significance was set at P < 0.05. significant differences (P < 0.001) are found between 103154 VOLUME 10, 2022 G. Company-Se et al.: Effect of Calibration for Tissue Differentiation Between Healthy and Neoplasm Lung FIGURE 3. Results of the non-calibrated (left) and calibrated (right) mean (continuous line) and SD (dashed lines) parameters extracted from the bioimpedance signal along the different frequencies analyzed (15 kHz – 307 kHz). In order, (a) Modulus, (b) Phase angle, (c) Resistance and (d) Reactance of the bioimpedance of all the different measures taken in healthy lung tissue classified according to cigarette consumption. Green: non-smoker; blue: ex-smokers; red: smokers. VOLUME 10, 2022 103155 G. Company-Se et al.: Effect of Calibration for Tissue Differentiation Between Healthy and Neoplasm Lung TABLE 1. Descriptions of bioimpedance measurements performed in healthy lung tissue for non-smokers, smokers and ex-smokers. The variables normally distributed are shown as mean ± SD, 95% confidence interval for mean (lower bound and upper bound). In addition, the Fisher (F) coefficient for variance analysis and the statistical significance (P) are also shown. healthy and neoplasm lung tissue for both calibrated and Lungs are organs that belong to the respiratory system non-calibrated data. whose principal function is to produce gas exchange. Struc- tures from the respiratory system include trachea, bronchi and terminal bronchioles. Each of these structures has its own E. EFFECTS OF CALIBRATION IN DATA VARIABILITY IN anatomical and histological characteristics [19]. Therefore, HEALTHY LUNG TISSUE AND IN NEOPLASM LUNG TISSUE differences in bioimpedance measurements can be expected Fig. 5 shows the effect of calibration in bioimpedance data based not only on the type of tissue but also on its state. variability for healthy lung tissue and neoplasm lung tissue This work reports the use of minimally invasive EIS in respectively for |Z| and R at 15 kHz and for PA and Xc at lungs through a bronchoscopy process using the 3-electrode 307 kHz. Results show a decrease in data dispersion within method to differentiate among smoker, non-smokers and the same tissue group, especially in |Z| and R parameters, ex-smoker healthy lung tissue samples, in order to analyze after the calibration of the bioimpedance data. Fig. 6 shows its potential role in the measurements variability and to dif- the receiver operating characteristic (ROC) curves for |Z|, PA, ferentiate between healthy lung tissue and neoplasm lung R and Xc before and after calibration for healthy lung tissue tissue. Both tissue differentiations are used to evaluate the and neoplasm lung tissue groups. Results show an increase inter-patient variability in the mentioned groups and to eval- of the area under curve (AUC) after the calibration of the uate the utility of calibration using a bioimpedance measure bioimpedance data in |Z|, R and Xc (AUC > 0.96) and a performed in a principal bronchus. This strategy of taking a decrease of the AUC in PA (AUC < 0.95). measure to calibrate the other measures has been previously used in heart applications [18]. IV. DISCUSSION The inflammatory response due to cigarette consumption This project evaluates the need of the calibration of the is not differentiable through bioimpedance EIS measures minimally invasive EIS bioimpedance measures performed neither with the non-calibrated measurements nor with the in lung tissue using a measure performed in bronchial tissue. calibrated measures. Therefore, the initial hypothesis that Moreover, it evaluates the influence of cigarette smoking the smoking condition could be a cause of dispersion in the in healthy lung tissue bioimpedance measures as a possible EIS-derived estimators can be discarded. According to Fig. 3 cause of dispersion. Finally, it differentiates between healthy the |Z| and R show a decrease in their values when calibrating and neoplasm lung tissue and assesses the possible improve- with respect to a bronchi measurement while PA and Xc ment of this differentiation using the calibration. increase their values (nearer to 0 than the non-calibrated 103156 VOLUME 10, 2022 G. Company-Se et al.: Effect of Calibration for Tissue Differentiation Between Healthy and Neoplasm Lung FIGURE 4. The mean (continuous line) and SD (dashed lines) values from the bioimpedance signal along the different frequencies analyzed before and after calibration. The (a) modulus, (b) phase angle, (c) resistance and (d) capacitive reactance. Green: healthy lung tissue; black: neoplasm lung tissue. VOLUME 10, 2022 103157 G. Company-Se et al.: Effect of Calibration for Tissue Differentiation Between Healthy and Neoplasm Lung TABLE 2. Descriptions of minimally-invasive bioimpedance measurements for healthy lung tissue and neo-plasm lung tissue. The variables normally distributed are shown as mean ± SD, 95% confidence interval for mean (lower bound and upper bound) while that non-normally distributed data are shown as statistic median (interquartile range, IQR) and minimum-maximum. In addition, the statistic of the Mann Whitney U test (U), the Fisher (F) coefficient for variance analysis and the statistical significance (P) are also shown. data). The non-calibrated |Z| and R as well as the PA show through bioimpedance analysis. Moreover, results show an slightly although non-significative higher values in those intra-sample dispersion reduction with the effect of calibra- samples in which cigarette consumption is present. How- tion, especially in |Z| and R, which depend on the geometrical ever, Xc present lower values in smoker samples. When factors. we calibrate the bioimpedance measurements we show an Regarding tissue differentiation between healthy lung tis- intra-sample variability reduction. This variability reduction sue and neoplasm lung tissue we have taken all the healthy specially affects |Z| and R due to the geometrical dependence lung tissue samples without considering the tabaco habits as of R and the high correlation between |Z| and R [5], [6]. it has been demonstrated that this factor is not significant Emphasizing the importance of the analysis of R and (P > 0.05). Lung cancer is a highly complex neoplasm and Xc according to the theory of Lukaski et al. [6], [7], comprise several histological types. The groups most fre- Piccoli et al. [20], and Lukaski et al. [21] we selected the quently are the non-small cell lung cancer (NSCLC) such as frequencies (15 kHz and 307 kHz) to check the hypothetical adenocarcinoma and squamous carcinoma, followed by small differentiation among non-smokers, smokers and ex-smokers cell lung cancer (SCLC) [23]. Lung cancer are the results of healthy lung tissue samples following the calculation of the the accumulation of genetic and epigenetic changes, includ- maximum distance between means of the three groups. From ing abnormalities of the inactivation of tumour-suppression the bioimpedance parameters, R describes the behavior of the genes and the activation of oncogenes [24]. For the tissue medium through which the current flows while Xc describes differentiation between healthy lung tissue and neoplasm the capacitive component of the cell membranes. The values lung tissue all cancer types have been included in the same of |Z| and PA are dependent of R and Xc [6], [7]. group so we assume that the remaining dispersion in neo- The significance of the test was determined with the plasm lung tissue might be due to the differences within lung p-value which is the probability of obtaining test results at cancer types. We have selected the frequencies (15 kHz and least as extreme as the result observed, assuming that the 307 kHz) that offered us a better discriminatory response null hypothesis is correct. Therefore, considering the level of between healthy lung tissue and neoplasm by taking the significance set, results will be statistically significant if a frequency with the maximum difference between the mean P < 0.05 is obtained in the test. Regarding tissue dif- of the healthy lung tissue and the mean of the neoplasm ferentiation among non-smokers, smokers and ex-smokers lung tissue. We have also visualized the mean impedance healthy lung tissue samples, one-way ANOVA reported spectrum and SD of the healthy lung tissue samples and the non-significant results (P > 0.05) for all variables (|Z|, PA, neoplasm lung tissue samples between the frequency range R and Xc) for both, the non-calibrated and the calibrated analyzed (15 kHz – 307 kHz) with the data non-calibrated measures. No post-hoc test has been done as no significant and calibrated to show the effects of the calibration. Accord- results have been found. The Fisher coefficient parameter (F) ing to the results obtained in Fig. 4 the calibration of the represents the relationship between the inter-group variance bioimpedance measures with respect to a measure performed and the intra-group variance. Therefore, a higher F coeffi- in bronchi reduces the intra-group variability and, in con- cient indicates a higher inter-group variance than intra-group sequence, increases the inter-patient distance in both, the variance [22]. According to the results obtained in Table 1, healthy lung tissue and the neoplasm lung tissue, especially in the F coefficient obtained in the non-calibrated data is higher |Z| and R, which are the two parameters that are dependent on in |Z|, R and Xc than in the calibrated data. In contrast, geometrical factors (Fig. 5). Results obtained show a higher F coefficient in PA obtained in the calibrated data is higher |Z| and R and a lower PA and Xc in healthy lung tissue than than in the non-calibrated data. Therefore, the statistical in neoplasm lung tissue. Moreover, |Z| and R show higher results obtained show that the effect of cigarette consumption difference between the lower frequencies and the higher fre- should not be considered to perform tissue differentiation quencies in healthy lung tissue than in neoplasm lung tissue. 103158 VOLUME 10, 2022 G. Company-Se et al.: Effect of Calibration for Tissue Differentiation Between Healthy and Neoplasm Lung FIGURE 5. Boxplot of bioimpedance calibrated (C) and non-calibrated (NC) data of healthy lung tissue and neoplasm lung tissue for (a) |Z| and (c) R at 15 kHz and for (b) PA and (d) Xc at 307 kHz. The central mark of each box indicates the median, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The whiskers extend to the most extreme data points that are not considered outliers. In addition, the bioimpedance data values for the calibrated (blue) and non-calibrated (orange) measures. Vertical axis are different for the calibrated and non-calibrated data for better data visualization. VOLUME 10, 2022 103159 G. Company-Se et al.: Effect of Calibration for Tissue Differentiation Between Healthy and Neoplasm Lung FIGURE 6. Receiver operating characteristic (ROC) curves to assess the predictive ability of the different electrical impedance parameters before and after calibration between healthy and neoplasm lung tissue. In (a) the results of the |Z| before calibration (NC) and after calibration (C) at 15 kHz. In (b) the results of the PA before calibration (NC) and after calibration (C) at 307 kHz. In (c) the results of the R before (NC) and after (C) calibration at 15 kHz. In (d) the results of the Xc before (NC) and after (C) calibration at 307 kHz. EIS assumes that current at low frequency flows through the medium, than the neoplasm lung tissue. These two main char- extracellular space while current at high frequencies flows acteristics produce a higher |Z| and R in healthy lung tissue through both, intracellular and extracellular space. Moreover, than in neoplasm lung tissue. Lung cancer produce multi- healthy lung tissue is composed of alveolar epithelial and ples histological changes of the normal bronchial mucosa. endothelial cells separated by a thin basement membrane Proliferation of epithelial cells with abundant cytoplasm and and interstitial space. Interstitial space is a non-conductive vesicular nuclei, intercellular bridging, thickening of alveolar 103160 VOLUME 10, 2022 G. Company-Se et al.: Effect of Calibration for Tissue Differentiation Between Healthy and Neoplasm Lung septa and others pathological changes [25]. The morphologic between healthy lung tissue among smoker, non-smoker and features in neoplasm lung tissue seem to contribute to lose ex-smoker measures, which was initially stated as a possible their capacitive behavior which is translated into a PA and cause of EIS measurement dispersion in lungs. Then, to per- Xc flat mean impedance spectrum, as compared to the healthy form tissue differentiation between healthy lung tissue and lung tissue mean impedance spectrum (Fig. 4). Regarding tis- neoplasm lung tissue the effect of tobacco habit will not be sue differentiation between healthy lung tissue and neoplasm considered. Also, this effect will not be considered in our lung tissue, one-way ANOVA for the calibrated data and future studies. Mann–Whitney U test for the non-calibrated data (Table 2) On the other hand, we found that there is a statistically show statistically significant differences between the two significant difference in both calibrated and non-calibrated groups (healthy and neoplasm lung tissue, P < 0.001). These measurements at 15 kHz (|Z| and R) and 307 kHz (Xc and statistical differences are probably due to the histological PA) between healthy and neoplasm lung tissue. This shows differences between both groups by minimally-invasive EIS that minimally invasive electrical impedance spectroscopy measurements. Focusing only in the calibrated data results measurements using the 3-electrode method are able to dis- in Table 2 show higher significance in |Z| and R than in criminate between healthy lung and neoplasm both with and PA and Xc as the F coefficient is higher in the first two without calibration. parameters. Calibration has, however, been demonstrated to reduce The study has shown that there is an effect on the mea- data variability and increase the tissue state separa- surement when calibrating, reducing the dispersion of the tion capability, which will be useful in future studies measurements (Fig. 5). Calibration doesn’t change the out- when including other pathologies with similar pathological come of the hypothesis test, showing a statistically significant mechanisms. difference in both cases, but the higher-F coefficient (Fisher Moreover, significant differences are found between cal- coefficient from one-way ANOVA test, used for comparing ibrated and non-calibrated paired samples of smoker, non- the factors of the total deviation) than U (statistic from smoker ex-smoker and neoplasm lung tissue showing that Mann–Whitney U test, used to assess whether two sampled calibration is beneficial to reduce intra-sample variability. groups are likely to derive from the same population) suggests The authors recommend calibrating the measures obtained stronger separation between the groups (Table 2), which is with respect to the bronchi given that it is demonstrated highly significant (P < 0.01) for both calibrated and non- that it increases the sensitivity of the 3-electrode mini- calibrated measures. On the other hand, according to the ROC mally invasive electrical impedance spectroscopy for tissue curve analysis, (Fig. 6) we have observed that the area under differentiation. the curve (AUC) is equally excellent in all the variables (AUC > 0.9) both calibrating and not calibrating, although higher AUTHOR CONTRIBUTIONS AUC values are observed when calibrating. After calibrating, Ramon Bragós, Virginia Pajares, Pere J. Riu, Javier Rosell, the AUC is greater than 0.96 for all cases except in PA. Georgina Company-Se, Alfons Torrego, and Lexa Nescolarde The |Z|, R and Xc increase the sensitivity (true positive designed the experiments; Georgina Company-Se, Virginia fraction) and decrease the specificity (false positive fraction) Pajares, and Alfons Torrego performed the experiments; Lexa after calibration. Only PA showed a decrease in sensitivity Nescolarde and Georgina Company-Se performed the data maintaining its specificity (Fig. 6). Considering that PA has processing; Lexa Nescolarde, Georgina Company-Se, and a trigonometric relationship between R and Xc and that these Ramon Bragós analyzed the data; Georgina Company-Se improve with calibration, the authors recommend perform- and Lexa Nescolarde drafted the manuscript and pre- ing the calibration of the measurements with respect to the pared the tables and figures; and Ramon Bragós, Virginia bronchi. Pajares, Pere J. Riu, Javier Rosell, Georgina Company-Se, In the previous study performed by Company-Se et al. [14] Alfons Torrego, and Lexa Nescolarde revised the paper and we performed tissue differentiation between healthy lung approved the final version of the manuscript. tissue and bronchi tissue. We proposed continuing with the study by including neoplasm lung tissue for lung ACKNOWLEDGMENT tissue differentiation. Results obtained in Table 2 show The authors would like to specially thank the patients that minimally invasive electrical impedance spectroscopy without whom this study would not have been possible. using the 3-electrode method is able to discriminate with In addition, they would like to thank Marta Navarro Colom, both, calibrated data (not considering geometrical factors) Laura Romero Roca, and Margarita Castro Jiménez from and with non-calibrated data. In future studies we aim the Interventional Pulmonology Unit, Respiratory Medicine to include other lung pathologies with other histological Department, Hospital de la Santa Creu i Sant Pau for the characteristics. invaluable support. V. CONCLUSION REFERENCES In conclusion, results of the healthy lung tissue bioimpedance [1] I. Annesi-Maesano and G. Viegi, Eds., Respiratory Epidemiology. measurements show that there are no significant differences Lausanne, Switzerland: European Respiratory Society, 2014. VOLUME 10, 2022 103161 G. Company-Se et al.: Effect of Calibration for Tissue Differentiation Between Healthy and Neoplasm Lung [2] H. J. de Koning, ‘‘Reduced lung-cancer mortality with volume CT screen- [21] K. R. Foster and H. C. Lukaski, ‘‘Whole body impedance what does ing in a randomized trial,’’ New England J. Med., vol. 382, no. 6, it measure?’’ Amer. J. Clin. Nutrition, vol. 64, no. 3, pp. 388S–396S, pp. 503–513, Feb. 2020, doi: 10.1056/NEJMoa1911793. Sep. 1996, doi: 10.1093/ajcn/64.3.388S. [3] D. E. Ost, ‘‘Diagnostic yield and complications of bronchoscopy for [22] D. F. Morrison, Multivariate Statistical Methods. Boca Raton, FL, USA: peripheral lung Lesions. Results of the AQuIRE registry,’’ Amer. J. Res- CRC Press, 1967. piratory Crit. Care Med., vol. 193, no. 1, pp. 68–77, Jan. 2016, doi: [23] M. Zheng, ‘‘Classification and pathology of lung cancer,’’ Surgical 10.1164/rccm.201507-1332OC. Oncol. Clinics North Amer., vol. 25, no. 3, pp. 447–468, Jul. 2016, doi: [4] J. S. Wang Memoli, P. J. Nietert, and G. A. Silvestri, ‘‘Meta-analysis of 10.1016/j.soc.2016.02.003. guided bronchoscopy for the evaluation of the pulmonary nodule,’’ Chest, [24] E. A. Engels, ‘‘Inflammation in the development of lung cancer: Epi- vol. 142, no. 2, pp. 385–393, Aug. 2012, doi: 10.1378/chest.11-1764. demiological evidence,’’ Expert Rev. Anticancer Therapy, vol. 8, no. 4, [5] S. Khalil, M. Mohktar, and F. Ibrahim, ‘‘The theory and fundamentals pp. 605–615, Apr. 2008, doi: 10.1586/14737140.8.4.605. of bioimpedance analysis in clinical status monitoring and diagnosis of [25] W. D. Travis, ‘‘International association for the study of lung diseases,’’ Sensors, vol. 14, no. 6, pp. 10895–10928, Jun. 2014, doi: cancer/american thoracic society/European respiratory society 10.3390/s140610895. international multidisciplinary classification of lung adenocarcinoma,’’ J. Thoracic Oncol., vol. 6, no. 2, pp. 244–285, Feb. 2011, doi: [6] H. C. Lukaski, N. V. Diaz, A. Talluri, and L. Nescolarde, ‘‘Classifica- 10.1097/JTO.0b013e318206a221. tion of hydration in clinical conditions: Indirect and direct approaches using bioimpedance,’’ Nutrients, vol. 11, no. 4, p. 809, Apr. 2019, doi: 10.3390/nu11040809. [7] H. C. Lukaski, ‘‘Biological indexes considered in the derivation of the bioelectrical impedance analysis,’’ Amer. J. Clin. Nutrition, vol. 64, no. 3, pp. 397S–404S, Sep. 1996, doi: 10.1093/ajcn/64.3.397S. [8] S. Toso, ‘‘Altered tissue electric properties in lung cancer patients as GEORGINA COMPANY-SE received the B.S. detected by bioelectric impedance vector analysis,’’ Nutrition, vol. 16, degree in biomedical engineering from the Univer- no. 2, pp. 120–124, Feb. 2000, doi: 10.1016/s0899-9007(99)00230-0. sitat Politècnica de Catalunya (UPC), Barcelona, [9] D. M. Nierman, D. I. Eisen, E. D. Fein, E. Hannon, J. I. Mechanick, Spain, in 2018, and the M.S. degree in compu- and E. Benjamin, ‘‘Transthoracic bioimpedance can measure extravascu- tational biomedical engineering from Universitat lar lung water in acute lung injury,’’ J. Surgical Res., vol. 65, no. 2, Pompeu Fabra (UPF), Barcelona, in 2019. She is pp. 101–108, Oct. 1996, doi: 10.1006/jsre.1996.0350. currently pursuing the Ph.D. degree in biomedi- [10] J. Orschulik, N. Hochhausen, S. A. Santos, M. Czaplik, S. Leonhardt, and cal engineering with UPC. Her research interests M. Walter, ‘‘Detection of acute respiratory distress syndrome using sectoral bioimpedance spectroscopy—A pilot study,’’ in Proc. 11th Int. Conf. Bio- include bioimpedance measures and analysis, sig- electromagn., Aachen, Germany, May 2018, pp. 1–10. [Online]. Available: nal processing, and data science. She is interested http://publications.rwthaachen.de/record/723538/files/723538.pdf in machine learning and data science. [11] B. Sanchez, G. Vandersteen, I. Martin, D. Castillo, A. Torrego, P. J. Riu, J. Schoukens, and R. Bragos, ‘‘in vivo electrical bioimpedance characteri- zation of human lung tissue during the bronchoscopy procedure. A feasi- bility study,’’ Med. Eng. Phys., vol. 35, no. 7, pp. 949–957, Jul. 2013, doi: 10.1016/j.medengphy.2012.09.004. [12] N. Coll, R. Bragós, A. M. Muñoz, V. Pajares, A. Torrego, and P. J. Riu. LEXA NESCOLARDE was born in Baracoa, Cuba, (2016). Espectrometría de Impedancia Eléctrica en Tejido Pulmonar. in 1970. She received the Ph.D. degree in biomed- [Online]. Available: https://upcommons.upc.edu/handle/2117/100980 ical engineering from the Universitat Politècnica [13] P. J. Riu, G. Company, R. Bragos, J. Rosell, V. Pajares, and A. Torrego, de Catalunya (UPC), Barcelona, Spain, in 2006, ‘‘Minimally invasive real-time electrical impedance spectroscopy diag- nostic tool for lung parenchyma pathologies,’’ in Proc. 42nd Annu. Int. under supervision of Professor Javier Rosell. Since Conf. IEEE Eng. Med. Biol. Soc. (EMBC), Jul. 2020, pp. 5077–5080, doi: January 2001, she has been a member of the 10.1109/EMBC44109.2020.9175860. Electronic and Biomedical Instrumentation Group [14] G. Company-Se, L. Nescolarde, V. Pajares, A. Torrego, P. J. Riu, (IEB) as well as the Biomedical Research Center J. Rosell, and R. Bragos, ‘‘Minimally invasive lung tissue differentiation (CREB-UPC). She is currently an Associate Pro- using electrical impedance spectroscopy: A comparison of the 3- and fessor at UPC. Since 2002, she has been partic- 4-electrode methods,’’ IEEE Access, vol. 10, pp. 7354–7367, 2022, doi: ipating in 22 research projects and has led three research and technology 10.1109/ACCESS.2021.3139223. transfer contracts of special relevance and two of which are still valid. Her [15] K. Subramaniam, A. R. Clark, E. A. Hoffman, and M. H. Tawhai, ‘‘Met- current research interests include use of non-invasive localized bioimpedance rics of lung tissue heterogeneity depend on BMI but not age,’’ J. Appl. measurement (L-BIA) for muscle assessment in high performance athletes, Physiol., vol. 125, no. 2, pp. 328–339, Aug. 2018, doi: 10.1152/japplphys- body composition analysis, and data analysis. iol.00510.2016. [16] S. Weitzman and L. Gordon, ‘‘Inflammation and cancer: Role of phagocyte-generated oxidants in carcinogenesis,’’ Blood, vol. 76, no. 4, pp. 655–663, Aug. 1990. [17] R. O’Donnell, ‘‘Inflammatory cells in the airways in COPD,’’ Thorax, vol. 61, no. 5, pp. 448–454, May 2006, doi: 10.1136/thx.2004.024463. VIRGINIA PAJARES was born in Girona, in 1979. [18] G. Amorós-Figueras, E. Jorge, C. Alonso-Martin, D. Traver, M. Ballesta, R. Bragós, J. Rosell-Ferrer, and J. Cinca, ‘‘Endocardial infarct scar recog- She received the Ph.D. degree in medicine nition by myocardial electrical impedance is not influenced by changes in and surgery from the Universitat Autónoma de cardiac activation sequence,’’ Heart Rhythm, vol. 15, no. 4, pp. 589–596, Barcelona, in 2015. She has been a Medical Doctor Apr. 2018, doi: 10.1016/j.hrthm.2017.11.031. and a Respiratory Specialist, since 2008. More- [19] J. F. Murray, ‘‘The structure and function of the lung,’’ Int. J. Tuberculosis over, she is a Consultant with the Department of Lung Diseases, Off. J. Int. Union Tuberculosis Lung Diseases, vol. 14, Respiratory, Hospital de la Santa Creu i Sant Pau, no. 4, pp. 391–396, Apr. 2010. Barcelona. She is part of the Bronchosocopy Unit [20] A. Piccoli, B. Rossi, L. Pillon, and G. Bucciante, ‘‘A new method Staff and also she works as a Respiratory Resi- for monitoring body fluid variation by bioimpedance analysis: The dents’ Mentor. On 2021, she is author of 20 scien- RXc graph,’’ Kidney Int., vol. 46, no. 2, pp. 534–539, Aug. 1994, doi: tific articles in PubMed. Her main research interests include interventional 10.1038/ki.1994.305. pulmonology, lung cancer, and pleural diseases. 103162 VOLUME 10, 2022 G. Company-Se et al.: Effect of Calibration for Tissue Differentiation Between Healthy and Neoplasm Lung ALFONS TORREGO was born in Barcelona, JAVIER ROSELL (Senior Member, IEEE) Spain, in 1971. He has been a Medical Doctor was born in Barcelona, Spain, in June 1959. and a Respiratory Specialist, since 2000, and a He received the Ingeniero de Telecomunicación Consultant with the Department of Respiratory, and Doctor Ingeniero de Telecomunicación Hospital de la Santa Creu i Sant Pau, Barcelona. degrees from the Polytechnic University of Cat- He is also the Bronchosocopy Unit Coordinator at alonia (UPC), Barcelona, in 1983 and 1989, the Hospital de la Santa Creu i Sant Pau. Moreover, respectively. he is an Associated Professor of medicine at the He is currently a Full Professor with the Depart- Universitat Autònoma de Barcelona and a Former ment of Electronic Engineering, UPC, and the Chairperson of the Spanish Respiratory Society Head of the Research Group, Biomedical Research Scientific Committee. On 2021, he is author of more than 60 scientific Center (CREB-UPC). His current research interests include non-invasive and articles in PubMed. His main research interests include interventional pul- non-obtrusive measurement methods in sports, medical and biological fields, monology, lung cancer, and severe asthma. particularly based on bio-electrical impedance spectroscopy and magnetic induction spectroscopy. PERE J. RIU (Senior Member, IEEE) received the M.Sc. degree in telecommunication engineer- ing and the Ph.D. degree in electronic engineer- ing from the Universitat Politècnica de Catalunya (UPC), Barcelona, Spain, in 1986 and 1991, RAMON BRAGÓS received the degree in elec- respectively. He was a Visiting Associate Profes- trical engineering (major in telecommunications sor with the Department of Bioengineering, Uni- engineering) and the Ph.D. degree in electronic versity of Pennsylvania, Philadelphia, PA, USA, engineering from the Technical University of in 1997. He is currently a Full Professor of elec- Catalonia (UPC), in 1991 and 1997, respectively. tronics with the Department of Electronic Engi- Since 1998, he has been an Associate Profes- neering, UPC. His research interests include electromagnetic compatibility, sor at the Department of Electronic Engineering, computational electromagnetics, interaction of EMF with biological tis- UPC. He belongs to the Electronic and Biomedical sues, and biomedical instrumentation design with emphasis on electrical Instrumentation Research Group and the Center bioimpedance techniques, including EIT. These activities are performed for Research in Biomedical Engineering (CREB). within the Centre for Research on Biomedical Engineering (CREB, UPC) His main research interests include design of methods and systems for and the Institut de Recerca Sant Joan de Déu (IRSJD). He is a member of the characterization of biological materials and systems using minimally the Committee on Man and Radiation (COMAR) (IEEE-BMES) and the invasive methods, mainly electrical impedance spectroscopy. Vice-President of the International Society for Electrical Bio-Impedance. VOLUME 10, 2022 103163