Lung Quantitative Ultrasound to Stage and Monitor Interstitial Lung Diseases

Chronic interstitial lung diseases (ILDs) require frequent point-of-care monitoring. X-ray-based methods lack resolution and are ionizing. Chest computerized tomographic (CT) scans are expensive and provide more radiation. Conventional ultrasound can detect severe lung damage via vertical artifacts (B-lines). However, this information is not quantitative, and the appearance of B-lines is operator- and system-dependent. Here we demonstrate novel ultrasound-based biomarkers to assess severity of ILDs. Lung alveoli scatter ultrasound waves, leading to a complex acoustic signature, which is affected by changes in alveolar density due to ILDs. We exploit ultrasound scattering in the lung and combine Quantitative Ultrasound (QUS) parameters, to develop ultrasound-based biomarkers that significantly correlate to the severity of pulmonary fibrosis and edema in rodent lungs. These innovative QUS biomarkers will be very significant for monitoring severity of chronic ILDs and response to treatment, especially in this new era of miniaturized and highly portable ultrasound devices.


Introduction
Lung ultrasound (LUS) remains challenging because of multiple scattering of ultrasound waves by airlled alveoli that prevents creating an image.Conventional ultrasound imaging is based on the assumption of a known speed of sound in tissue.In highly heterogeneous media such as lung tissue, ultrasound waves encounter numerous re ections and scattering events, leading to complex propagation paths.This prevents obtaining images that accurately represent lung structures using conventional ultrasound.
Although conventional LUS imaging does not provide detailed structural information about the parenchyma, physicians have used it to detect signs of potential lung damage by observing artifacts in LUS images.Efforts have been made to standardize the interpretation of these artifacts, such as vertical and horizontal artifacts known as B-lines and A-lines, respectively.For example, B-lines in LUS images have been utilized to detect the presence of pulmonary brosis (1).The appearance of these artifacts has also been shown to correlate with increased extravascular lung water (2), interstitial lung diseases (ILDs) (2-4), cardiogenic and non-cardiogenic lung edema (5), interstitial pneumonia(6), and lung contusion (7).
The physical origin of these artifacts is commonly attributed to ultrasound beam reverberations in acoustical traps made of poorly aerated regions(8).
However, interpreting these artifacts is highly subjective since they are qualitative and dependent on imaging parameters such as transmit ultrasound frequency (9).Consequently, the effectiveness of diagnosing and monitoring ILDs using conventional LUS is not possible.There is a clear need for a quantitative approach to assess lung parenchymal microstructure.This would enable monitoring severity of some ILDs, especially in point-of-care settings.The rationale for the present study is to leverage this purported weakness of LUS: we exploit scattering as a source of contrast for characterizing lung tissue.
Our novel hypothesis is that ILD-induced changes in alveolar structure related to interstitial lung diseases will lead to altered scattering patterns and acoustic signatures that can be detected and quanti ed (10).
We quantify and characterize this acoustic signature by using four families of quantitative ultrasound (QUS) parameters, related to i) the relative contributions of multiple and single scattering (MS and SS), ii) the backscatter coe cient (BSC), iii) envelope statistics (EnvS) and iv) attenuation (Att).The output of these methods (i.e., the lung QUS (LQUS) parameters) is powerful to quantify severity of pulmonary edema and brosis because each of these LQUS parameters is a function of the scattering properties of lung tissue that is related to lung microstructure.Pulmonary diseases all affect the parenchymal microstructure in unique ways.Since alveoli act as scatterers, LQUS parameters will re ect parenchymal changes such as alveolar density and size, alveolar ooding, alveolar wall thickness, or spatial heterogeneity.Measuring a variety of LQUS parameters is critical to ensure speci city of the method, because we expect different combinations of these LQUS parameters to be sensitive to different ILDs.
This study shows that different combinations of LQUS parameters enable creation of two different, easyto-interpret LQUS scores that are respectively sensitive and speci c to pulmonary brosis and pulmonary edema.

Raw ultrasound data in rodent models of lung edema and brosis
We used a surgical model of ischemia-reperfusion injury to create various degrees of pulmonary edema in 18 Sprague-Dawley rats.We used a pharmaceutical model of bleomycin inhalation to create various degrees of pulmonary brosis in 24 Sprague Dawley rats.Among the brotic rats, six were administered Nintedanib, to demonstrate the impact of treatment.Twelve healthy untreated animals were used as controls (six for edema and six for brosis).Raw ultrasound data were acquired using a fully programmable ultrasound research scanner (Verasonics, Kirkland, WA).A 128 element linear array transducer (L11-5v, Verasonics), with a central frequency of 7.8 MHz was placed directly on the lung surface using ultrasound coupling gel.After ultrasound data acquisition, the animals were euthanized, and the lungs were excised.Pulmonary edema severity was quanti ed by the wet/dry weight ratio (W/D) (11).Pulmonary brosis severity was quanti ed histologically by a board-certi ed veterinary pathologist using the modi ed Ashcroft score (12,13).

Quantitative ultrasound biomarkers of pulmonary edema and brosis
Raw ultrasound data were processed in order to measure LQUS parameters from four families of methods; Table 1 summarizes all the LQUS parameters that were obtained from a single ultrasound data set.
Multiple LQUS parameters showed signi cant differences between the edematous group and the control group, between the brotic group and the control group, and between the brotic and the edematous group (Fig. 1).For example, the Intercept of the BSC Intercept was on average equal to 40.9 +/-10.5 dB/sr/m in the control group and 64.0 +/-4.8 dB/sr/m in the edematous group.The Intercept of the BSC Intercept showed signi cant differences between these two groups (p = 3.4*10 − 4 ) (Fig. 1a).As another example of a measured LQUS parameter, the Scattering Mean Free Path (SMFP) represents the average distance between scatterers.It is expected to increase with brosis, due to increasing inter-alveolar distances.In the control group, the SMFP was found to be 0.60 +/-0.09mm.In the brotic group, the SMFP was 1.43 +/-0.24mm.In the brotic group treated with Nintedanib, the SMFP was 0.94 +/-0.08 mm, falling between the control and the brotic group.ANOVA demonstrated differences in SMFP between the control, brotic and treated groups (p = 1.3*10 − 6 ) as well as between the brosis and Nintedanib-treated groups (p = 1.9*10 − 2 ) (Fig. 1b).Multiple parameters allowed discrimination between the brosis and edema groups, including the BSC Spectral Slope, as shown in Fig. 1c.BSC Spectral Slope

Speci city to edema and brosis
Measuring a large variety of LQUS parameters provides a detailed description of ultrasound scattering and of the lung microstructure.This large LQUS parameter space has the potential to provide sensitivity to various ILDs.From Table 2 and Fig. 2, it is encouraging to note that the most signi cantly contributing LQUS parameters are not the same for edema and brosis, suggesting that using speci c combinations of LQUS parameters will ensure the speci city of each combination to different ILDs.As an example, Fig. 3 shows that in a parameter space comprised of SMFP (from the MS/SS family), the Midband Fit Intercept and the Spectral Slope (from the BSC family), control, brotic, and edematous lungs can be discriminated.

Discussion
The accumulation of uid in the alveoli and/or interstitium of edematous lungs, or the modi cation of the material properties of the alveolar walls in brotic lungs affect lung function by reducing volume of air within the alveoli.This reduction in air volume also alters the distribution of ultrasound scatterers throughout lung tissue, leading to an increased average distance between scatterers during propagation in lung tissue.It is through elegantly quantifying this phenomenon that LQUS parameters show promising potential as biomarkers for assessing pulmonary edema and brosis, allowing for the quanti cation of scatterer size and distribution within the lungs.
Although not yet tested, this may be true for other forms of ILDs.
Individually, these LQUS parameters moderately but signi cantly (p < 0.05) correlated with the severity of edema or brosis with R values ranging from 0.50 to 0.61.When combined using a stepwise multilinear approach, large and highly signi cant correlations were obtained with the degree of severity of brosis (measured by histology), or edema (measured by W/D), because these LQUS parameters each re ect different features of the microstructure of lung tissue.The strength of the proposed method relies in the diversity of LQUS parameters measured, and the diversity of microstructural changes they quantify.For example, the BSC Midband Fit re ects the relative acoustic impedance between scatterers and the surrounding matrix, which in lungs will be sensitive to alveolar ooding.This is likely why this parameter is signi cantly different for brotic and edematous lungs, and signi cantly correlated to the severity of edema.As another example, changes in scatterer size leads to signi cant changes in the BSC Midband Fit, as well as in the BSC Midband Fit Spectral Slope.Larger alveolar wall thickness observed with brosis leads to larger distances between smaller air-lled alveoli on average.This leads to the larger SMFP values observed in brotic lungs.It can be expected that this diversity of parameters will enable LQUS methods to be sensitive and speci c to various ILDs.
The analysis of ultrasound scattering in lung tissue holds signi cant potential as a routine monitoring modality, with distinct advantages over existing techniques for quantifying the severity of ILDs, such as CT scans.Combining LQUS parameters allows creation of quantitative biomarkers for lung brosis and edema, and potentially other ILDs.Utilizing LQUS to assess lung condition and follow response to treatment could revolutionize point-of-care lung evaluation.Now that there are effective therapies to reduce progression of brosis for idiopathic pulmonary brosis, and alterations in pulmonary edema due to treatments for heart failure, non-invasive point-of-care assessment based on interpretation of ultrasound data may become a useful clinical tool.Adopting a multi-parameter approach is particularly crucial for assessing lung health, as it allows for the differentiation of various pulmonary diseases and enhances the speci city of the quantitative method.The proposed LQUS approach will enable us to tailor the diagnosis and monitoring of many pulmonary diseases.Because the output of LQUS methods is quantitative measures of ultrasound-based predictors that will be easy to interpret by clinicians, adoption should be straightforward even in fast-paced clinical environments providing care for ILD patients or patients with heart failure.
Combining different scattering based, BSC and EnvS LQUS parameters can improve detection and staging of pulmonary brosis and edema with high sensitivity and speci city.Such LQUS predictors will enable a convenient, non-ionizing method for diagnosis of pulmonary brosis, as well as provide a routine method to monitor severity of pulmonary brosis (indicating a potential need for performing a chest CT scan) and assessing severity of heart failure by quantifying severity of pulmonary edema.

Materials and Methods
Surgical model of pulmonary edema: Rats were anesthetized with an intraperitoneal (ip) injection of ketamine (100 mg/kg) and xylazine (10 mg/kg), and intubated via tracheotomy through a cervical incision.Rats were ventilated with FiO2 1.0, tidal volume (TV) of 0.75 ml/100 gm body weight, rate 70/min, 3 cm H2O PEEP.Anesthesia was maintained with iso urane (0.5-4%).Anesthesia depth was judged by toe pinch every 30 min.To maintain hydration, 1-2 cc sterile saline was injected subcutaneously (sc) each hour based on skin turgor and disappearance of the sc uid.Animal temperature was monitored with a rectal probe and maintained at 37 o C on a heating pad intermittently turned on and off.
Pulmonary edema was induced in the left lung of rats using an established lung ischemia-reperfusion injury (IRI) model (1).The anesthetized rat was placed in the right lateral decubitus position, and a left lateral thoracotomy was performed in the 4th intercostal space.The left pulmonary hilum was dissected free.A small laparotomy incision exposed the liver.600 U of heparin was injected intrahepatically.After 5 minutes, the left lung hilum was clamped with an aneurysm clip to render the left lung ischemic.The rat was placed in the supine position.To induce varying degrees of pulmonary edema, 3 different durations (20, 40, and 60 minutes) of lung ischemia were used before the clip was removed and the left lung reperfused for 60 minutes.

Model of pulmonary brosis:
After sedation (ketamine/xylazine as described above), rats were intubated with a 12-gauge catheter.Bleomycin 2mg/kg, dissolved in 100 µl sterile phosphate buffered saline (PBS), was administered into the trachea(2).The rat was extubated and allowed to recover.This bleomycin dose causes development of pulmonary brosis within 2-3 weeks.Rats were studied in groups of n = 6 (3 male, 3 female) 2, 3, and 4 weeks after bleomycin administration.This provided a range of severity of brosis for assessment.Six rats (3 male, 3 female) who received no treatment served as controls.Six more rats (3 males and 3 females) were treated with Nintedanib (10mg/kg) administered daily by gavage after intra-tracheal bleomycin administration for 14 days.Nintedanib is an inhibitor of tyrosine kinase and is used in humans with idiopathic pulmonary brosis to delay progression of the disease(3).Animal care was in accordance with institutional guidelines at the University of North Carolina at Chapel Hill, and all experiments were performed under Institutional Animal Care and Use Committee (IACUC) number 23-029.0The UNC IACUC is under the oversight of the UNC Vice-Chancellor for Research.
Surgery for ultrasound measurements, and euthanasia: After sedation (ketamine/xylazine as described above), a tracheotomy was performed.Rats were ventilated with a Harvard rodent ventilator.Anesthesia was maintained with titrated iso urane (0.5-4%).
A sternotomy was performed, both pleural spaces were opened, and the sternal edges spread maximally to expose both lungs.Hydration was maintained by subcutaneous administration of normal saline, 1-2 ml/hr, based on skin turgor and uid absorption.All rats were sacri ced by cardiectomy under anesthesia (a method of exsanguination under anesthesia, approved by the American Veterinary Medical Association).The incision was extended inferiorly into the abdomen to expose the liver.Heparin was administered intra-hepatically 5 minutes before sacri ce by incising the left and right atria to prevent clotting after the lungs are removed.Wet to dry ratio: Pulmonary edema was quanti ed by the wet/dry weight ratio (W/D).W/D was measured by excising the apical portion of the left lung, while the lung was in the chest with the rat in the supine position.This provided a combination of dependent (posterior) and non-dependent (anterior) lung tissue.Samples were placed on a previously weighed piece of aluminum foil, and each sample was weighed, then dried in an oven at 60ºC for 48 hours, and weighed again.W/D is the ratio of the weights of the lung sample before and after desiccation and is an accepted method of assessing the amount of water in a lung sample (17,18).

Histology:
Following the acquisition of ultrasound data, lung tissues were excised with the airway clamped at endinspiration, and underwent in ation xation by in ating them to the volume they appeared to be at when excised, then instilling 20 ml 10% paraformaldehyde through the main pulmonary artery from a height of 15 cm, followed by immersion in 10% paraformaldehyde for 24-48 hours, Lung blocks were then placed in 70% ethanol.Subsequently, the lung blocks were embedded in para n and each lung was sectioned in the middle of the lung block from the apex to the base of the lung and 5 µm slices were obtained and mounted on glass slides.Sections were stained with hematoxylin and eosin (H&E), Sirius red special, and Masson's Trichrome.Each lung was examined under a microscope at 100x total magni cation.For each lung, 10 different areas were selected for microscopic assessment, so 20 elds were averaged for each lung to create a score.
Severity of brosis was evaluated by assigning scores based on the modi ed Ashcroft scale(6, 7) by a veterinary pathologist (SM).A score of zero denoted completely healthy lung tissue (within normal limits), while a score of 8 indicated lung tissue eld entirely affected by brosis.It should be noted that the maximum brosis score using the Ashcroft scale observed in this study was 4. Scores from 20 elds were then averaged to derive the nal histology score for each lung (Figure .4).
Histological examinations were carried out on both brotic and control lungs.Importantly, control lungs could be assigned a non-zero score (up to 0.25 on the modi ed Ashcroft brosis scale) due to a slight excess of normal collagen in some lung sections.

BSC-based LQUS parameter estimation:
The mean magnitude squared of the Fourier transform of every US segment (i.e., transmit-receive pair over the windowed time) within each Region of Interest (ROI) was computed and log-compressed, then compensated for attenuation, to derive an attenuation-compensated ROI spectrum S ROI (f).The BSC spectral slope, intercept(I0), and midband t (MF) are computed in overlapping, rectangular time gates in each IRM, using the reference phantom method (13,15).The average and standard deviation of each parameter, as well as the slope and intercept of a linear function t to each parameter over propagation time, were obtained (e.g., MF average, MF standard deviation, MF slope, and MF intercept were derived for MF).
EnvS parameters based on the Nakagami and Homodyned-K distribution are computed in overlapping, rectangular time gates in each IRM.The envelope of every US segment in each ROI is attenuation compensated in the spectral domain.The Nakagami probability density function (PDF) was t to the PDF of the ROI using a maximum likelihood estimator, yielding two LQUS parameters: the Nakagami shape parameter, m, and the Nakagami scale parameter Ω.The Homodyned-K PDF was t to the PDF of the ROI using the standard XU estimator, which employs the rst moment of envelope intensity and two logmoments to yield two LQUS parameters (Κ and β, Table 1) (19).
Scattering Mean Free Path (SMFP): SMFP represents the average distance between scatterers.It is expected to increase with edema due to increasing inter-alveoli distances.SMFP is obtained by extracting the incoherent intensity from the IRM and tracking the rate of growth of the incoherent halo over time (9).The SMFP calculation method was improved over the method presented in (20), by adding a region of interest selection step, to ensure only lung tissue is taken into account on the measurement.This is especially critical in rat lungs, which can be smaller than the footprint of the ultrasound probe.
Singular Value Distributions measures (E and λ max ): E and λ max are parameters related to the shape of the singular value distribution of the IRM.According to RMT, the distribution of singular values will resemble a quarter circle if MS dominates, and a Hankel function if single scattering (SS) dominates( 21).E, the expected value, and λ max , the value corresponding to the highest density of singular values relate to the shape of the singular value distribution and are therefore used to quantify the amount of MS in lung tissue.Because unhealthy lungs will create less multiple scattering, E is expected to be larger for unhealthy lungs, and λ max is expected to be smaller (Table 1).

Separation of SS and MS contributions:
The BSC and Env QUS methods described above are based on the assumption of weak and single scattering.In order to improve the sensitivity and speci city to lung pathologies of LQUS parameters, LQUS parameter estimation described above was performed on i) the raw original IRM, ii) only the MS part of the IRM (MS-IRM), and iii) only the SS part of the IRM (SS-IRM).MS-IRM and SS-IRM are obtained using singular value decomposition based on RMT and on the fact that SS is responsible for a deterministic coherence along the anti-diagonal of the IRM.SVD is performed for each time window and the eigenvalues are obtained.An eigenvalue threshold is automatically determined according to (22)and gives the rank of separation.According to RMT, the stronger eigenvalues correspond to the SS contribution and the weaker eigenvalues correspond to the MS contribution.Once the SS and MS contributions are separated, each LQUS parameter was obtained separately from the native IRM, the SS-IRM, and the MS-IRM (Table 1).This very comprehensive framework allows description of the microstructure of the lung tissues quantitatively and in great detail.
Single Intensity halo decay (i.e., SS_IDR): MS by alveoli should yield tremendous acoustic losses as a function of US propagation time.
was on average equal to 0.35 +/-0.23 dB/sr/m/MHz in the edema group, and equal to -3.73 +/-0.56 dB/sr/m/MHz in the brosis group.Through an unpaired t-test, BSC Intercept was found signi cantly different in the edema and brosis groups (p < 10 − 15 ).Table 2: Left: Most signi cant correlations between single LQUS parameters and severity of edema as measured by W/D.Right: Most signi cant correlations between single LQUS parameters and severity of brosis as measured by histology.Pearson correlations between each LQUS parameter and the brosis histology score or the W/D were computed.Overall, 29 LQUS parameters were signi cantly correlated (i.e., p < 0.05) to the W/D ratio in the edema study, and 23 LQUS were signi cantly correlated to the histology score in the brosis study.The LQUS parameters yielding the four strongest correlations are listed in Table2.A stepwise multilinear regression including all LQUS parameters demonstrates that combining three (for brosis) or two (for edema) LQUS parameters leads to ultrasound-based predictors of edema and brosis, with excellent correlations to the severity of edema (R = 0.71, p = 1*10 − 4 ) or brosis (R = 0.84, p = 3*10 − 7 ), as shown in Fig.2.The ultrasound-based predictor of pulmonary edema was obtained by combining the BSC Midband Fit Slope (BSC MF_Slope) and the Nakagami shape parameter (Nakagami m).The ultrasound-based predictor of pulmonary brosis was obtained by combining the Single Scattering intensity decay rate (SS_IDR), the BSC Spectral Slope (BSC SS) and the BSC Midband Fit Intercept (BSC MF_I 0 ).

Figure 4 Examples
Figure 4

Figure 5 Single
Figure 5