Quantitative CT analysis is one area in the rapidly growing field of radiomics, defined as the practice of extracting, analyzing, and interpreting quantitative data from medical images to aid in disease diagnosis and prognosis.1 The use of quantitative CT methods has been shown to be successful in the detection, staging, and management of diffuse lung diseases in human patients. Canine bronchomalacia (BM), defined as the dynamic collapse of segmental and subsegmental airways with clinical signs of airflow limitation, cannot be diagnosed with thoracic radiography, the most used imaging tool in veterinary practice. It is an important cause of morbidity and mortality in canine patients, highlighting the importance of correct diagnosis. Bronchoscopy is the criterion standard for diagnosis of BM2,3 but may underestimate the severity or miss dynamic collapse in anesthetized patients with shallow breathing or apnea and fails to visualize subsegmental airways of higher generations.4 Paired inspiratory and expiratory breath-hold CT scans allow global assessment of dynamic changes in airway calibers and shape and may reflect downstream effects of airflow limitation on the pulmonary parenchyma, providing supportive subjective evidence of BM.5,6 Disparity between identification of BM on bronchoscopy and inspiratory and expiratory breath-hold CT scans has recently been described, suggesting the need for better means to identify BM.7 Quantitative CT analysis providing objective data may play an important role improving disease characterization, rather than solely relying on subjective methods, where visual assessments can be subject to inter- and intraobserver variation and may also be beneficial for cases with subtle changes undetected using subjective methods.
Computed tomographic attenuation, measured in HU, is described by a scale with values expressed relative to attenuation of water (0 HU), ranging from − 1,000 HU (air) to + 3,000 HU (mineral).8 Lung attenuation is negative, varying by species, respiratory phase, and underlying pathology; it is determined by amounts of air, soft tissue, and blood in each volume element known as a voxel. The relative frequency of lung voxels can be calculated and displayed as a histogram showing the distribution of HU values. Computed tomography attenuation is decreased (more negative) with relatively more air as in air trapping or cystic lung disease. It is increased (more positive) with more soft tissue density and less air (eg, infiltrative lung disease, fibrosis, atelectasis). Any condition increasing soft tissue attenuation and decreasing air will increase mean lung attenuation (MLA), increase percent high lung-density volume, blunt degree of leftward skewness (extent of histogram asymmetry), and decrease kurtosis (sharpness of the histogram peak).1,9 Subjectively, assuming the absence of air trapping, it is postulated that increased pulmonary parenchymal attenuation can occur with loss of aeration due to airway collapse. Most quantitative tools use volumetric datasets, and various types of information can be extracted from the data to quantify lung disease. Threshold measurements generally count voxels above or below a certain attenuation value and can be utilized to obtain objective metrics to evaluate parenchymal changes. The use of attenuation histograms allows for the visualization of the distribution in HU, while subjectively the pulmonary parenchyma may appear more hypo- or hyperattenuating, quantifying metrics of attenuation are anticipated to be more sensitive, especially with subtle changes hard to detect with the naked eye, reliable, and reproducible. The purpose of this study is to employ semiautomated objective metrics of CT attenuation histograms to discriminate dogs with BM and without BM (NoBM) using paired inspiratory and expiratory breath-hold CT.
We hypothesized that inspiratory and expiratory breath-hold CT scans with generated CT lung attenuation histograms would demonstrate significant differences in objective metrics reflecting changes in airway caliber on downstream parenchymal aeration in BM and NoBM dogs. Semiautomated software analysis (3D Slicer; Brigham and Women's Hospital) of inspiratory and expiratory breath-hold CT served to determine MLA, lung volume, percent low-attenuation area (LAA) at −856 HU (percentage of voxels more negative than −856 HU), percent high-attenuation area (HAA) at −700 HU (percentage of voxels more positive than −700 HU), percent attenuation area (AA) between −600 and −250 HU (percentage of voxels between −600 and −250 HU), skewness, and kurtosis. We predicted BM dogs would have an increased inspiratory and expiratory MLA, decreased percent LAA at −856 HU, increased percent HAA at −700 HU and percent AA between −600 and −250 HU, and decreased kurtosis and skewness versus NoBM dogs.
Methods
Animals and selection criteria
Medical records for dogs presenting to the Veterinary Health Center, University of Missouri, between January 1, 2014, and July 15, 2022, with a paired inspiratory and expiratory breath-hold CT available in a high–spatial-frequency algorithm, were retrospectively reviewed. Dogs met inclusion criteria if they had a diagnosis of BM based on paired inspiratory and expiratory breath-hold CT, bronchoscopic examination, or both. A CT diagnosis of BM relied on the identification of changes in segmental and subsegmental bronchial caliber and shape on the expiratory versus inspiratory phase including airway wall flattening, distorted circular appearance, or near disappearance of airway lumen.6,10 Subjective parenchymal changes on the expiratory phase of paired inspiratory and expiratory breath-hold CT were not used to make a diagnosis of BM. A bronchoscopic diagnosis of BM relied on regional to diffuse segmental and subsegmental reduction in airway caliber of ≥ 25%.10 Control dogs (NoBM) were selected based on the absence of a diagnosis of BM as well as the absence of lesions encompassing > 5% of the lung on the inspiratory series (transverse images). Dogs having a minimal amount of gravity-dependent atelectasis were not excluded from the study. NoBM dogs were selected from a bank of 46 dogs with negligible observed parenchymal, airway, or vascular disease on paired inspiratory and expiratory breath-hold CT studies. Of those, 20 were similar with respect to age, body weight, and body condition score (BCS) and were selected as controls.
Clinical information gathered from the medical record included sex, neuter status, age, breed, cephalic index (brachycephalic, mesocephalic, or dolichocephalic), and BCS. Final respiratory diagnoses were assigned after review by 2 internists with a special interest in respiratory disease (CR and AVP) after a comprehensive assessment of the entire clinical picture and diagnostic results. Criteria for diagnosis, including confidence in the diagnosis (ie, definitive or suspect), were based on previously published definitions.4,11
Procedures and data acquisition
General anesthesia protocols were individually tailored under the guidance of a board-certified anesthesiologist for thoracic CT and tracheobronchoscopy. Thoracic CT images were performed using a standardized protocol of inspiratory and expiratory breath-hold CTs with the assistance of a mechanical ventilator (Engstrom Carestation ventilator; GE Healthcare). The volume-controlled ventilation setting was used with the tidal volume set to 10 mL/kg, inspired oxygen at 40%, and positive end-expiratory pressure at 5 cm H2O. The positive end-expiratory pressure was set to 0 cm H2O for the expiratory scan. Between October 2020 and July 2022, CT images were acquired with an Aquilion One Genesis CT scanner (Canon Medical Systems). Scan parameters included 0.5-mm slice thickness, 120 kVp, and 300 to 400 mA. Before October 2020, CT images were acquired with an Aquilion 64, a 64-slice CT scanner (Canon Medical Systems). Scan parameters included 0.5-mm slice thickness, 120 kVp, and 170 to 500 mA. The CT images were reconstructed to 0.5- to 3-mm slice thickness, using the thinnest whenever 2 different slice thicknesses were available. All CT images were reconstructed with a high–spatial-frequency algorithm. Direct visualization of the pressure graph displayed on the critical care ventilator monitor was utilized to facilitate timing between CT acquisition and ventilator-assisted breath holds, which provided for an apneic state.7 The inspiratory images were acquired by using the plateau phase of inspiration, and the expiratory images were acquired during a short breath-hold pause (5 to 12 s) induced at the end of the expiration.7 Tracheobronchoscopy was performed using flexible pediatric video bronchoscopes (Fujinon EB-530 H and EG-530NP).
To obtain objective measures for evaluation of parenchymal changes, paired inspiratory-expiratory breath-hold CT images were analyzed by use of a commercially available computer program 3D Slicer software (version 5.0.3; https://www.slicer.org). The Chest Imaging Platform was utilized to analyze data uploaded from DICOM beginning with the Lung CT Segmenter. The right and left lungs were selected as regions of interest for analysis with the exclusion of the trachea, esophagus, and stomach. Three points were placed in transverse and dorsal planes inside the right and left lung fields followed by one point in the trachea (Figure 1). Additional points to exclude the stomach or esophagus were used as needed. Density thresholds expressed in HU were generated utilizing the Parenchymal Analysis extension module. The following metrics were obtained based on studies in humans: MLA, lung volume (in mL), percent LAA at −856 HU (used in human quantitative densitometry as a surrogate marker of air trapping), percent HAA at −700 HU, percent AA between −600 and −250 HU (a surrogate marker of interstitial lung disease), kurtosis (sharpness of the peak of a histogram of attenuation distribution of the lungs, which should normally be sharp to reflect parenchymal homogeneity), and skewness (measure of asymmetry and deviation from the normal distribution of histogram values).1,12–14 Percent change in lung attenuation between inspiration and expiration for each dog was calculated using the equation: (lung volumeI – lung volumeE)/lung volumeI X 100.
Lung segmentation in a dog without bronchomalacia (BM) using the Lung Segmenter module of 3D Slicer. Inspiratory and expiratory (the latter not shown in this example) CT images were uploaded to 3D Slicer. A and B—Using the Lung Segmenter module, regions of interest (arrows) were placed in the right (R-1 to R-6; red dots) and left (L-1 to L-6; blue dots) lungs on transverse (A) and dorsal (B) images. A region of interest was also placed in the trachea (Tr; T-1, teal dot) and stomach (S) for exclusion. C—The area used for analysis is colored green and yellow and rendered in a 3-D model.
Citation: American Journal of Veterinary Research 86, 1; 10.2460/ajvr.24.07.0186
Data analysis and statistics
A regression analysis was performed for each metric with the independent variables being BM, breath phase, and the interaction effect for BM and phase. Statistically significant (P < .05) interactions were followed by pairwise comparisons. The CT acquisition slice thickness was used as a covariate. The MLA metric is strictly negative with a strongly skewed distribution. Median regression15 was used for the analysis of MLA. The decision to model the median rather than the mean was motivated by the fact that for MLA the frequency distribution of MLA is heavily skewed making the mean a poor choice of summary measure. Median regression does not assume normality, is robust to outliers, and obviates the need for transformations. The relative area measures expressed as percentages were rescaled to proportions and analyzed using beta regression.16,17 Each dog was imaged under inspiratory and expiratory breath phases creating a repeated measures layout. To accommodate this, all regression models employed cluster-robust SEs that allow for intrasubject correlation.18 Regression analyses were performed using Stata version 18 (StataCorp) Beta regression made use of the Stata betareg module, and median regression used qreg2.19 As historical data were used, CT images were acquired at 4 slice thicknesses. Slice thickness was treated as a covariate in each model and retained when statistically significant. The beta model is a generalized linear model parameterized in terms of the mean and a scale factor. For those models, CT slice thickness was considered as a covariate for both the mean and scale factor.
Results
Four hundred thirteen dogs had inspiratory-expiratory breath-hold CT over the study time frame; of these, 123 dogs met the inclusion criteria for a diagnosis of BM. Twenty dogs comprised the control population of NoBM. Dogs with BM were comprised of 62 spayed females, 54 neutered males, 4 intact males, and 3 intact females. Their mean age was 10 years (range, 1 to 18 years), median BCS was 6/9 (range, 2 to 9), and mean body weight was 11.2 kg (range, 2.5 to 40.0 kg). The skull shape included mesocephalic (78), brachycephalic (31), and dolichocephalic (14). In the NoBM group, there were 8 spayed females, 11 neutered males, and 1 intact male. Their mean age was 9 years (range, 2 to 14 years), median BCS was 6/9 (range, 4 to 9), and mean body weight was 15.6 kg (range, 1.9 to 40.0 kg). The skull shape included mesocephalic (11), brachycephalic (3), dolichocephalic (3), and unknown (3). A summary of the breeds and skull shapes in each group is provided (Supplementary Table S1).
Suspect and definitively diagnosed comorbid respiratory diseases in BM and NoBM dogs are provided (Supplementary Table S2). Dogs with BM had a mean of 5 comorbid respiratory and aerodigestive diseases (range, 1 to 14), with most dogs having ≥ 3 diagnoses. Overall dogs with BM had 386 definitive and 163 suspect respiratory and aerodigestive diagnoses. Dogs in the NoBM group had a mean of 2 comorbid diseases (range, 1 to 7), with 33 being definitive and 7 being suspect.
The main effects and the BM by respiratory phase interaction were statistically significant for MLA, percent LAA at −856 HU, percent HAA at −700 HU, and percent AA between −600 and −250 HU. Computed tomography slice thickness was a statistically significant covariate for each outcome and thus retained. Model-based summary statistics and comparisons were adjusted for slice thickness.
The pairwise summary of covariate-adjusted medians of MLA, percent LAA at −856 HU, percent HAA at −700 HU, and percent AA between −600 and −250 HU between phases in dogs with BM and NoBM with a 95% CI for differences are provided (Supplementary Table S3). The median MLA values with 95% CIs are shown (Figure 2). There was a significantly greater increase in MLA when comparing the difference between inspiration and expiration in the BM (156 HU) versus NoBM (90 HU) groups (P = .001). On average, compared with inspiration, dogs with BM had a 22% increase in MLA on expiration versus 11% for NoBM dogs supporting impaired parenchymal aeration downstream of airway collapse in BM dogs over what would be expected with typical exhalation. On imaging, dogs with BM that had the most severe increase in attenuation on exhalation were characterized by marked ground-glass opacity or consolidation in areas surrounding the collapsed airways (Figure 3). In contrast, NoBM dogs had an overall minor increase in lung attenuation on expiration. Dynamic changes in lung attenuation from inspiration to expiration in a dog with BM are provided (Supplementary Video S1). Tidal volume to percent change in lung volume between inspiration and expiration for BM and NoBM dogs are provided (Supplementary Figure S1).
Median values of inspiratory and expiratory mean lung attenuation (MLA) with 95% CIs in dogs with BM and without BM (NoBM). The MLA was significantly increased in dogs with BM versus NoBM on inspiration and expiration (P < .001 for both), likely reflecting comorbid airway and/or parenchymal diseases in dogs with BM. Importantly, BM dogs had a significantly greater change in difference between inspiratory and expiratory MLA (P = .001) than NoBM dogs reflecting downstream effects of segmental and subsegmental collapse on the pulmonary parenchyma.
Citation: American Journal of Veterinary Research 86, 1; 10.2460/ajvr.24.07.0186
A—Inspiratory (A) and expiratory (B) breath-hold CT images of a 12-year-old female spayed Miniature Schnauzer with severe (grade III) BM. On inspiration, segmental (white arrows) and subsegmental (arrowheads) airways are opened and have a relatively circular shape. Most of these airways collapse or are severely narrowed during expiration. On expiration, the pulmonary parenchyma is markedly increased in attenuation especially surrounding the collapsed airways, reflective of impaired downstream parenchymal aeration. Standardized nomenclature is used to describe segmental and subsegmental airways. B—Inspiratory (C) and expiratory (D) breath-hold CT images of a 12-year-old female spayed Border Collie without BM. On inspiration, segmental (white arrows) and subsegmental (arrowheads) airways are opened and have a relatively circular shape. On expiration, the airways maintain their shape although they are slightly narrower than on inspiration. Overall, the lung parenchyma is minimally increased in attenuation on expiration compared to inspiration although it remains homogenous including that surrounding the airways. A sedated upper airway examination revealed complete lack of abduction of the arytenoid cartilages on inspiration diagnostic for laryngeal paralysis. Tracheobronchoscopy showed hyperemia and increased vascularity of the proximal trachea and the distal trachea and bronchi were unremarkable. These changes were consistent with gastroesophageal reflux. Ao = Descending aorta. CVC = Caudal vena cava. E = Esophagus. LB2 = Left caudal lobar bronchus. LB2V2 = Second ventral segmental bronchus of LB2. RB3D1 = First dorsal segmental bronchus emerging from the right accessory lobar bronchus. RB3V2 = Second ventral segmental bronchus from the right accessory lobar bronchus. RB4 = Right caudal lobar bronchus. RB4D2 = Second dorsal segmental bronchus of RB4. RB4V1 = First ventral segmental bronchus of RB4.
Citation: American Journal of Veterinary Research 86, 1; 10.2460/ajvr.24.07.0186
Dogs with BM had a significantly greater reduction in percent LAA at −856 HU from inspiration to expiration than NoBM dogs (Figure 4; P = .016). In dogs with BM, the median percentage of lung voxels below −856 HU decreased from 29.6% (95% CI, 27.2% to 32.0%) on inspiration to 8.2% (95% CI, 7.1% to 9.4%) on expiration, representing an approximately 3.6-fold decrease. NoBM dogs had a median of 40.7% (95% CI, 34.3% to 47.1%) and 17.3% (95% CI, 13.4% to 21.3%) of their lung with an attenuation below −856 HU on inspiration and expiration, respectively, reflecting a 2.4-fold decrease between the 2 phases of respiration.
Mean percent attenuation for the percentage of voxels more negative than −856 HU (low-attenuation area [LAA]) with 95% CIs as estimated from beta-regression modeling. Dogs with BM had a significantly greater reduction in percent LAA at 856 HU from inspiration to expiration than dogs without BM (P = .016).
Citation: American Journal of Veterinary Research 86, 1; 10.2460/ajvr.24.07.0186
For percent HAA at −700 HU, a significant difference between groups (BM vs NoBM) was attributed to the phase of respiration (P < .001). This is illustrated by an increase in the percentage of lungs with voxels above −700 HU on expiration in dogs with BM compared to that of NoBM dogs, as reflected by the absence of parallelism between the lines for each group (Figure 5). Compared with inspiration, dogs with BM had a 2.2-fold increase in percent HAA at −700 HU on expiration, while the increase in expiration of NoBM dogs was 2.0-fold. In other words, the proportion of lungs having attenuation values above (ie, more positive than) −700 HU in dogs with BM was higher than expected for the expiratory phase alone. This discrepancy from inspiration to expiration between BM and NoBM dogs was further accentuated for AA between −600 HU and −250 HU (P < .001), wherein BM dogs had a 2.5-fold increase and NoBM dogs had a 2-fold increase.
A—Mean percent attenuation for the percentage of voxels more positive than −700 HU (high-attenuation area [HAA]) with 95% CIs as estimated from beta-regression modeling. There was a significant difference in percent HAA at 700 HU between phases of respiration in dogs with and without BM (P = .001). B—Mean percent attenuation area (AA) for the percentage of voxels between −600 and −250 HU with 95% CIs as estimated from beta-regression modeling. There was a significant difference in percent AA between −600 and −250 HU between phases of respiration in dogs with and without BM (P < .001).
Citation: American Journal of Veterinary Research 86, 1; 10.2460/ajvr.24.07.0186
Dogs with BM demonstrated an approximately 0.6-fold lower kurtosis (P < .001) and a 0.6-fold decreased skewness (P < .0001) than NoBM dogs, although this effect was independent of the phase of respiration. In both groups, the kurtosis (P = .014) and skewness (P < .001) decreased from inspiration and expiration. Representative CT histograms from a BM and NoBM dog are illustrated demonstrating the less peaked and less skewed CT histogram from the dog with BM versus NoBM on both phases of respiration (Figure 6).
Representative CT histograms of a dog with BM and a dog without BM (NoBM) on inspiration (Insp) and expiration (Exp). The dog with BM has a density histogram curve that is less peaked (ie, decreased kurtosis) and less skewed to the left than the NoBM dog.
Citation: American Journal of Veterinary Research 86, 1; 10.2460/ajvr.24.07.0186
Discussion
Quantitative, semiautomated CT analysis employing inspiratory and expiratory breath-hold CT lung attenuation histograms identified objective metrics supporting changes to pulmonary parenchymal attenuation in dogs with BM compared to NoBM that may be complimentary to assessing alterations in segmental and subsegmental airway calibers and shapes. Dogs with BM versus NoBM had a higher MLA than expected on expiration supporting the presence of impaired parenchymal aeration downstream of affected airways. This is consistent with findings of several dogs with BM having greater than 50% change in lung attenuation between inspiration and expiration while none of the NoBM dogs were above this threshold. The percent LAA at −856 HU, a surrogate marker for air trapping in people, decreased more significantly from inspiration to expiration in dogs with BM versus no BM. This provides further evidence that, perhaps counterintuitively, air trapping is not a feature of segmental and subsegmental airway collapse (ie, BM).5 The percent HAA at −700 HU is a metric that relates to increased attenuation that could arise from an increase in cells, matrix, extracellular fluid, or blood, decreased air, or both.20 Dogs with BM had a greater increased percent HAA at −700 HU from inspiration to expiration supporting accentuated diminished aeration of the lung during exhalation compared to NoBM dogs. This was similar for the percent AA between −600 and −250 HU, which is a surrogate marker of interstitial lung disease in humans.1,12–14 Lower kurtosis and skewness in dogs with versus without BM were observed supporting the presence of greater parenchymal heterogeneity and rightward shift of attenuation values. Collectively, these findings demonstrate that dogs with BM exhibit a greater increase in lung attenuation than that attributable to the expiratory phase alone and greater lung heterogeneity on expiration than dogs with NoBM. Together, these metrics document that segmental and subsegmental airway collapse are accompanied by parenchymal changes. These results could be assessed in future studies to determine their utility in improving BM characterization, translating into more accurate diagnoses, sensitive disease monitoring, and prognosis.
Quantitative CT analysis assessment has been shown in human studies to produce objective, reproducible, and quantifiable evaluation of lung parenchymal changes.21 Objective metrics of CT lung attenuation capable of identifying disease extent are correlated with clinical parameters in humans with interstitial lung disease.22 Interstitial lung diseases are complex due to their diverse nature, and accurate interpretation of high-resolution CT has demonstrated wide intra- and interobserver variability.21 Evaluating the progression or regression of the disease has prognostic significance and is necessary to provide treatment guidelines; therefore, quantitative tools that are reliable and reproducible are desirable.23 Research pertaining to the recent pandemic demonstrated significant challenges with diagnosis, especially when tests were not readily available or were unreliable. This disease highlighted the need for alternative diagnostic modalities, driving the investigation of the role of quantitative tools, especially important as CT imaging features of COVID-19 can be very nonspecific overlapping with other pulmonary diseases.24 The COVID-19 patients with comorbid lung diseases also have a poorer prognosis, and the use of advanced imaging tools may help to better quantify disease burden and phenotypes, improving prognostication and management.24
Due to the semiautomated nature of radiomics programs, human software can be adapted for use in veterinary patients expanding our diagnostic capabilities.25 Although the primary means of assessing airway collapse is to investigate changes in the caliber and shape of airways, subjective increases in lung attenuation on the expiratory phase of inspiratory and expiratory breath-hold CT scans led to the study hypothesis that dogs with BM have measurable objective changes in lung attenuation compared to control dogs without BM. Objective metrics of CT lung attenuation capable of identifying disease extent are correlated with clinical parameters in humans with interstitial lung disease.22 Recognizing associated parenchymal changes downstream of collapsing segmental and subsegmental airways using objective histogram-based quantitative techniques was shown in this study to be a viable technique and could improve the detection of BM in dogs.
Radiomics offers the ability to extract image data that would otherwise not be apparent visually and indexes are derived from lung density histograms. The MLA represents the average attenuation value of the segmented lung parenchyma. Increasing MLA reflects a decrease in aeration of the lungs resulting from changes in lung volume or filling of air spaces. Lung attenuation on CT is affected by tissue within the alveolar walls, the interstitium, and amount of air present in the alveoli and small airways.5 Increased MLA could result from the presence of abnormal cells, increased vascular blood volume, or decreased alveolar air volume from either a reduction in the expansion of air spaces from collapse or replacement of air with abnormal cells or fluid. Increased attenuation can be visualized as ground-glass opacity, consolidation, and attenuation greater than soft tissue opacity.6 Compared to control dogs, dogs with BM had significantly greater increased MLA on expiration supporting loss of aeration.
The percent LAA at −856 HU reflects the percent voxels more negative than −856 HU, documenting areas with abnormally low attenuation. This metric is useful when approximating the extent and severity of diseases that result in lower lung density, for example, in people with airway disease (ie, asthma or chronic obstructive pulmonary disease) resulting in air trapping.23,26 Dynamic air trapping results from the retention of air within the lungs distal to a complete or partial airway obstruction during expiration.26 Factors such as inflammation, bronchoconstriction, excessive mucus, and decreased elastic recoil associated with obstructive airway diseases result in increased airway resistance. This in turn leads to premature airway closure increasing the volume of air retained in the lungs at the end of expiration.27 Findings in our study suggest that air trapping is not a feature of dogs with BM.
High-attenuation areas include voxels more positive than −700 HU and are threshold measurements used to evaluate the presence of respiratory disorders resulting in increased attenuation. Dogs with BM had an increase in percent HAA at −700 HU attributed to the phase of respiration, most notable on expiration. Evaluation of the percent AA between −600 and −250 HU in humans is used as a surrogate marker of interstitial lung disease.1,12 While this parameter has not yet been applied to dogs, it is reasonable to associate diseases having an accumulation of cells, edema, or matrix with a higher percent AA between −600 and −250 HU. Interestingly, in dogs with BM, the higher percent AA between −600 and −250 HU is dynamic and based on phase of respiration; in other words, the increased percent AA between −600 and −250 HU on expiration dramatically improves (returns to more normal) with inspiration (see Supplementary Video S1). This suggests that dynamic loss of air downstream of segmental and subsegmental airways can cause increased attenuation. What has not previously been reported but is supported by the results of this study is that a dog with BM having a CT scan obtained during or close to the expiratory phase (eg, dogs that are hyperventilated to achieve a breath hold, sedated dogs with low tidal volumes, or dogs where there is no control of the phase of respiration caught on exhalation) will have lungs that appear hyperattenuating and consistent with a primary parenchymal disease. Bronchomalacia should be considered as a differential diagnosis in dogs having expiratory or near-expiratory CT scans with airway-centric regions of the lung having more positive HU, especially if segmental or subsegmental airways are narrowed (Figure 3).
Kurtosis and skewness represent the distortion or disparity of a histogram when compared to a normal distribution.12,28 Voxels are plotted to obtain the average lung attenuation to determine kurtosis (sharpness of the peak) and skewness (measurement of distribution asymmetry).1,12 In normal lungs, there is generally a sharp peak (high kurtosis) compared to a normal Gaussian distribution, and the data are skewed to the left and are less symmetric (leftward skewness).1,12 The overall attenuation in normal lungs depends mostly on the degree of aeration (air content) and degree of perfusion (blood content); any regional abnormality of either aeration or perfusion translates into heterogeneity of the parenchyma. This fits with the lower kurtosis and skewness reflecting parenchymal heterogeneity associated with BM and comorbid airway and lung parenchymal abnormalities. As the downstream aeration of lung parenchyma is impaired on expiration, lung attenuation increases in areas surrounding the airways, the severity of which may be related to the extent (ie, unilateral vs bilateral, multifocal vs diffuse) and severity of the collapse. Our study was unable to identify additional contributions of BM to lung heterogeneity from inspiration to expiration (absence of interaction of respiratory phase). This may suggest that kurtosis and skewness are not sensitive markers for BM dogs with concurrent airway and/or parenchymal disease.
Our study had several limitations. First, there are no comparable veterinary CT radiomic applications in dogs with lung disease, making it necessary to apply the human parenchymal analysis extension module of the 3D slicer Chest Imaging Platform. The use of different protocols for image acquisition (ie, image reconstruction, postprocessing, ventilation), equipment (ie, CT), and anatomic differences could impact the ability to translate applications between species. A standardized protocol is imperative to allow for disease quantification and prevent variability in interpretation. Second, there was a relatively small number of dogs in the NoBM group used as the control population. This was largely because the study was retrospective and required the absence of BM with negligible observed parenchymal, airway, or vascular disease on paired inspiratory and expiratory breath-hold CT. Third, as has been shown previously, dogs with BM commonly have multiple comorbid conditions, which may affect parenchymal attenuation. While this likely explains the higher MLA in BM dogs versus NoBM dogs on inspiration (Figure 2), the focus of this article is on describing the dynamic changes (from inspiration to expiration) due to segmental and subsegmental airway collapse and thus provides meaningful data.
In conclusion, radiomics may offer advantages as an imaging biomarker for detecting BM in dogs. In dogs with BM versus NoBM having inspiratory and expiratory breath-hold CT, quantitative CT-generated lung attenuation histograms showed that multiple objective metrics supported impaired downstream parenchymal aeration associated with BM. While changes in airway caliber and shape have traditionally been used to identify and assess the severity of BM, objective CT attenuation metrics on paired breath-hold CT scans may be of benefit as a complementary diagnostic tool. Further prospective studies are needed to better establish reference ranges of objective lung attenuation metrics over a wider range of canine breeds and conformations (eg, deep chested vs barrel chested), thus opening the field for additional studies using quantitative CT lung densitometry in other disease states. Radiomics provides the possibility for objective, reproducible characterization of lung diseases with the potential to improve our diagnostic capabilities.
Supplementary Materials
Supplementary materials are posted online at the journal website: avmajournals.avma.org.
Acknowledgments
The authors thank Heidi Weideman for technical assistance with CT acquisition.
Disclosures
Drs. Reinero and Vientos-Plotts speak at conferences on respiratory disease and may receive honoraria and reimbursement for travel and accommodations. These presentations may encompass topics on bronchomalacia, but compensation is independent of and unrelated to the current study. Drs. Reinero and Vientos-Plotts have received gifts for respiratory research, but this study was not funded by those gifts. No other authors have conflicts of interest to disclose.
No AI-assisted technologies were used in the generation of this manuscript.
Funding
The authors have nothing to disclose.
References
- 1.↑
Chen A, Karwoski RA, Gierada DS, Bartholmai BJ, Koo CW. Quantitative CT analysis of diffuse lung disease. Radiographics. 2020;40(1): 28–43.
- 2.↑
Bottero E, Bellino C, De Lorenzi D, et al. Clinical evaluation and endoscopic classification of bronchomalacia in dogs. J Vet Intern Med. 2013;27(4):840–846.
- 3.↑
Johnson LR, Pollard RE. Tracheal collapse and bronchomalacia in dogs: 58 cases (7/2001–1/2008). J Vet Intern Med. 2010;24(2):298–305.
- 4.↑
Gamracy J, Wiggen K, Vientos-Plotts A, Reinero C. Clinicopathologic features, comorbid diseases, and prevalence of pulmonary hypertension in dogs with bronchomalacia. J Vet Intern Med. 2022;36(2):417–428.
- 5.↑
Masseau I, Reinero CR. Thoracic computed tomographic interpretation for clinicians to aid in the diagnosis of dogs and cats with respiratory disease. Vet J. 2019;253:105388.
- 6.↑
Bianco Z, Bukoski A, Masseau I, Reich C, Schultz L, Reinero C. Risk factors and outcomes in dogs with respiratory disease undergoing diagnostic airway lavage. Front Vet Sci. 2020;7:165.
- 7.↑
Levy A, Reinero C, Masseau I. Ventilator-assisted inspiratory and expiratory breath-hold thoracic computed tomographic scans can detect dynamic and static airway collapse in dogs with limited agreement with tracheobronchoscopy. Animals (Basel). 2022;12(22):3091.
- 8.↑
Bibb R, Eggbeer D, Paterson A. Medical Modelling: the Application of Advanced Design and Rapid Prototyping Techniques in Medicine. 2nd ed. Elsevier/WP; 2015.
- 9.↑
Weatherley ND, Eaden JA, Stewart NJ, et al. Experimental and quantitative imaging techniques in interstitial lung disease. Thorax. 2019;74(6):611–619.
- 10.↑
Reinero CR, Masseau I. Lower airway collapse: revisiting the definition and clinicopathologic features of canine bronchomalacia. Vet J. 2021;273:105682.
- 11.↑
Howard J, Grobman M, Lever T, Reinero CR. Videofluoroscopic swallow study abnormalities identify aerodigestive disorders in dogs with respiratory disease versus healthy controls. J Vet Intern Med. 2023;37(3):1166–1178.
- 12.↑
Mascalchi M, Camiciottoli G, Diciotti S. Lung densitometry: why, how and when. J Thorac Dis. 2017;9(9):3319–3345.
- 13.
Manichaikul A, Wang XQ, Sun L, et al. Genome-wide association study of subclinical interstitial lung disease in MESA. Respir Res. 2017;18(1):97.
- 14.↑
Lederer DJ, Kawut SM, Hoffman EA, Hunninghake G, van Beek EJ, Austin JH, Jiang R, Lovasi GS, Barr RG. Cigarette smoking is associated with subclinical parenchymal lung disease: the Multi-Ethnic Study Of Atherosclerosis (MESA)-lung study. Am J Respir Crit Care Med. 2009;180(5):407–414.
- 15.↑
Wei Y, Pere A, Koenker R, He X. Quantile regression methods for reference growth charts. Stat Med. 2006;25(8):1369–1382.
- 16.↑
Ferrari SLP, Cribari-Neto F. Beta regression for modelling rates and proportions. J Appl Stat. 2004;31(7):799–815.
- 17.↑
Smithson M, Verkuilen J. A better lemon squeezer? Maximum-likelihood regression with beta-distributed dependent variables. Psychol Methods. 2006;11(1):54–71.
- 19.↑
Machado JAF, Parente PMDC, Santos Silva JMC. QREG2: Stata module to perform quantile regression with robust and clustered standard errors. In: Statistical Software Components; Boston College Department of Economics; 2021.
- 20.↑
Verschakelen JA, De Wever W, Matamoros A. Computed tomography of the lung: pattern approach. J Nuclear Med. 2008;49(1):164–164.
- 21.↑
Ufuk F, Demirci M, Altinisik G. Quantitative computed tomography assessment for systemic sclerosis-related interstitial lung disease: comparison of different methods. Eur Radiol. 2020;30(8):4369–4380.
- 22.↑
Camiciottoli G, Orlandi I, Bartolucci M, et al. Lung CT densitometry in systemic sclerosis: correlation with lung function, exercise testing, and quality of life. Chest. 2007;131(3):672–681.
- 23.↑
Schroeder JD, McKenzie AS, Zach JA, et al. Relationships between airflow obstruction and quantitative CT measurements of emphysema, air trapping, and airways in subjects with and without chronic obstructive pulmonary disease. AJR Am J Roentgenol. 2013;201(3):W460–W470.
- 24.↑
Nagpal P, Guo J, Shin KM, et al. Quantitative CT imaging and advanced visualization methods: potential application in novel coronavirus disease 2019 (COVID-19) pneumonia. BJR Open. 2021;3(1):20200043.
- 25.↑
Bouhali O, Bensmail H, Sheharyar A, David F, Johnson JP. A review of radiomics and artificial intelligence and their application in veterinary diagnostic imaging. Vet Sci. 2022;9(11):620.
- 26.↑
Tan S, Saffar B, Wrobel J, Laycock A, Melsom S. Air trapping in small airway diseases: a review of imaging technique and findings with an overview of small airway diseases. J Med Imaging Radiat Oncol. 2023;67(5):499–508.
- 28.↑
Cheng GZ, San Jose Estepar R, Folch E, Onieva J, Gangadharan S, Majid A. Three-dimensional printing and 3D slicer: powerful tools in understanding and treating structural lung disease. Chest 2016;149(5):1136–1142.
