Received: 6 March 2019 Revised: 12 January 2020 Accepted: 30 January 2020

DOI: 10.1111/vru.12859

O R I G I N A L I N V E S T I G AT I O N

Radiographic differentiation of mediastinal versus pulmonary

masses in dogs and cats can be challenging

Jennifer Ruby Scott Secrest Ajay Sharma

Department of Biosciences and Diagnostic

Imaging, College of Veterinary Medicine,
University of Georgia, Athens, Georgia
Correspondence
Jennifer Ruby, Department of Biosciences
and Diagnostic Imaging, College of Veterinary
Medicine, University of Georgia, 2200 College
Station Rd, Athens, GA 30605.
Email: jlr81370@uga.edu
Abstract
The ability to differentiate thoracic masses of mediastinal and pulmonary origins is often con-
founded by their complex spatial relationship. The objectives of this retrospective, observational
cross-sectional study were to assess radiographic differentiation of mediastinal versus pulmonary
masses, and to determine if there are any correlations with specific radiographic findings. Tho-
racic radiographs of 75 dogs and cats with mediastinal and/or pulmonary masses identified on CT
were reviewed. Radiographic studies were anonymized, randomized, and reviewed twice by three
reviewers. Reviewers categorized the origin of each mass(es) as mediastinal, pulmonary, or both.
On the second review, the presence or absence of 21 different radiographic findings was recorded
for each mass. Agreement between the radiographic and CT categorization of mass origin, as well
as inter- and intraobserver agreement, was calculated. Overall agreement between radiographs
and CT was moderate for both mediastinal (68.6%) and pulmonary masses (63%). Overall, inter-
observer agreement was moderate (𝜅 = 0.50-0.74), with moderate to strong intraobserver agree-
ment (𝜅 = 0.58-0.93). Masses within the mediastinum were significantly more likely to displace
other mediastinal structures. Alternatively, masses lateral to midline and in the caudal thorax were
found to be significantly positively correlated with a pulmonary origin. The results of this study
highlight the limitations of radiography for differentiation of mediastinal and pulmonary masses,
with mass location and displacement of other mediastinal structures potentially useful for radio-
graphic findings that may help improve accuracy.

KEYWORDS
computed tomography, lung, thorax

1 INTRODUCTION branchial cyst, esophageal enlargement, hernia, abscess, hematoma, or

Mediastinal masses are not uncommon in the dog and cat, and may
be found incidentally on screening thoracic radiography or in asso-
ciation with coughing, dyspnea, or other respiratory-related clinical
signs.1–3 The most common neoplastic etiologies of mediastinal ori-
gin in the dog and cat include malignant lymphadenomegaly (eg, sec-
ondary to multicentric lymphoma) and thymoma, with ectopic thy-
roid tumor, heart base tumors, and other neoplasms occurring less
frequently.1 Mediastinal masses of nonneoplastic etiologies include
reactive lymphadenopathy (sternal, mediastinal, or tracheobronchial),
granuloma (such as secondary to Spirocerca lupi infection).4,5
Given the complex spatial relationship between the mediastinum
and surrounding pulmonary parenchyma and the spatial and contrast
resolution limitations of two-dimensional radiography, oftentimes it
can be difficult to distinguish between a mass of mediastinal versus pul-
monary origin.4 Generally, pulmonary masses can be found lateral to
midline with distinct margins contrasted by adjacent gas-filled lung.4
However, medially located pulmonary masses or mediastinal masses
that deviate away from midline can easily be misclassified.4 If a mass is
determined to be arising from the pulmonary parenchyma, an alternate

Portions of this study were presented at the ACVR Scientific Conference in Fort Worth, TX on October 19th, 2018.

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided
the original work is properly cited.

c 2020 American College of Veterinary Radiology

Vet Radiol Ultrasound. 2020;1–9. wileyonlinelibrary.com/journal/vru 1

2 RUBY ET AL .

differential diagnoses list is considered including neoplasia (primary 2.2 Imaging analysis
pulmonary carcinoma, histiocytic sarcoma, squamous cell carcinoma,
Computed tomography studies were randomized, anonymized, and
or metastatic neoplasia), abscess, hematoma, or granuloma.1,4,6 It is
initially reviewed by three readers (second year radiology resident
critical to distinguish between masses of mediastinal and pulmonary
and two ACVR-certified radiologists). By consensus, readers classified
origin for prioritization of differential diagnoses, guiding tissue sam-
intrathoracic masses as mediastinal, pulmonary, or both (eg, patient
pling, and for surgical planning.
has two masses with one each in the mediastinum and pulmonary
Computed tomography (CT) is superior to radiography in the detec-
parenchyma). For the radiographic review, in each group (mediasti-
tion and characterization of intrathoracic lesions providing better
nal, pulmonary, both), half of the radiographic studies were randomly
contrast resolution and elimination of superimposed anatomy, with
selected and duplicated to assess intraobserver agreement. All orig-
increasing availability in veterinary medicine.7 Additionally, CT is par-
inal and duplicated radiographic studies were anonymized, random-
ticularly useful for tumor staging and for the evaluation of surgical
ized, and evaluated by the same three readers individually. Readers
resectability and presence of malignant vascular invasion.2 However
were blinded to signalment and diagnosis, reviewing all studies on stan-
radiography remains the initial diagnostic test of choice given the
dard image archiving and communication system software (version
widespread availability and relatively low cost.
9.5.2, OsiriX Foundation, Geneva, Switzerland). Reviewers were able
Based on our review of the literature, the accuracy of diagnos-
to manipulate the images as needed, including changing window width,
ing mediastinal masses verses pulmonary masses on radiography in
window level, and magnification.
small animals has not been previously published. In humans, the radio-
During the first radiographic evaluation, reviewers determined if
graphic characteristics of mediastinal masses include obtuse mar-
radiographs were of diagnostic quality. If radiographic quality was
gins with the surrounding lung, a lack of air bronchograms within
deemed adequate, reviewers then determined if a mass or masses
the mass, and lack of movement of the mass between inspiratory
were present, and categorized them as mediastinal, pulmonary, or
and expiratory radiographs.8,9 In veterinary medicine, it has been
both. This review was again repeated on the same cases after an 8-
suggested that masses with the following features are likely to be
week interval. In addition, the presence or absence of 21 different
mediastinal in origin: the mass lies adjacent to midline, the mass
radiographic features including thoracic location of the mass (midline
is in a position consistent with the cranioventral or caudoventral
or lateral to midline; cranial, middle or caudal thorax; dorsal or ven-
mediastinal reflection, or the mass causes deviation of a mediastinal
tral thorax), margination (ill-defined or well-defined margins), opac-
structure.4
ity (internal mineralization, air bronchograms within the mass), and
The objectives of this study were twofold: to determine the agree-
relationship to adjacent structures (displacement of mediastinal struc-
ment of diagnosing a mediastinal versus pulmonary mass on thoracic
tures, broad-base of contact with thoracic wall, summation vs. silhou-
radiography as compared to CT, and to determine if there are any
etting of adjacent pulmonary vessels, abnormal adjacent pulmonary
discriminatory radiographic imaging characteristics associated with
pattern, displacement and/or narrowing of bronchi, bronchus leading
mediastinal or pulmonary origins.
into mass) was assessed by each reader. For the purposes of this study,
displacement of mediastinal structures referred to masses that dis-
placed an individual or a few mediastinal structures in relation to each
2 MATERIALS AND METHODS other; masses that were associated with a complete lateralized shift
of all mediastinal structures did not qualify for this criterion. Addition-
2.1 Sample population ally, the presence of pleural effusion, pulmonary nodules, and osseous
changes adjacent to the mass were recorded.
In this retrospective, observational cross-sectional study, the elec-
tronic medical records of the University of Georgia Veterinary Teach-
ing Hospital were searched from 2008 to 2017 for dogs or cats that had
2.3 Data analysis
either a mediastinal and/or a pulmonary mass diagnosed on thoracic
CT, and also had thoracic radiographs available. A mass was defined Statistical analyses were selected and performed by a PhD statisti-
as a structure approximately 3 cm or larger. Pathologic alterations of cian, and all analyses were performed in R (R Core Team) in the rel
normal anatomy (eg, lung lobe torsion or diaphragmatic hernia) were 1.3.1 (2017) package or SAS V 9.4 (Cary, NC). Fleiss kappa coefficients
included if a concurrent mass effect was present. Cases were included were calculated to quantify intraobserver variability for each of the
if two- or three-view thoracic radiographs (right lateral and/or left lat- three readers, and to calculate interobserver variability for the origi-
eral and dorsoventral or ventrodorsal projections) were available for nal studies on the first and second radiographic reviews. Cohen’s kappa
review and obtained within 35 days of the CT exam. A single reviewer coefficients were calculated to compare each of the three readers’
recorded the following from the medical record: patient signalment first and second readings to the CT reference standard. Additionally,
(species, age, sex), CT protocols, and if available, surgery or necropsy a second reference standard (“best right answer”) was created that
reports, and/or cytologic or histopathologic diagnoses. Cases were accounted for any cytologic and/or histopathologic results available,
excluded if a diagnosis was not obtained via fine needle aspiration, core which were considered to supersede the CT findings. Cohen’s kappa
or surgical biopsy, and/or necropsy exam. coefficients were calculated to compare each of the three readers’ first
RUBY ET AL .
or second reviews to this “best right answer” standard. A 95% confi-
dence interval was calculated for all Fleiss and Cohen’s kappa statistics.
Interpretations of strength of agreements were made using published
guidelines.10 For those cases with both a mediastinal and a pulmonary
mass, each type of mass was evaluated separately within the medi-
astinal and pulmonary categories, respectively. Using the “best right
answer” reference standard to categorize cases, the 21 radiographic
features were analyzed using generalized linear mixed models. In order
to simplify the multivariable analysis, only cases where all reviewers
agreed there was a single mass (either mediastinal or pulmonary) were
included for analysis. A single (univariable) fixed binary factor for each
criteria was used for each generalized linear mixed model. A random
intercept for each animal was included to model within animal correla-
tion since there were three sets of criteria identified by three review-
ers for each animal. A binary distribution with a logit link function was
used. Criteria with P < .10 in univariable analyses were considered
for inclusion in a multivariable generalized linear mixed model. Simple
correlations and collinearity diagnostics were used to evaluate crite-
ria multivariable analyses for collinearity. All criteria with P < .10 not
identified as causing collinearity were entered into a multivariable gen-
eralized linear mixed model. Odds ratios and a 95% confidence interval
were calculated.
3 RESULTS
3.1 Animals
A total of 75 patients (65 dogs, 10 cats) met the inclusion criteria.
Using CT as the reference standard, cases were assigned to the
following groups: 26 mediastinal, 42 pulmonary, and seven “both.”
In the mediastinal group, there were 23 dogs and three cats. In the
pulmonary group, there were 36 dogs and six cats and the “both”
group consisted of six dogs and one cat. Breeds represented in the
mediastinal group included Labrador Retrievers (five), mixed-breed
dogs (five), Yorkshire Terriers (three), German Shepherd Dogs (two),
Domestic Longhair cats (two), and one each of nine other breeds. Gen-
der distribution in the mediastinal group included 13 castrated males,
12 spayed females, and one intact male. The age-range for the medi-
astinal group was 3-14 years (median 9 years). Breeds represented in
the pulmonary group included Labrador Retrievers (seven), Domestic
Shorthair cats (five), mixed-breed dogs (three), West Highland Terriers
(three), Australian Shepherds (three), Boxers (two), Shih Tzus (two),
Cocker Spaniels (two), Miniature Schnauzers (two), Dachshunds (two),
Doberman Pinschers (two), and one each of nine other breeds. Gender
distribution in this group included 25 spayed females, 16 castrated
males, and one intact male. The age range for the pulmonary group
was 6-16 years (median 10 years). The “both” group included two
Boston Terriers, one each of a Boxer, Mastiff, Miniature Schnauzer,
Basenji, and Domestic Shorthair cat consisting of three spayed females
and four castrated males with an age-range of 6-12 years (median
10 years).
3
3.2 Retrospective computed tomographic data
Computed tomography examinations were performed using either a
dual slice (HiSpeed NX/i Pro; GE Medical Systems, Milwaukee, WI) or
64-slice multidetector CT unit (Sensation 64; Siemens, Malvern, PA).
Studies were performed with patients under sedation or general anes-
thesia with various anesthesia protocols, in either sternal or dorsal
recumbency. Images were acquired as a helical volume with a medium-
frequency soft tissue algorithm, a slice thickness of 0.6-3 mm, and a
variable pitch (0.8-1.5). Most cases had several reconstructions avail-
able including a high-frequency (lung) algorithm with a 0.6-2.0 mm slice
thickness in transverse, dorsal, and/or sagittal planes. All studies had
both pre- and postintravenous contrast series performed (iohexol 600
mgI/kg; GE Healthcare, Vacaville, CA) in either two (precontrast and
delay) or three (precontrast, arterial, and delay) phases.
3.3 Final diagnosis
In the mediastinal group, 22 of 26 cases (84.6%) had a definitive diag-
nosis based on core or surgical biopsy or necropsy exam, while the
remaining four of 26 cases (15.4%) had a diagnosis based on fine nee-
dle aspiration. Diagnoses included thymoma (eight), lymphoma (three),
thyroid carcinoma (two), and one each thymic lymphoma, thymic squa-
mous cell carcinoma, fibrosarcoma, esophageal leiomyosarcoma, aor-
tic body carcinoma, neuroendocrine tumor, undifferentiated malignant
neoplasia, metastatic adenocarcinoma, abscess, and infiltrative lipoma.
Three cases were misclassified into the mediastinal group based on
consensus of CT findings. Two cases, misclassified in the mediasti-
nal group, were surgically confirmed lung masses (pulmonary papil-
lary adenocarcinoma and undifferentiated sarcoma). The third case,
incorrectly classified in the mediastinal group, was later classified in
the “both” category due to necropsy findings of a severely enlarged
tracheobronchial lymph node (malignant undifferentiated neoplasia
from an unknown primary), which was poorly delineated from adjacent
smaller lung masses on the CT study.
In the pulmonary group, 24 of 42 cases (57.11%) had a definitive
diagnosis based on core or surgical biopsy or necropsy, which included
pulmonary papillary adenocarcinoma (eleven), primary pulmonary
histiocytic sarcoma (three), bronchoalveolar carcinoma (three),
adenosquamous carcinoma (two), hematoma (two), and one each
hemangiosarcoma, pulmonary papillary adenoma, and lung lobe tor-
sion. Eighteen pulmonary cases (42.9%) had a presumptive diagnosis
based on fine needle aspiration, which included 14 cases of carcinoma,
two cases of a poorly differentiated neoplasm, and one case each of
squamous cell carcinoma and metastatic mammary carcinosarcoma.
In the “both” group, three of seven cases (42.9%) had a defini-
tive diagnosis based on endoscopic or surgical biopsy or necropsy,
and included a case of multicentric plasmacytoma in the lung with
a concurrent thymoma, a case of blastomycosis in the lung and
intrathoracic lymph nodes, and a case of pulmonary adenosquamous
carcinoma with concurrent thymic lymphoma. The remaining four
cases were presumptively diagnosed based on fine needle aspira-
tion of the lung and/or mediastinal mass (all remaining mediastinal
4 RUBY ET AL .

TA B L E 1 Agreement of radiographic evaluation with CT by reader

First radiographic review (95% CI) Second radiographic review (95% CI)
Reader Mediastinal Pulmonary Mediastinal Pulmonary
Reviewer 1 0.71 (0.54-0.87) 0.64 (0.46-0.82) 0.75 (0.62-0.89) 0.72 (0.57-0.87)
Reviewer 2 0.49 (0.28-0.70) 0.42 (0.18-0.66) 0.64 (0.47-0.8) 0.66 (0.49-0.83)
Reviewer 3 0.67 (0.50-0.85) 0.57 (0.37-0.78) 0.75 (0.61-0.89) 0.67 (0.50-0.84)

Abbreviation: CI, confidence interval.

TA B L E 2 Intraobserver agreement of the 36 duplicated thoracic The results of univariable analysis of the 21 different radiographic
radiography studies features are listed in Table 4. Criteria identified with P < .05 in uni-
Reader Mediastinal (95% CI) Pulmonary (95% CI) variable analyses for mediastinal and pulmonary “best right answer”
Reviewer 1 0.93 (0.79-1.0) 0.84 (0.71-0.97) were as follows: on/near midline, displacement of other mediastinal
Reviewer 2 0.61 (0.47-0.75) 0.61 (0.47-0.75) structures, air bronchograms within the mass, lateral to midline, bor-

Reviewer 3 0.58 (0.44-0.72) 0.73 (0.59-0.87)

Abbreviation: CI, confidence interval.
TA B L E 3 Interobserver agreement of the 91 thoracic radiographic
studies
Review Mediastinal (95% CI) Pulmonary (95% CI)
1 0.62 (0.49-0.75) 0.50 (0.37-0.63)
2 0.74 (0.61-0.87) 0.72 (0.59-0.85)
Abbreviation: CI, confidence interval.
masses represented enlarged sternal, cranial mediastinal, and/or tra-
cheobronchial lymph nodes), and included two cases of histiocytic sar-
coma and one case each of pulmonary carcinoma and chronic lympho-
cytic leukemia.
3.4 Radiographic reviews
Agreement in mass location between radiographs and CT for the first
and second radiographic reviews of the 75 original cases is listed in
Table 1. On the first review, the combined overall agreement for the
three reviewers was 62.3% for mediastinal masses and 54.3% for pul-
monary masses. Overall agreement was similar on the second review at
71.3% for mediastinal masses and 68.3% for pulmonary masses. Using
the “best right answer” standard (available cytology or histopathol-
ogy findings supersede CT categorization) compared to radiographs,
comparing the combined agreement of mass origin between the first
and second reviews was 67.3% and 74.3% for mediastinal masses, and
60.0% and 71.7% for pulmonary masses, respectively. Between both
reviews, overall agreement with CT was similar for the radiology res-
ident (73.0% for mediastinal and 68.0% for pulmonary) and boarded
radiologists (63.8% for mediastinal and 58.0% for pulmonary).
Intraobserver agreement of the 36 duplicated cases is listed in
Table 2. Overall intraobserver agreement varied between moderate
to strong (𝜅 = 0.58-0.93), and was similar for mediastinal (𝜅 = 0.71)
and pulmonary (𝜅 = 0.73) masses. Interobserver agreement is listed in
Table 3 with moderate agreement for both mediastinal (𝜅 = 0.62-0.74)
and pulmonary masses (𝜅 = 0.50-0.72) on both reviews.
der effacement of pulmonary vessels, bronchus leading into mass, cra-
nial thorax, mid-thorax, caudal thorax, dorsal thorax, and ventral tho-
rax. Additional criteria with P < .10 in univariable analyses for medi-
astinal “best right answer” were as follows: ill-defined margins, dis-
tinct margins, and mass mineralization, and for pulmonary “best right
answer” the criterion was narrowed and/or displaced bronchi, which
brought the total number of criteria under consideration for the multi-
variable model up to 14 criteria for mediastinal and 12 criteria for pul-
monary. Mass mineralization was removed from consideration since
it caused convergence problems due to very few masses identified
with this criterion. High collinearity was identified for the following
radiographic features, which were excluded from multivariable analy-
sis: on/near midline (collinearity with lateral to midline), distinct mar-
gins (collinearity with ill-defined margins), and cranial and ventral
thoracic locations (collinearity with caudal and dorsal locations,
respectively). This brought the total number of criteria under consider-
ation for the multivariable model up to 10 criteria for mediastinal and
10 criteria for pulmonary.
On the multivariable analysis (Table 5), the criterion that signifi-
cantly (P < .05) positively correlated to a mediastinal mass was dis-
placement of other mediastinal structures (Figure 1). For pulmonary
masses, caudal thoracic and lateral to midline location were found to
be significantly positively correlated (Figure 2).
4 DISCUSSION
Differentiating masses of mediastinal versus pulmonary origin is
important for generating an accurate list of differential diagnoses, tis-
sue sampling, and treatment planning. The objectives of this study
were to assess radiographic differentiation of pulmonary and medi-
astinal masses and to determine if there were any specific radio-
graphic findings associated with a mass of pulmonary or mediastinal
origin. The study results suggest that radiographic differentiation of
pulmonary and mediastinal masses can be difficult with only moderate
accuracy and inter- and intraobserver agreement. Despite analysis of
21 different radiographic criteria, the only factor that proved to sig-
nificantly positively correlate with a mass within the mediastinum was
whether the mass displaced other mediastinal structures. In addition,
RUBY ET AL . 5

TA B L E 4 Univariable analysis of 21 radiographic criteria of mediastinal and pulmonary masses using best right answer

Criteria Mediastinal P-value Pulmonary P-value

On/near midline .0000* .0000*
Vessels overlying mass on both lateral images .3208 .3208
Broad base of contact with thoracic wall .1373 .1373
Ill-defined margins .0758* .0758*
Displacement of other mediastinal structures .0124* .0124*
Air bronchograms within the mass .0162* .0162*
Narrowed and/or displaced bronchi .0472* .0472*
Lateral to midline .0000* .0000*
Distinct margins .0949* .0949*
Abnormal pulmonary pattern adjacent to mass .3067 .3067
Border effacement of pulmonary vessels .0000* .0000*
Bronchus leading into mass .0035* .0035*
Mass mineralization n/a† n/a†
Cranial thorax .0000* .0000*
Mid-thorax .0141* .0141*
Caudal thorax .0123* .0123*
Dorsal thorax .0027* .0027*
Ventral thorax .0018* .0018*
Pleural effusion .9501 .9501
Pulmonary nodules .5049 .5049
Adjacent osseous changes n/a‡ n/a‡

*All criteria that reached significance of P < .1 in the univariable analysis were included in the multivariable analysis.
†This criterion was removed from the univariable analysis due to quasi-separation of data.
‡No cases were identified with this criterion.

TA B L E 5 Final multivariable model for mediastinal and pulmonary masses

P-value Odds ratio (95% CI)

Mediastinal criteria
Displacement of other mediastinal structures .0213* 16.9 (1.5-185.9)
Lateral to midline .0037* 0.05 (0.01-0.4)
Mid-thorax .0089 0.03 (0.003-0.4)
Caudal thorax .0027* 0.01 (<0.001-0.2)
Pulmonary criteria
Displacement of other mediastinal structures .0213* 0.06 (0.005-0.7)
Lateral to midline .0037* 20.1 (2.7-148.7)
Mid-thorax .0089 29.6 (2.4-370)
Caudal thorax .0027* 138 (5.7 → 1000)

*Criterion reached significance of P < .05 on multivariable analysis.

masses lateral to midline and in the caudal thorax were found to be sig-
nificantly positively correlated with a pulmonary origin.
Overall agreement of radiographic evaluation with CT studies was
surprisingly limited, varying from weak to moderate (𝜅 = 0.42-0.75)
depending on the reviewer. In general, accuracy was somewhat
better for mediastinal masses (66.8%) as compared to pulmonary
masses (61.3%). The authors attribute this to the more predictable
location of mediastinal masses, generally occurring on midline, which
is an unavoidable bias of the dataset.4 Additionally, canine primary
pulmonary adenocarcinomas have been reported to more commonly
occur in the center or periphery of the caudal lung lobes.6,11–13
Despite these known associations, radiographic accuracy for dis-
tinguishing masses in these two locations was only moderate. These
findings underscore the inherent limitations of radiography. Computed
tomography is well known to be superior to radiography for the assess-
ment of intrathoracic structures and subtle or complex pathologic
6 RUBY ET AL .

F I G U R E 1 Left lateral (A) and ventrodorsal (B) thoracic radiographs and transverse postcontrast CT image (C) in a soft tissue window (window
width 290 HU, window level 45 HU) at the level of the fourth thoracic vertebra of a 12-year-old female spayed Airedale terrier with a cranial
mediastinal mass diagnosed as a thymic squamous cell carcinoma on surgical biopsy. Note the dorsal displacement of the trachea by the cranial
mediastinal mass on the radiographs

F I G U R E 2 Right lateral (A) and ventrodorsal (B) thoracic radiographs and transverse postcontrast CT image (C) in a lung window (window
width 1400 HU, window level −500 HU) at the level of the eighth thoracic vertebra of an 11-year-old male castrated West Highland White Terrier
with a left caudal lung lobe mass diagnosed as a pulmonary adenocarcinoma on surgical biopsy. Note the lateral and caudal location of the mass on
the radiographs

lesions due to the three-dimensional nature of CT imaging and
superior contrast resolution.7,14–18 In a prior study, CT was found
to provide new information on pathologic location in 80% of cats
and 67.8% of dogs when compared to thoracic radiographs.7 The
limited accuracy of radiographic assessment in this study supports the
continued use of CT for investigation of intrathoracic masses.
Despite the utility of CT for assessing intrathoracic masses, this
modality is not a perfect test to distinguish mass location as evidenced
by the 4% misdiagnosis rate in the study reported here of CT com-
pared with histopathology. All three miscategorized studies were per-
formed on the dual-slice CT scanner with only a soft tissue algorithm
available for review. Two of the misclassified studies were primary pul-
monary masses miscategorized as “mediastinal.” Interestingly, both of
these pulmonary masses were within the caudal thorax on midline and
displaced the esophagus dorsolaterally (Figure 3). The authors specu-
late that availability of additional images reconstructed with a high fre-
quency (lung) algorithm may have improved accuracy for these cases
specifically. However, these cases particularly highlight that while the
criterion of a mass being lateral to midline was found to be sig-
nificantly positively correlated with a pulmonary origin, pulmonary
masses can still be found on midline (Figure 4). Thus, while the criteria
found to be significantly correlated with masses of mediastinal or pul-
monary location may help prioritize lesion location, these criteria are
not absolute, and other mass origins should still be considered, albeit
deprioritized.
Intraobserver agreement was moderate to strong for all review-
ers, with a slight difference between mediastinal and pulmonary cases
based on the reviewer. This is suggestive of the inherent biases of each
RUBY ET AL . 7

F I G U R E 3 Challenging cases: Left lateral (A) and ventrodorsal (B) thoracic radiographs and transverse postcontrast CT image (C) in a lung
window (window width 1400 HU, window level −500 HU) at the level of the second intercostal space of 12-year-old female spayed Cocker Spaniel
with a left cranial lung lobe mass diagnosed as a pulmonary papillary adenocarcinoma on surgical biopsy. While this case was correctly classified
on CT evaluation, none of the three reviewers correctly classified this mass as pulmonary in origin on the radiographic review. This case is one of
the few examples of pulmonary masses that did not follow the pattern of being located lateral to midline and in the caudal thorax. Images (D-F)
represent a case in which there was a lack of consensus between reviewers; right lateral (D) and ventrodorsal radiographs (E) and a transverse
postcontrast CT image (F) in a soft tissue window (window width 290 HU, window level 45 HU) at the level of the third intercostal space of a
6-year-old male child castrated Norfolk Terrier with a left cranial lung lobe torsion. Only two of the three reviewers correctly classified this lesion
as pulmonary. While this lesion was located lateral to midline, the cranial location was misleadingly suggestive of a mediastinal origin

reviewer based on various training, clinical experience, and/or the pref-
erential use/emphasis of specific roentgen signs in prioritizing lesion
location. Interestingly, intraobserver agreement showed higher accu-
racy for the resident reviews compared to the radiologists. Possible
reasons include resident bias during data collection/organization, rel-
atively more time spent on evaluating images and/or criteria for dif-
ferentiation of mass origin by the resident, influence of level of train-
ing where novices tend to be more detail oriented compared with
experts, or a combination thereof.19 Of note, however, there is the
slight improvement in overall accuracy of all reviewers between radio-
graphic reviews one and two. This may represent an interpretation
bias of having seen the same cases before, despite an 8-week inter-
val between the first and second reviews. Alternatively, while evalua-
tion of the 21 different criteria on the second review was not intended
to guide decision making for determining a mass’s origin, critical anal-
ysis of each of these factors may have caused the reviewers to more
carefully consider categorization of each mass as compared to the first
review, subsequently improving agreement.
A potential limitation of this study is the time delay of up to 35 days
between acquisition of radiographic and CT studies. While lesions may
have undergone some changes during this time frame, cases were only
included for evaluation if a distinct mass or masses were identified on
radiographs. For the purposes of this study to solely determine mass
origin, the time delay between studies was considered inconsequential.
This study is further limited by the number of cases reviewed. There
were several criteria in the univariable analysis, and one criterion in the
multivariable analysis that trended toward significance (P < .1), how-
ever, did not reach a significance of P < .05 on the final analysis. With
a larger dataset, these criteria could potentially prove to be significant.
Further studies could be useful to prospectively investigate whether
the inclusion of those criteria that were found to be significant in dis-
tinguishing mediastinal and pulmonary masses improves radiographic
8 RUBY ET AL .

F I G U R E 4 Misclassified case: Left lateral (A) and ventrodorsal (B) radiographs and a transverse postcontrast CT image (C) in a soft tissue
window (window width 290 HU, window level 45 HU) at the level of the eight intercostal space of a 14-year-old female spayed Jack Russell Terrier
with a surgically confirmed pulmonary papillary adenocarcinoma. This case was incorrectly classified as mediastinal on the CT evaluation based on
reviewer consensus predominately due to subjective midline location and displacement of mediastinal structures. Additionally, based on the
radiographic review this case represents another case in which there was a lack of consensus between reviewers, while the primary bronchus of
the right caudal lung lobe is abruptly tapered adjacent to the mass on the radiographs, the mass is located close to midline, which were
confounding characteristics. Only two of the three reviewers correctly classified this mass as pulmonary in origin on the radiographic review

agreement with CT. An additional bias of the study is the exclusion of Category 3
cases without a definitive diagnosis and those that did not proceed to
(a) Final Approval of the Completed Article: Ruby, Secrest,
CT for further evaluation. Generally, excluded masses included those
Sharma
that were deemed to be nonresectable and/or difficult to sample, such
as heart base tumors, or lesions with a typical appearance on radio-
graphs or CT, such as branchial cysts or primary pulmonary masses for ACKNOWLEDGMENTS
which the owner chose not to pursue additional diagnostics.
The authors thank Dr. Deborah Keyes for assistance in statistical
Accurate assessment of the origin of a mass is critical for prioritiza-
analysis.
tion of differential diagnoses, guiding tissue sampling, and for surgical
planning. This study found that masses within the mediastinum were
significantly more likely to displace other mediastinal structures. Alter- CONFLICT OF INTEREST

natively, masses lateral to midline and in the caudal thorax were found The authors declare no conflict of interest.
to be significantly positively correlated with a pulmonary origin. These
imaging features may be helpful in differentiating pulmonary and medi-
ORCID
astinal masses and may improve overall radiographic accuracy.
Jennifer Ruby https://orcid.org/0000-0002-4818-8693
Scott Secrest https://orcid.org/0000-0002-3964-7587

LIST OF AUTHOR CONTRIBUTIONS

REFERENCES
Category 1
1. Withrow SJ. Thymoma. In: Withrow SJ, MacEwan EG, eds. Small Ani-
(a) Conception and Design: Ruby, Secrest, Sharma mal Clinical Oncology. 4th ed. Philadelphia, PA: WB Saunders Company;
(b) Acquisition of Data: Ruby, Secrest, Sharma 2007:795-796.
2. Scherrer WE, Kyles AE, Samii VF, Hardie EM, Kass PH, Gregory
(c) Analysis and Interpretation of Data: Ruby, Secrest, Sharma CR. Computed tomographic assessment of vascular invasion and
resectability of mediastinal masses in dogs and a cat. N Z Vet J.
2008;56(6):330-333.
3. Yoon J, Fenney DA, Cronk DE, Anderson KL, Ziegler LE. Computed
Category 2
tomographic evaluation of canine and feline mediastinal masses in 14
(a) Drafting the Article: Ruby, Secrest, Sharma patients. Vet Radiol Ultrasound. 2004;45(6):542-546.
4. Thrall DE. The mediastinum. In: Thrall DE, ed. Textbook of Veteri-
(b) Revising Article for Intellectual Content: Ruby, Secrest, nary Diagnostic Radiology. 6th ed. St. Louis, MO: Elsevier Saunders;
Sharma 2013:554-565.
RUBY ET AL .
5. Dennis R, Kirberger RM, Wrigley RH, Barr FJ. Small Animal Radiologi-
cal Differential Diagnosis. 1st ed. London: Harcourt Publishers Limited;
2001:150-152.
6. Barrett LE, Pollard RE, Zwigenberger A, Zierenberg-Ripoll A, Skorup-
ski KA. Radiographic characteristics of primary lung tumors in 74 dogs.
Vet Radiol Ultrasound. 2014;55(5):480-487.
7. Prather AB, Berry CR, Thrall DE. Use of radiography in combination
with computed tomography for the assessment of noncardiac thoracic
disease in the dog and cat. Vet Radiol Ultrasound. 2005;46(2):114-121.
8. Felson B. The mediastinum. Semin Roentgenol. 1969;4(1):41-58.
9. George PP, Irodi A, Keshava SN, Lamont AC. Felson Signs’ revisited. J
Med Imaging Radiat Oncol. 2014;58:64-74.
10. McHugh ML. Interrater reliability: the kappa statistic. Biochem Med
(Zagreb). 2012;22(3):276-282.
11. Suter PF, Lord PF. Thoracic Radiography: A Text Atlas of Thoracic Diseases
of the Dog and Cat. Wettswil, Switzerland: PF Suter; 1984:517-624.
12. Barr FJ, Gibbs C, Brown PJ. The radiological features of primary lung
tumours in the dog: a review of thirty-six cases. J Small Anim Pract.
1986;27:493-505.
13. Marolf AJ, Gibbons DS, Podell BK, Park RD. Computed tomographic
appearance of primary lung tumors in dogs. Vet Radiol Ultrasound.
2011;52:168-172.
14. Paoloni MC, Adams WM, Dubielzig RR, Kurzman I, Vail DM, Hardie
RJ. Comparison of results of computed tomography and radiology
with histopathologic findings in tracheobronchial lymph nodes in dogs
with primary lung tumors: 14 cases (1999-2002). J Am Vet Med Assoc.
2006;228:1718-1722.
15. Ballenger EA, Adams WM, Dubielzig RR, et al. Computed tomography
characteristics of canine tracheobronchial lymph node metastasis. Vet
Radiol Ultrasound. 2010;51:397-403.
9
16. Schultz RM, Zwingenberger A. Radiographic, computed tomographic,
and ultrasonographic findings with migrating intrathoracic grass awns
in dogs and cats. Vet Radiol Ultrasound. 2008;49:249-255.
17. Nemanic S, London CA, Wisner ER. Comparison of thoracic radio-
graphs and single breath-hold helical CT for detection of pul-
monary nodules in dogs with metastatic neoplasia. J Vet Intern Med.
2006;20:508-515.
18. Esterline ML, Radlinsky MG, Biller DS, et al. Comparison of radio-
graphic and computed tomography lymphangiography for identifi-
cation of the canine thoracic duct. Vet Radiol Ultrasound. 2005;46:
391-395.
19. Woods NN, Howey EHA, Brooks LR, Norman GR. Speed kills?
Speed, accuracy, encapsulation and causal understanding. Med Educ.
2006;40(1):973-979.
SUPPORTING INFORMATION
Additional supporting information may be found online in the Support-
ing Information section at the end of the article.
How to cite this article: Ruby J, Secrest S, Sharma A. Radio-
graphic differentiation of mediastinal versus pulmonary masses
in dogs and cats can be challenging. Vet Radiol Ultrasound.
2020;1-9. https://doi.org/10.1111/vru.12859