AJR:172, January 1999 107

Contrast-Enhanced Helical CT forPulmonary Embolism Detection:Inter- and Intraobserver AgreementAmong Radiologists with VariableExperience

Carl Chartrand-Lefebvre1’2Nigel Howarth1’3

Olivier Lucidarme1Catherine Beigelman1Philippe Cluzel1Isabelle Mourey-G#{233}rosa1Mehdi Cadi1Philippe Grenier1’4

OBJECTIVE. The objective of this study was to evaluate inter- and intraobserver agree-

ment in the diagnosis of central pulmonary embolism using contrast-enhanced helical CT

among observers with variable experience in the interpretation of pulmonary CT angiograms.

MATERIALS AND METHODS. Helical CT angiograms of 60 patients clinically sus-

pected of having pulmonary embolism were analyzed retrospectively and independently by

two chest radiologists, one cardiovascular radiologist, and three general radiologists. The

films were rated a second time by the chest radiologists to assess intraobserver variability.

Findings for pulmonary embolism were categorized as positive, negative, or indeterminate at

the main, lobar, and segmental pulmonary artery levels.

RESULTS. The observers interpreted 19-21 CT angiograms as positive for pulmonary em-

holism (mean, 19.7) and one to six as indeterminate (mean, 3.2). Agreement occurred among all

observers in 50 patients (83.3%), among five observers in six patients (10.0%), among four ob-

servers in three patients (5.0%), and among three observers in one patient ( I .7%). Interobserver

agreement was very good (s, .85) for the diagnosis of pulmonary embolism on a per-patient ba-

sis. Agreement on a per-artery basis for all arteries was moderate (66%; K, .56); for lobar arteries

was good (83%; K, .75); and for segmental arteries was moderate (57%; K, .47). Mean intraob-

server agreement on a per-patient basis was very good (93%; K, .87).

CONCLUSION. Inter- and intraobserver agreement in the diagnosis of pulmonary em-

bolism with helical CT is very good despite a wide variety ofexperience among radiologists.

Received April 15, 1998; accepted after revisionJuly6, 1998.

1 Department of Radiology, H#{244}pitalPiti#{233}-Salp#{234}tri#{232}re,47-83, Blvd. de l’H#{244}pital,75651 Paris cedex 13, Universit#{233}Pierre et Marie Curie (Paris 6), France. Addresscorrespondence to P. Grenier.

2Present address: Department of Radiology, CentreHospitalier de l’Universit#{233}de Montr#{233}al,Campus St Luc,1058 rue St-Denis, Montr#{233}al,Qu#{233}becH2X 3J4, Canada.

3Present address: Department of Radiology, H#{244}pitalCantonal Universitaire de Gen#{232}ve,CH-1211, Geneva 14,Switzerland.

4lnstitut National de Ia Sante et de Ia Recherche M#{233}dicale,U494, Paris, France.

AJR 1999;172:1O7-1 12

0361-803X/99/1721-107

© American Roentgen Ray Society

H elical CT angiography of the pul-

monary arteries has emerged re-

cently as a potentially useful

diagnostic method for the detection or exclu-

sion ofpulmonary embolism [1-6]. As a mini-

mally invasive examination, this technique is

becoming widely available and has a low rate

of technical failure [3-5, 7]. Although helical

CT angiography is unable to reliably identify

small clots located beyond the segmental arter-

ies, its integration into the diagnostic algorithm

of pulmonary embolism is under discussion

[7-I I ]. Recently published prospective studies

have reported good sensitivity, specificity, and

interobserver agreement for helical CT in the

diagnosis of pulmonary embolism to the level

of the segmental arteries [3-51. It has been sug-

gested, however, that the interpretation of pul-

monary Cf angiograms may be subject to

personal experience [3, 1 1, 12]. The objective

of this study was to evaluate the inter- and in-

traobserver variability in the diagnosis of pul-

monary embolism, using contrast-enhanced

helical CT, among radiologists with variable

experience in the interpretation of CT pulmo-

nary angiograms.

Materials and Methods

Patients

Sixty patients were randomly selected from a

population of 104 patients referred between Septem-

ber 1996 and February 1997 for helical CT evalua-

tion because of a clinical suspicion of pulmonaryembolism. This randomization was done to avoid

any selection bias that could occur if we limited thestudy to the patients with a confirmation of the ab-sence or presence of pulmonary embolism by pul-

monary angiogram or by a concordant result ofventilation-perfusion lung scintigraphy and helical

CT angiography. The study group included 26 men

and 34 women who were 30-95 years old (mean, 62

years old). The clinical suspicion of pulmonary em-

holism was based on history, physical findings, chest

radiography, and blood gas. Ventilation-perfusionlung scintigraphy was performed in 54 of these pa-tients and pulmonary angiography in 38. Twenty pa-

#{149}1�:IU�lnterpretadon of6O Pulmonary CT Angiograms by Six Observers

FindingObserver

Mean (SD)1 2 3 4 5 6

Pulmonaryembolism

No pulmonary embolism

Indeterminate

Nonanalyzable arteries (n)

Nonanalyzable arteries(%)

21

37

2

141

8.1

20

39

1

78

4.5

19

40

1

77

4.4

19

36

5

307

17.6

20

36

4

89

5.1

19

35

6

174

10.0

19.7 (0.8)

37.2 (1.9)

3.2 (2.1)

144.3 (88.7)

8.3 (5.1)

Note-A total of 1740 arteries were studied by each observer for diagnosis and technical accuracy. lnterobserver agree-ment on a per-patient basis: K = .85 (p < .001).

Chartrand-Lefebvre et al.

108 AJR:172, January 1999

tients were considered to have pulmonary embolism

on the basis of positive findings on pulmonary an-giography (n = 10), a high probability result of venti-

lation-perfusion scintigraphy and positive findings

on helical CT angiography (n = 6), oran indeterminateprobability result on ventilation-perfusion scintig-

raphy and positive findings on helical CT angiogra-

phy(n=4).

Forty patients were considered not to have pul-monary embolism on the basis of normal findings onpulmonary angiography (n = 28), a normal or low-

probability result on ventilation-perfusion scintigra-

phy and normal findings on helical CT angiography

(n = 8), or an indeterminate probability result of venti-

lation-perfusion scintigraphy and normal findings on

helical CF angiography (n = 4). In these four patients,

because normal findings on helical CF angiography

cannot exclude the possibility of isolated subsegmen-tal emboli, the decision not to treat the patient was

made also on the basis of negative findings on a sono-

graphic examination of the limbs and the relatively

low clinical probability ofpulmonary embolism.

Among the 10 patients having positive findings

on pulmonary angiography, helical CT angiogra-

phy had positive findings in eight, indeterminate

findings in one, and negative findings in one.

Among the 28 patients having negative findings on

pulmonary angiography, helical Cl’ angiography

showed negative results in 20, indeterminate results in

six. and positive results in two.

CT

CFwas performed on a 7000 SR scanner (Philips

Medical Systems, Tokyo, Japan, and Best, the Nether-

lands). Helical CT pulmonary angiography was per-

formed with an acquisition (250 mA, 120 kV) from

the diaphragm to the aortic arch during a single

breath-hold. Forty-six patients were scanned with 5-

mm collimation, a 3-mm reconstruction interval, and

a pitch of 1; and 14 patients with 3-mm collimation, a

2-mm reconstruction interval, and a pitch of 1.7. All

axial images were displayed at lung (window level,-600 H; window width, 1600 H) and mediastinal

(window level, 30 H; window width, 300 H) win-

dows on a I 5-on-one film format. The mean field ofview was 295 cm (range, 205-380 mm).

The protocol for injection consisted of 110 ml of30% iodinated contrast medium(iopamidolor iohexol)

injected through a 20-gauge catheter placed in the left

antecubital vein. The injection rate was 2.7 mI/sec us-

ing an automatic injector (Medrad, Pittsburg, PA). The

delay between the beginning of the injection and the

start ofCT angiography was 10-15 sec. the longer de-

lay being chosen for patients with cardiac failure.

Interpretation ofPulmonary CTAngiograms

After a minimum of 3 months from the time of

the examination, the 60 CT examinations were inter-

preted retrospectively by six radiologists (hereafter

called the observers). All examinations were inter-

preted independently in a random fashion. The oh-servers had knowledge of the clinical suspicion of

pulmonary embolism in all patients but were un-aware of other information about the patients. Before

starting the evaluation, the six observers underwent a

training session for reading consensual interpretation

using 10 helical CT pulmonary angiograms obtained

because of a clinical suspicion of pulmonary embo-

lism. These CT angiograms were not included in the

present study.Detection of pulmonary embolism on CT angio-

grams included analysis of the main, lobar, and seg-mental pulmonary arteries. Vascular signs of

pulmonary embolism were central partial intravascu-

lar filling defects surrounded by contrast medium, ec-

centric partial filling defects surrounded by contrast

medium, complete filling defects not surrounded by

contrast medium and occupying the total vessel sec-tion, and mural defects. These criteria of positive pul-

monary embolism are defined elsewhere [1, 7, 13].

The pulmonary vascular bed was divided into 29zones: right and left pulmonary arteries; right and leftinterlobar arteries; lobar arteries of the right upper

lobe, the upper part of the left upper lobe, the right

middle lobe, the lingula (considered as a lobe), and

the right and left lower lobes; and segmental arteries

ofthe right upperlobe (apical, anterior, posterior), the

left upper lobe (apicoposterior, anterior), the right

middle lobe (lateral, medial), the lingula (superior, in-ferior), and the lower lobes (superior, mediobasal, an-

terobasal, laterobasal, posterobasal). Analysis of

subsegmental arteries was not included in the study.

Each ofthe 29 zones was categorized by the observeras positive or negative for pulmonary embolism or asinadequately depicted. Inadequate depiction could be

caused by nonoptimal pulmonary artery opacifica-

tion, presence of respiratory or motion artifacts, or

horizontal or oblique orientation of the vessel. A con-

clusion was then reached as to whether the CT studyshowed positive or negative findings for pulmonary

embolism on the basis of the pulmonary arteries con-sidered analyzable. The CT study could also be cate-

gorized as indeterminate because of ambiguous

findings or because of overall technical inadequacy if

considered so by the observer. Parenchymal signs of

pulmonary embolism were not analyzed in the study.After a 3-month interval, the CT angiograms were re-scored by two observers for intraobserver variability.

Observers

All six observers were board-certified staff ra-

diologists working in an academic hospital and

having variable clinical experience in the interpre-

tation of CT scans for pulmonary embolism diag-

nosis and other indications. Two chest radiologists

had 8 and 20 years of experience in chest CT; a

cardiovascular and interventional radiologist had 8

years of experience in conventional and digital an-

giography and 6 years of experience in body CT;

and three general radiologists had 4, 5, and 8 years

of experience in body CT (chest, abdominal, and

pelvic). The clinical experience of the observers inthe interpretation of helical CT angiograms for

pulmonary embolism diagnosis before starting this

study was 4, 8, 12, 12, 20, and 24 months (mean,

13.3 months).

StatisticalAnalysis

Inter- and intraobserver agreement was expressed

as a percentage of agreement and using the kappa

coefficient. The latter accounts for the chance agree-

ment between two or more observers I 141. Given the

large sample size of 60 patients, a normal distribu-

tion was assumed for the kappa statistic. The kappa

values obtained were compared with the null hy-

pothesis (agreement by chance only) to assess statis-

tical significance, using the t test. Kappa values were

interpreted as described by Altmann 115] (<.20,

poor; .2l-.40. fair; .41-60, moderate; .61-80,

good; .81-I .0, very good).

Results

!nterobserver Variability

A total of 1740 arterial zones in 60 CT pul-

monary angiograms were reviewed by the six

observers. These zones were 600 nonsegmen-

tal arteries (main, interlobar, lobar) and 1 140

segmental arteries.

Among the 60 CT angiograms, a mean of

19.7 (range, I 9-2 1 , depending on the ob-

server) were interpreted as positive for pulmo-

nary embolism. A mean of 3.2 CT angiograms

(range 1-6, depending on the observer) were

classified as indeterminate (Table I). Interob-

server agreement was very good for the diag-

nosis of pulmonary embolism on a per-patient

basis. Agreement occurred among the six ob-

servers in 50 patients (83.3%), among five ob-

servers in six patients (10.0%), among four

observers in three patients (5.0%), and among

three observers in one patient (1.7%). The

kappa value was .85 (p < .001) (Table 1). Al-

though the difference was not statistically sig-

nificant, the agreements between the two chest

radiologists (K, .95) were better than between

two of the general radiologists (K, .87). In the

same way, the agreements among the three

most experienced radiologists (chest, cardio-

vascular) (K, .93) were better than those among

the three general radiologists (K, .76).

Among the 1740 arterial zones studied by

each observer, the mean number of zones con-

sidered positive for pulmonary embolism was

I 97.8 (1 1.4%; range, 180-233 depending on

the observer). According to all the observers,

the distribution per lobe of the diagnosed em-

boli was as follows: right lower lobe, 25%; left

lower lobe, 19%; right upper lobe, 13%; lin-

gula, 9%; right middle lobe and left upper lobe,

both 8%. Agreement on a per-zone basis oc-

curred among the six observers in 66% of the

arterial zones, with a kappa value of .56 (mod-

erate) (p < .001). When only the lobar arteries

were considered, the interobserver agreement

was 83%, with a kappa value of .75 (good)(p <

.001 ). When only the segmental arteries were

considered, the interobserver agreement was

57%, with a kappa value of .47 (moderate) (p <

.001). Overall, discord was most frequent in the

posterior segmental artery of the right upper

lobe, the inferior and superior segmental arter-

ies of the lingula, the lateral segmental artery of

the right middle lobe, and the posterior and lat-

eral segmental arteries of the right lower lobe.

On a per-artery basis, the agreements be-

tween the two chest radiologists (K, .86) were

significantly better (p < .05) than those be-

tween pairs of general radiologists (K, .68).

lnterobserverAgreementAccorciing to Scan Collimation

On a per-patient basis, the kappa values

were not significantly different when only 3-mm

(K, .87) or 5-mm (K, .73) collimation scans

were considered.

On a per-artery basis, the kappa values were

significantly better on 3-mm (K, .63) collima-

tion scans than on 5-mm (K, .53) (p < .05). Re-

garding the lobar and segmental arteries, the

kappa values were .78 and .55, respectively, on

3-mm collimation scans versus .74 and .45, re-

spectively, on 5-mm collimation scans. These

differences, however, were not significant.

Second Interpretation by the Two Chest Radiologists

for lntraobserver Variability

An intraobserver agreement occurred in 59

patients (98.3%) and in 53 patients (88.3%) for

the two chest radiologists. The intraobserver

agreement on a per-patient basis was very good,

with a mean kappa value of .87 (range, .77-.96).

Mean intraobserver agreement on a per-zone

basis occurred in 90.4% (range, 90.1-90.7%) of

arterial zones, with a mean kappa value of .70

(range, .68-.72, good; p < .001). When only the

lobar arteries were considered, the mean in-

traobserver agreement was 95.5% (range, 94.7-

96.3%), with a mean kappa value of .86 (range,

.84-87, very good; p < .001). When only the

segmental arteries were considered, the mean

intraobserver agreement was 87% (range,

86.3-87.5%), with a mean kappa value of .63

(range, .59-66, good: p < .001).

TechnicalAdequacy byArterial Zone

The mean numbers of zones considered

nonanalyzable by a given observer were as fol-

lows: 144.3 (8.3%; range, 77-307 depending on

the observer) of 1740 arterial zones, 13 (2.2%;

range, 5-21 depending on the observer) of 600

nonsegmental arteries (main, interlobar, lobar),

and 13 1 .3 (1 1.5%; range, 70-288 depending on

the observer) of I 140 segmental arteries. A

mean of 83.5 zones (4.8%) was considered

nonanalyzable by the two chest radiologists,

compared with a mean of 174.8 (10.0%) by the

one cardiovascular and three general radiolo-

gists. The arteries most often considered nonan-

alyzable were the arteries of the inferior and

superior segments of the lingula, of the lateral

segment of the right middle lobe, and of the

posterior segment ofthe right upper lobe.

Discussion

The overall prevalence of pulmonary embo-

lism in our study was 33% and was the same as

that in the Prospective Investigation of Pulmo-

nary Embolism Diagnosis (PIOPED) study

[16]. We found very good inter- and intraob-

server agreements for the diagnosis of pulmo-

nary embolism with helical CT to the level of

the segmental arteries, with kappa values of .85

and .87. Agreement occurred among all six oh-

servers in 83.3% ofpatients. In previous studies,

the interobserver agreement has been evaluated

among pairs of observers who apparently had a

high level of experience with the interpretation

of pulmonary embolism using helical or elec-

tron-beam CT [4, 5, 7, 17]. The interobserver

agreement reported in these studies was good.

Teigen et al. [17] obtained an agreement of

95%, Goodman et al. [7] of 75%, and Mayo et

al. [5] of 92.8%. Van Rossum et al. [4] reported

a kappa value of .77 and Mayo et al. [5] of .85

(Table 2). Only the studies of Goodman et al.

[7] and van Rossum et al. [41 included the anal-

ysis of subsegmental arteries. In spite of the

greater variability of expertise in chest CT and

pulmonary CT angiography among our six oh-

servers, the interobserver agreement in our

study was very good. However, a better exper-

tise in the interpretation of helical CT angio-

#{149}fi11�*lnter- and lntraobserver Agreement for Diagnosis of Pulmonary Embolism with CT, Angiography. and Lung Scintigraphy

Study Year. .

Imaging TechniqueNo.of

.

Patients

No. ofPulmonary

Embolisms(%)

No.ofObservers

Interobserve r Agreement lntraobserver Agreement

K % K

Teigen et al. [17] 1993 Electron-beam CT� 86 39 (45) 2 95 NA 96 NA

Goodman et al. [7] 1995 Helical CT 20 11 (55) 2 75 NA NA NA

van Rossum et al. [4] 1996 Helical CIa 149 60 (4Q)b 2 NA .77 NA NA

Mayo et al. 151 1997 Helical CTa 139 46 (33) 2 92.8 .85 NA NA

This study 1998 Helical CTa 60 19.7 (33)b 6 83 .85 93,3b 87b

Stein etal. [181 1992 Conventional angiography 1111 383 (35) 2 81 NA NA NA

Quinn et al. [19] 1987 Conventional angiography 60 26 (43) 3 86 NA NA NA

van Beek et al. [201 1996 Conventional angiography 45 6 (13) 2 64.5 .28 NA NA

van Beek et al. [201 1996 Digital angiography 85 14 (17) 2 91 .77 NA NA

PIOPED [161 1990 Scintigraphy 931 (33)C 2 70-95 NA NA NA

Mayo et al. [5] 1997 Scintigraphy 136 46 (33) 2 NA .61 NA NA

Note-NA =not available, PIOPED =Prospective Investigation of Pulmonary Embolism Diagnosis.‘Did not include analysis of subsegmental arteries.

bMean number from more than one observer.

C755of the 931 patients underwent pulmonary angiography; 251 (33%) of those had positive results.

Fig. 1-48-year-old man with sudden pain in right hemithorax and clinical suspicion of pulmonary embolism.A and B, Mediastinal (A) and lung (B) window settings on helical CT scans reveal low-attenuation areas within lu-men of segmental and subsegmental pulmonary arteries of superior segment of right lower lobe. Both images wereinterpreted as positive for pulmonary embolism by two observers and as negative by four. Areas of hypoattenuation(arrows, A) could be interpreted as tortuosity of vessels and partial volume averaging or as filling defect

Chartrand.Lefebvre et al.

110 AJR:172, January 1999

grams of pulmonary arteries improves the

interobserver agreement, as suggested in our

study by significantly better agreements among

chest radiologists than among general radiolo-

gists on a per-artery basis. Expertise also had a

positive effect on the number of pulmonary ar-

teries considered analyzable on helical CT

scans, as shown in our study.

Our results and the results of previous studies

14, 5, 7, 17j show that the agreement among oh-

servers of CT is favorably comparable to the

agreement among observers of pulmonary an-

giography and lung scintigraphy in the diagno-

sis of pulmonary embolism. In a subsequent

analysis of the PIOPED study data by Stein et

al. I 18J, an 8lC/e agreement was reached be-

tween pairs of angiographers for the presence or

absence of pulmonary embolism. Quinn et al.

I l9J reported a mean interobserver agreement

of 86% in the diagnosis of pulmonary embo-

lism in a series of 60 angiograms reviewed by

three observers. In a more recent study, the

agreement between two observers of 85 digital

subtraction angiograms achieved a kappa value

of .77 (91% agreement) 1201. However, when

the same observers interpreted 45 conventional

angiograms, the kappa value decreased to .28

(64.5% agreement). These three studies using

angiography included the analysis of subseg-

mental arteries. Regarding lung scintigraphy,

the PIOPED study I 16j showed that in the mi-

nority of cases in which the test is conclu-

sive-namely, the high probability and normal

categories-the interobserver agreement was

95% and 94%, respectively. However, the

agreement was 75% and 70% in the intermedi-

ate- and low-probability categories, respec-

tively, which represent most cases.

In their recent prospective study comparing

ventilation-perfusion scintigraphy and helical

CT angiography, Mayo et al. 151 showed that

agreement between two observers was better

with helical CT than with ventilation-perfusion

scintigraphy (K, .85 versus .61).

Studies using pulmonary angiography have

shown that inter()bserver agreement is strongly

influenced by the magnitude and the site of the

thmmboembolism. Quinn et al. [191 showed that

massive pulmonary embolism is associated with

a high agreement among observers for the diag-

nosis of pulmonary embolism but with a poor

agreement when the embolism is limited to the

subsegmental level. Agreement between two oh-

servers in the interpretation of pulmonary angio-

grams from the subsequent analysis of the

PIOPED study data by Stein et al. [l8( was 98%

for lobar pulmonary embolism, 90% for seg-

mental pulmonary embolism, and 66% for sub-

segmental pulmonary embolism. Our study

shows a similar decreasing trend from central to

more peripheral arteries, with an agreement of

83% for lobar pulmonary embolism and of 57%

for segmental pulmonary embolism.

Difficulties in interpreting helical CT angio-

grains may be due to various causes. Normal or

slightly enlarged hilar lymph nodes and associ-

ated soft tissues seen as marginal areas of low

attenuation, most often located between major

bronchi and pulmonary arteries. may be a patent

cause of a false-positive diagnosis of pulmonary

embolism [21]. Although reformatted images

that depict the pulmonary artery in the longitu-

dinal axis may be helpful, an intimate knowl-

edge of the true location of the hilar lymph

nodes is desirable (22, 23j.

An oblique or transverse course of a vessel

may generate areas of low attenuation that may

be confused with pulmonary embolism on axial

images (Fig. I). These areas are mainly caused

by partial volume averaging between paren-

chyma, vessel walls, and blood. Such partial

identification of vessels may create false-positive

and false-negative results or lead the observer to

an indeterminate answer 13, 2 1 j. This cause of

misdiagnosis may explain why in our study the

inter- and intraobserver discord was higher in the

posterior segmental artery of the right upper

lobe, the inferior and superior segmental arteries

of the lingula, and the lateral segmental artery of

the right middle lobe: because these vessels are

almost parallel to and within the axial plane of

scanning (Fig. 2). Reduction of scan collimation

permits a decrease in the volume averaging ef-

fect and prtJbahly improves interobserver agree-

ment. as shown in our study.

Cardiac and respiratory motion may induce

segmental artery pseudo-filling defects I 13,

21 . Respiratory motion artifacts affect mainly

the lower lobes because of the greater difficulty

in maintaining the breath-hold at the end of the

scanned volume and could explain the higher

rate of inter- and intraobserver discord in the

analysis of the posterior and laterobasal seg-

mental arteries of the left lower lobe. The vascu-

lar contrast in the lumen of pulmonary arteries

on CT scans is largely influenced by the display

window setting (Fig. 3). A modified setting may

be useful when embc)li are obscured by dense

contrast material 1241. In our study, the retro-

spective analysis of the cases was performed on

previously printed films. Analysis at the work-

station would improve interpretation by allow-

ing the observer to modify the window settings

and to visualize the successive axial images in a

rapid display using the track ball. This cine

mode permits the fillowing of every segmental

artery from its origin to its branches. Such

equipment could have provided better inter- and

intraobserver agreement.

The evaluation of diagnostic accuracy of he-

lical CT angiography for the detection or exclu-

sion of pulmonary embolism was not the

objective of our study. We know from the litera-

ture that the technique has a good accuracy for

the identification of clot within the central lobar

D

111

Helical CT for Detection of Pulmonary Embolism

AJR:172, January 1999

and segmental pulmonary arteries. Three recent

prospective studies including 75-149 patients

reported sensitivities of 82-94% and specifici-

ties of 78-96% 13-5]. Although clot within a

subsegmental artery may be identified, many

factors currently conspire to limit helical CT to

the examination of vessels of greater than sub-

segmental size [71, which is the main reason we

did not include analysis of the subsegmental ar-

teries in our standard retrospective interpreta-

tion of the CT scans. Optimization of technical

factors, such as using a thinner collimation (2

mm) and a faster scanner (0.85-sec rotation

time), as shown by Remy-Jardin et al. [25],

should enable marked improvement in the anal-

ysis of the peripheral arteries with helical CT.

Future work would then be needed to document

interobserver variability when the subsegmental

arteries are considered.

To date, we can conclude that when the pul-

monary bed is considered to the level of seg-

mental arteries, the inter- and intraobserver

agreement in the diagnosis of pulmonary em-

Fig. 2.-55-year-old woman with unexplained dyspnea and suspected pulmonary embolism.A and B, Mediastinal (A) and lung (B) window settings on helical CT scans reveal presence of subtle small areaof hypoattenuation within lumen of segmental pulmonary artery of laterobasal segment of lower lobe (arrow, A)interpreted as positive for pulmonary embolism by one observer, indeterminate by one, and negative by four.

Fig. 3-43-year-old man with thoracic pain 8 days after hip surgery and clinical suspicion of pulmo-nary embolism.A-C,Mediastinal (A, B) and lung (C) window settings on helical CT scans reveal segmental and sub-segmental artery branches within anterior segment of right upper lobe running almost parallel to andwithin axial plane (arrows). Scans were interpreted as negative for pulmonary embolism by five ob-servers and as indeterminate by one.D,Digital right selective pulmonary angiogram reveals presence of clot within subsegmental pulmo-nary artery in anterior segment of right upper lobe (arrow).

Chartrand.Lefebvre et al.

112 AJR:172, January 1999

bolism with helical CT is very good among ra-

diologists with variable experience in the

interpretation of CT pulmonary angiograms.

The interobserver agreement with helical CT

appears to be at least equivalent to the interob-

server agreement reported in studies using

lung scintigraphy and pulmonary angiography.

References

I. Remy-Jardin M, Remy J, Wattinne L, Giraud F.

Central pulmonary thromboembolism: diagnosis

with spiral volumetric CT with the single-breath-

hold technique-comparison with pulmonary an-

giography. Radiology 1992:185:381-387

2. Blum AG, Delfau F, Grignon B, et al. Spiral-com-

puted tomography versus pulmonary angiography

in the diagnosis of acute massive pulmonary em-bolism. Am J Cardiol 1994;74:96-98

3. Remy-Jardin M, Remy J, Deschildre F, et al. Di-

agnosis of pulmonary embolism with spiral CT:comparison with pulmonary angiography and

scintigraphy. Radiology 1996;200:699-706

4. van Rossum AB, Pattynama PM, Tjin A Ton ER,

et al. Pulmonary embolism: validation of spiral

CT angiography in 149 patients. Radiology 1996;

201:467-470

5. Mayo JR. Remy-Jardin M, MUller NL, et al. Pul-

monary embolism: prospective comparison of

spiral CT with ventilation-perfusion scintigra-phy. Radiology 1997:205:447-452

6. Ferretti GB, Bosson JL, Buffaz PD, et al. Acute pul-

monary embolism: role ofhelical CT in 164 patientswith intermediate probability at ventilation-perfu-

sion scintigraphy and normal results at duplex US ofthe legs. Radiology 1997:205:453-458

7. Goodman LG, Cm-tin JJ, Mewissen MW, et al.

Detection of pulmonary embolism in patientswith unresolved clinical and scintigraphic diag-

nosis: helical CT versus angiography. AiR 1995;164:1369-1374

8. Goodman LR, Lipchik Ri. Diagnosis of acute

pulmonary embolism: time for a new approach.Radiology 1996;199:25-27

9. van Erkel AR, van Rossum AB, Bloem JL, Kievit

J, Pattynama PMT. Spiral CT angiography for

suspected pulmonary embolism: a cost-effective-

ness analysis. Radiology 1996;201:29-36

10. Gefter WB, Hatabu H, Holland GA, Gupta KB,

Henschke CI, Palevsky HI. Pulmonary throm-boembolism: recent developments in diagnosis

with CT and MR imaging. Radiology 1995;l97:

561-574

1 1. Rubin GD, Goodman LR, Lipchik RJ, et al. Heli-

cal CT for the detection of acute pulmonary em-bolism: experts debate. J Thorac Imaging 1997;

12:81-102

12. Woodard PK, Sostman HD, MacFall JR. et al. De-tection of pulmonary embolism: comparison ofcontrast-enhanced spiral CT and time-of-flight MRtechniques. J Thorac Imaging 1995:10:59-72

13. Kuzo RS, Goodman LR. CT evaluation of pulmo-

nary embolism: technique and interpretation. AiR

1997;l69:959-965

14. Fleiss IL. Measuring nominal scale agreementamong many raters. Psvchol Bull 1971:76:378-382

15. Altmann Dci. Practical statistics for medical re-

search. London: Chapman & Hall, 1992

16. The PIOPED investigators. Value of ventilation!

perfusion scan in acute pulmonary embolism: re-

sults of the Prospective Investigation of Pulmo-nary Embolism Diagnosis (PIOPED). JAMA

1990:263:2753-2759

17. Teigen CL, Maus TP, Sheedy PF II, Johnson CM,

Stanson AW, Welch TJ. Pulmonary embolism: di-

agnosis with electron-beam CT Radiology 1993;

188:839-845

I 8. Stein PD, Athanasoulis C, Alavi A, et al. Compli-

cations and validity of pulmonary angiography inacute pulmonary embolism. Circulation 1992:85:

462-468

19. Quinn MF, Lundell Ci, Klotz TA, et al. Reliabil-

ity of selective pulmonary arteriography in the di-

agnosis of pulmonary embolism. AJR 1987:149:

469-47120. van Beek EJR, Bakker AJ, Reekers JA. Pulmo-

nary embolism: interobserver agreement in the

interpretation of conventional angiographic and

DSA images in patients with nondiagnostic lung

scan results. Radiology 1996:198:721-724

21 . Remy-Jardin M, Remy J, Artaud D, Deschildre F,

Fribourg M, Beregi JP. Spiral CT ofpulmonary em-

bolism: technical considerations and interpretative

pitfalls.J Thorac Imaging 1997:12:103-117

22. Remy-Jardin M, Remy J. Cauvain 0, Petyt L,

Wannebroucq J, Beregi JP. Diagnosis of central

pulmonary embolism with helical CT: role oftwo-dimensional multiplanar reformations. AiR

1995:165:1131-1138

23. Remy-Jardin M, Duyck P. Remy J. et al. Hilar

lymph nodes: identification with spiral CT and his-

tologic correlation. Radiology 1995:196:387-394

24. Brink JA. Woodard PK, Horesh L, et al. Depic-

tion of pulmonary emboli with spiral CT: optimi-

zation of display window settings in a porcine

model. Radiology 1997:204:703-70825. Remy-Jardin M, Remy J, Artaud D, Deschildre F,

Duhamel A. Peripheral pulmonary arteries: opti-

mization of the spiral CT acquisition protocol.

Radiology 1997:204:157-163