Ppn

CMLa

b

c

d

e

a

ARRA

KPRC

1

dwa

U

h0

European Journal of Radiology 85 (2016) 150–157

Contents lists available at ScienceDirect

European Journal of Radiology

j ourna l h o mepage: www.elsev ier .com/ locate /e j rad

ulmonary arteriovenous malformation (PAVM) reperfusion afterercutaneous embolization: Sensitivity and specificity ofon-enhanced CT

hantale Bélanger a, Carl Chartrand-Lefebvre a,b, Gilles Soulez a,b, Marie E. Faughnan c,d,e,uhammad Ramzan Tahir b, Marie-France Giroux a, Patrick Gilbert a, Pierre Perreault a,

ouis Bouchard a, Vincent L. Oliva a, Eric Therasse a,b,∗

Department of Radiology, Centre hospitalier de l’Université de Montréal (CHUM), Montreal, QC, CanadaCentre de recherche, Centre hospitalier de l’Université de Montreal (CRCHUM), Montreal, QC, CanadaRespirology Division, Department of Medicine, St. Michael’s Hospital, University of Toronto, Toronto, ON, CanadaLi Ka Shing Knowledge Institute, St. Michael’s Hospital, University of Toronto, Toronto, ON, CanadaMontreal HHT Centre, Pneumology Division, CHUM, Montreal, QC, Canada

r t i c l e i n f o

rticle history:eceived 28 July 2015eceived in revised form 4 November 2015ccepted 5 November 2015

eywords:ulmonary arteriovenous malformationeperfusionT

a b s t r a c t

Purpose: To evaluate the sensitivity and specificity of non-enhanced chest CT to detect reperfusion afterpulmonary arteriovenous malformation (PAVM) embolization.Materials and methods: The Institutional Review Board approved this retrospective HIPAA-compliantstudy and waived the need for patient consent. All consecutive patients who underwent PAVM emboliza-tion between January 2000 and April 2011 were included. Complex PAVMs and patients without availablepre- and/or post-embolization CT were excluded. PAVM artery, aneurysm and vein diameters weremeasured on non-enhanced chest CT before and after PAVM embolization. Pulmonary angiography(PA) was the reference standard to assess PAVM reperfusion. Reperfusion detection was analyzed withreceiver operating characteristic (ROC) curves according to percentage of diameter reduction cut-off.Inter-observer concordance was ascertained with intra-class correlation coefficients (ICCs).Results: Out of 68 patients with PAVM embolizations, 42 (62%) had 108 PAVMs that met inclu-sion/exclusion criteria. Areas under the ROC curves for PAVM reperfusion detection were 0.84, 0.87,and 0.78, respectively, for PAVM artery, aneurysm and vein (p > 0.05). Sensitivity varied between 51%and 56%, and specificity between 86% and 98% for the <30% diameter reduction cut-off. Sensitivity was

between 98% and 100%, and specificity, between 20% and 47% for the <70% diameter reduction cut-off.ICCs for inter-observer concordance were 0.58, 0.88 and 0.68 for percentage reduction of PAVM artery,aneurysm and vein, respectively.Conclusion: PAVM diameter reduction cut-offs of <30% and <70%, to detect PAVM reperfusion on non-enhanced CT reported in the literature, would respectively result in low sensitivity and specificity.

© 2015 Elsevier Ireland Ltd. All rights reserved.

. Introduction

Pulmonary arteriovenous malformations (PAVMs) represent

irect communications between the pulmonary artery and vein,ith these right-left shunts resulting in hemorrhagic complications

nd paradoxical embolization leading to stroke and brain abscesses

∗ Corresponding author at: Department of Radiology, CHUM, 3840, rue Saint-rbain, Montréal, QC H2W 1T8, Canada. Fax: +1 514 412 7193.

E-mail address: eric.therasse.chum@ssss.gouv.qc.ca (E. Therasse).

ttp://dx.doi.org/10.1016/j.ejrad.2015.11.014720-048X/© 2015 Elsevier Ireland Ltd. All rights reserved.

[1–6]. Percutaneous PAVM embolization is technically very suc-cessful and is now the standard of care for the treatment of theselesions [7,8]. However, PAVM reperfusion rates after embolizationare estimated to be between 2 and 25% [9,10] due most often to therecanalization of embolized vessels [11,12]. Pulmonary angiogra-phy (PA) is the imaging reference standard for PAVM reperfusionassessment [7,13] but, given its cost and invasive nature, CT is the

preferred follow-up imaging modality.

CT detection of PAVM reperfusion after embolization is basedon either PAVM enhancement on contrast-enhanced CT or PAVMdiameter reduction on non-enhanced CT [9,10,14–17]. Today novel

rnal o

tsrieleid[oCtatmoHh7uoaanPsc

2

2

ste2eaamia

tBlvsae

2

tSdO4wl1o

C. Bélanger et al. / European Jou

echniques such as dual energy scanning and iterative recon-truction algorithms can help reduce such artifacts while limitingadiation dose. However, assessment of PAVM enhancement afterntravenous contrast injection may still be impaired by metallicmbolization devices and requires good contrast resolution thatimits CT radiation dose reduction to this population with long lifexpectancy subjected to repeated CT examinations. Enhanced CTs also associated with risks of contrast media and air embolismuring tubing connection to automatic contrast media injectors14,18]. Small- and moderate-size air emboli are estimated toccur in 12% to 23% of patients undergoing contrast-enhancedT examination [18,19]. As patients with hereditary hemorrhagicelangiectasia (HHT) will almost always incur tiny residual PAVMsfter treatment, there is some concern over intravenous adminis-rations which may introduce air bubbles [16]. For these reasons,

any centers have established PAVM diameter reduction cut-offsn non-enhanced CT to evaluate PAVM reperfusion [9,10,14–16].owever, the cut-offs under which reperfusion is suspected areighly variable, arbitrary and non-validated. Some authors follow0% reduction criteria for the aneurysmal sac and/or draining veinnder which PAVM reperfusion is suspected [9,11,14,15], whereasthers consider >30% reduction sufficient for successful outcomefter embolization [10,12,13]. To our knowledge, the sensitivitiesnd specificities of these cut-offs are yet unknown, and there areo data in the literature comparing CT measurement of differentAVM structures to detect PAVM reperfusion. The purpose of thistudy is to evaluate the sensitivity and specificity of non-enhancedhest CT to detect reperfusion after PAVM embolization.

. Materials and methods

.1. Patients and PAVM embolization

The Institutional Review Board approved this retrospectivetudy and waived informed consent. We included all consecu-ive patients from a database of subjects who received PAVMmbolotherapy in our institution between January 2000 and April011. The inclusion criteria were: all patients with de novo PAVMmbolization who had chest CT scan before and at least 6 monthsfter embolization. Complex PAVMs, defined as multiple feedingrteries, and those without CT imaging follow-up of at least 6onths after embolization, were excluded from this study. Imag-

ng and medical files were reviewed to obtain clinical data as wells laboratory and procedural reports.

PAVM embolization was generally performed via a venousransfemoral approach, with a 7F guide catheter (Guider Softip XF,oston Scientific, Freemont, CA, USA), with 40 degree tip angu-

ation, through a 7F introducer. Embolization was secured witharious detachable plugs, balloons and coils until complete occlu-ion was confirmed by PA. The PAVM feeding artery was embolizeds close as possible to the aneurysmal sac. Multiple PAVMs werembolized during a single session in some patients.

.2. Follow-up CT

Follow-up chest CT was performed without intravenous con-rast with a 16-detector row scanner (Somatom Sensation 16,iemens Medical Solutions, Forchheim, Germany) and a 128-etector row scanner (Brilliance iCT, Phillips Healthcare, Cleveland,H, USA). X-ray tube voltage was 120 kV, and tube current was0 mA (low-dose protocol). The following scanning parameters

ere used: for the 16—detector row scanner, 16 × 0.75-mm col-

imation, a pitch of 1.15, and a rotation time of 0.5 s and for the28—detector row scanner, 128 × 0.625-mm collimation, a pitchf 0.915 and a rotation time of 0.50 s. At the time of the study,

f Radiology 85 (2016) 150–157 151

iterative reconstruction algorithm was not available on eitherscanner and filtered back projection was used for image recon-struction. The effective radiation dose of MDCT angiography wasestimated by the dose-length product (DLP), as indicated on thedose report of the CT scanner, and a conversion coefficient for thechest (k = 0.014 mSv × mGy−1 × cm−1). The mean effective doseswere 2.16 ± 1.12 mSv and 1.49 ± 0.105 mSv, respectively for the128- and for the 16—detector row scanners.

Axial images were reconstructed with a medium soft-tissue ker-nel (B70 lung for Somatom Sensation 16 and L (lung) for BrillianceiCT), with a slice thickness of 2 mm and an increment of 2 mm forboth scanners. Image quality and diagnostic performance of the twoscanners were comparable and there was no impact on subjectiveimage quality and detection of CT findings of PAVM reperfusion.In our institution both CT could be used interchangeably for allapplications using a collimation of 2 mm or greater, such as in thisstudy.

Patients were evaluated in a specialized HHT center underthe supervision of one of the authors. Unenhanced chest CT wasperformed routinely 1 year after embolization. PAVMs were con-sidered to be successfully treated if there was at least a 70% decreasein size of the aneurysm sac and draining vein, as assessed subjec-tively on CT. Patients with PAVMs suspicious for recanalizationwere referred to PA and recanalized PAVMs were re-embolizedduring the same session.

2.3. Follow-up PA

Follow-up PA was performed either in the setting of emboliza-tion of new or not previously-embolized PAVMs or if at least 1previously-embolized PAVM was suspected to be reperfused withCT. Only follow-up PAs at ≥6 months after embolization were con-sidered for the assessment of PAVM reperfusion. When PA wasperformed because one PAVM was suspected to be reperfused onCT, all other PAVMs of that patient were also assessed at the sameoccasion. Most PAs were performed with an Artis Zee angio suite(Siemens, Erlangen, Germany). Selective right and left postero-anterior and contralateral anterior oblique view PAs were obtainedvia a transfemoral approach with a 5 F pigtail catheter and, in case ofdoubt, super-selective injections were also frequently performed.Iodixanol (270 mg I/mL, Visipaque 270, GE Healthcare, Princeton,NJ, USA) was injected at 20 mL/s for a total 40 mL. Follow-up PAswere reviewed by a board-certified interventional radiologist with18 years of experience, to confirm or exclude reperfusion. OccludedPAVMs were defined by the absence of contrast material flowingthrough the treated PAVM aneurysmal sac on PA. PAVM reperfu-sions were evaluated and classified as either: (a) feeding arteryrecanalization, defined as restoration of blood flow through theembolized feeding vessel, (b) collateral reperfusion of feeding arter-ies by adjacent pulmonary artery collaterals, or (c) reperfusion ofdistal feeding arteries via systemic artery branches. Systemic arte-riography was not routinely undertaken to rule out reperfusion bysystemic collaterals.

2.4. CT assessment of PAVMs

For each PAVM, aneurysmal sac, draining vein and feeding arterydiameters were measured by an experienced, board-certified inter-ventional radiologist with 20 years of experience, both on pre- andpost-embolization non-enhanced CT scans, using 2-mm axial sliceseries in lung window reformat, at the same location. Drainingveins were measured as close as possible to the aneurysmal sac,

where vessel walls were parallel, beyond any transitional dilationto the aneurysmal sac, if present. Aneurysmal sacs were measuredin their lesser diameters, perpendicular to the long axis. On pre-embolization CT scan, the feeding artery was measured proximally

152 C. Bélanger et al. / European Journal of Radiology 85 (2016) 150–157

Fig.1. Flowchart explaining which patients wer

Table 1Descriptive statistics on patients, PAVM and time interval of procedures.

Patients N = 23PAVMs N = 82

Age (y)a 49 (13.5)No. of men 7 (30%)Hereditary hemorrhagic telangiectasia 15 (65%)Patients with ≥1 symptom at presentation 18 (78%)Stroke/TIA 8 (35%)Cerebral abscess 4 (17%)Migraine 3 (13%)Dyspnea 7 (30%)Hypoxemia 3 (13%)Cyanosis 1 (4%)Pulmonary hemorrhage 1 (4%)Patients with more than 3 PAVMs 9 (39%)Number of PAVMs per patienta 3.57 (3.89)Pre-embolization diameter (mm)a

Feeding artery 3.39 (1.70)Aneurismal sac 8.78 (6.38)Draining vein 3.87 (2.09)Post-embolization diameter (mm)a

Feeding artery 1.91 (1.20)Aneurismal sac 4.43 (3.77)Draining vein 2.14 (1.21)Duration (months)a

Between pre-embolization and post-embolization CT 16.64 (14.41)Between post-embolization CT and pulmonary

angiography4.81 (11.37)

Between embolization and control pulmonaryangiography

21.46 (17.15)

Except when indicated, the data are patient numbers with percentages in parenthe-ses.TIA = Transient ischemic attack.

a Data are mean ± SD unless indicated otherwise.

e included in or excluded from the study.

to the aneurysmal sac where vessel walls were parallel, or prox-imally to any transitional dilation before the aneurysmal sac, ifpresent. On post-embolization CT scan, the feeding artery was mea-sured just distally to the embolization material, in a portion clearof beam-hardening artifacts. If artifacts impeded measurement atthat site, pre- and post-embolization feeding artery measurementswere both done immediately proximal to the embolization mate-rial. All measures were taken after optimal magnification by using apicture-archiving and communication system workstation (Impax,Agfa, Mortsel, Belgium) in a different session and blinded to follow-up PA data. To establish inter-observer variability, a board-certifiedchest radiologist with 15 years of experience independently re-measured the diameters of the 3 PAVMs structures from a randomselection of 39 PAVMs (14 patients) according to the same methodsand guidelines.

2.5. Statistical analysis

We used receiver operating characteristic (ROC) curves todemonstrate the sensitivity and specificity of non-enhanced chestCT in detecting PAVM recanalization according to varying cut-offs of percentage of diameter reduction of PAVM structures. ROCcurves for PAVMs with feeding arteries <3 mm versus ≥3 mm werealso compared. Areas under the curves (AUCs) and optimal sen-sitivity/specificity cut-offs were adjusted for verification bias byusing CT measurement of PAVM structures of all patients who hadpre- and post-embolization CT but who did not have follow-up PA[20]. ROC curves were compared by the bootstrap method with

the pROC package in R software [21]. Bland–Altman plot is used toshow the differences in percentage reduction for both observers. A2-way mixed effect model calculated intra-class correlation coef-ficients (ICCs) for 3 measures. Percentage diameter reductions of

C. Bélanger et al. / European Journal of Radiology 85 (2016) 150–157 153

Table 2Characteristics of reperfused versus occluded PAVMs.

ReperfusedPAVMsN = 43

Occluded PAVMsN = 39

p value

PAVM locationRight upper lobe 6 (14.0) 4 (10.3)Right middle lobe 6 (14.0) 2 (5.1)Right lower lobe 11 (25.6) 11 (28.2)Left upper lobe 3 (7.0) 2 (5.1)Lingula 2 (7.0) 4 (10.3)Left lower lobe 15 (34.9) 16 (41.0)Embolization materialAmplatzer vascular plugs 6 (14.0) 11 (28.2) 0.1878Detachable balloons 4 (9.3) 4 (10.3) –Coils 33 (76.7) 22 (56.4) 0.0852Amplatzer vascular plugs and coils 0 2 (5.1)PAVM with feeder <3 mm 25 (48.7) 19 (58.1) 0.5269Pre-embolization diameters (mm)a

Feeding artery 3.19 (1.62) 3.62 (1.77) 0.19Aneurismal sac 8.47 (6.79) 9.11 (5.97) 0.73Draining vein 3.58 (1.86) 4.18 (2.31) 0.32Post-embolization diameter (mm)a

Feeding artery 2.37 (1.35) 1.40 (0.75) <0.0001Aneurismal sac 5.64 (4.58) 3.10 (1.92) 0.0001Draining vein 2.44 (1.28) 1.83 (1.04) 0.0043Post-embolization diameter reduction (mm)a

Feeding artery 0.88 (1.01) 2.28 (1.72) <0.0001Aneurismal sac 2.93 (3.32) 6.26 (4.87) 0.0002Draining vein 1.14 (1.25) 2.35 (1.63) 0.0004Post-embolization diameter reduction (%)a

Feeding artery 23.73 (21.89) 56.86 (22.15) <0.0001Aneurismal sac 29.89 (25.16) 62.581 (19.14) <0.0001Draining vein 27.15 (22.35) 51.82 (22.22) <0.0001

E ses.

rs

3

3

iPpf(etTplgwwflPP

3

tlPP

xcept when indicated, the data are patient numbers with percentages in parenthea Data are mean ± SD unless indicated otherwise.

eperfused and occluded PAVMs were compared by the Wilcoxonign rank test at 5% significance level.

. Results

.1. Patients and PAVM embolizations

Twenty-six of the 68 patients who had PAVM embolization dur-ng the study period were excluded, 5 because they had complexAVMs, and 21 others, because they had missing CT, either the onere- or the one at least 6 months after embolization, or both. There-

ore, our study group included 42 patients (108 PAVMs). Nineteen45%) of these 42 patients with 26 PAVMs had both pre- and post-mbolization CT scans available but no follow-up PA and were usedo adjust sensitivities and specificities for verification bias [20].wenty-three (55%) patients with 82 PAVMs had both pre- andost-embolization CT scans available as well as follow-up PA at

east 6 months after embolization (Fig. 1). Table 1 reports the demo-raphic data and descriptive PAVM variables of these 23 patientsho had a total of 82 embolized PAVMs. All PAVM embolizationsere technically successful with complete occlusion of the PAVM

eeding artery. Embolizations were achieved with detachable bal-oons (8 PAVMs), Amplatzer vascular plugs (17 PAVMs), coils (55AVMs) or a combination of Amplatzer vascular plugs and coils (2AVMs).

.2. Follow-up PA

Post-embolization PAs were performed for additional emboliza-

ion in 52 (63%) PAVMs, or because of suspected reperfusion of ateast 1 PAVM on follow-up CT scan in 30 (37%) PAVMs. Follow-upA demonstrated that 39 (48%) PAVMs were occluded and 43 (52%)AVMs were reperfused. Of the 43 reperfused PAVMs, 34 (79%)

were re-embolized. Reperfusions were due to recanalization in 39(91%) PAVMs, to reperfusion by accessory feeding artery in 3 (7%)PAVMs, and to reperfusion by systemic collaterals in 1 (2%) PAVM.

3.3. Follow-up CT

Table 2 shows the characteristics of reperfused versus occludedPAVMs in the 23 patients who had follow-up PA. Before PAVMembolization, there were no statistically significant differencesin mean PAVM feeding artery, aneurysmal sac and draining veindiameters between PAVMs that were reperfused and those thatremained occluded. After embolization, all 3 PAVM structures werestatistically significantly smaller in occluded than in reperfusedPAVMs. However, there was no statistically significant differencein percentages of diameter reduction between the feeding artery,aneurysmal sac and draining vein either for all PAVMs, for occludedPAVMs or for reperfused PAVMs. Fig. 2 presents ROC curves ofsensitivity and specificity of non-enhanced CT to detect PAVMreperfusion, adjusted for verification bias, according to the percent-age of diameter reduction of PAVM feeding artery, aneurysmal sacand draining vein. Table 3 gives sensitivities and specificities forselected percentage diameter reduction cut-offs. Areas under theROC curves to detect PAVM reperfusion on non-enhanced CT scan,according to the percentage of diameter reduction cut-off, were0.84 (95% CI 0.78–0.94), 0.87 (95% CI 0.80–0.93), and 0.78 (95% CI0.69–0.87), respectively for PAVM feeding arteries, aneurysmal sacsand draining veins. There were no statistically significant differ-ences between the 3 PAVM structures and between PAVMs withfeeding arteries <3 mm in comparison to feeding arteries ≥3 mm.

There were no statistically significant differences in mean per-centage diameter reduction between the 2 observers for feedingartery (41.0% ± 28.7% versus 37.4% ± 26.3%, p = 0.39), aneurysmalsac (50.3% ± 28.3% versus 43.7% ± 29.6%, p = 0.28) and draining

154 C. Bélanger et al. / European Journal of Radiology 85 (2016) 150–157

Fig. 2. Receiver operating characteristic (ROC) curves, adjusted for verification bias, of PAVM diameter reduction to detect post-embolization reperfusion of (a) feeding artery,(b) aneurysmal sac, and (c) draining vein on non-enhanced CT. Shaded areas represent 95% confidence intervals. Cut-off points are the most distant points of the curve fromthe diagonal and represent optimal compromise between sensitivity and specificity.

Table 3Sensitivity and specificity of non-enhanced CT scan to detect PAVM reperfusion according to the percentage diameter reduction cut-off of PAVM structures.

Diameter Feeding artery Aneurismal sac Draining vein

reduction Se (%) Sp (%) Se (%) Sp (%) Se (%) Sp (%)

20% 35.38 89.74 41.82 97.96 39.06 90.9130% 50.77 87.18 50.91 97.96 56.25 86.3640% 69.23 82.05 60.00 91.84 65.62 75.0050% 83.08 69.23 65.45 83.67 79.69 68.1860% 93.85 58.97 83.64 67.35 85.94 50.0070% 100.00 23.08 96.36 46.94 98.44 20.45

100.0

S

va2ma

ifdfli

80% 100.00 7.69

e = sensitivity; Sp = specificity.

ein (42.5% ± 23.8% versus 34.1% ± 19.5%, p = 0.11). Averages ofbsolute percentage reduction difference between observers were1.3% ± 21.3% for the feeding artery, 13.8% ± 14.6% for the aneurys-al sac and 17.6% ± 14.6% for the draining vein (p = 0.04 for

neurysm versus feeding artery).Fig. 3 presents Bland–Altman plots of inter-observer differences

n percentage of diameter reduction against mean diameter in mmor feeding arteries, aneurysmal sacs and draining veins, with mean

ifference and 95% limits of agreement. Almost 95% of points (dif-

erences between measurements by 2 observers) lie within theimits and are equally distributed above and below the zero hor-zontal line so that no observer overestimates or underestimates

0 28.57 100.00 4.55

diameters throughout the PAVM diameter range, and there is noapparent consistent bias of one observer versus the other. No asso-ciations between variations (differences) and mean values werefound and, therefore, no systematic or proportional measurementerror.

Table 4 reports ICCs of inter-observer correlations of PAVM feed-ing artery, aneurysmal sac and draining vein diameters before andafter embolization and percentage of diameter reduction of PAVM

structures. Inter-observer correlations of diameter measurementswere better before than after embolization for all 3 structures. ICCsof diameter measurements were statistically significantly better foraneurysmal sac than for feeding artery pre-embolization (p = 0.010)

C. Bélanger et al. / European Journal of Radiology 85 (2016) 150–157 155

Fig. 3. Bland–Altman plots of inter-observer difference of percentage diameter reduction of PAVM (a) feeding artery, (b) aneurysmal sac, and (c) draining vein against meandiameter reduction after PAVM embolization. X-axis: mean of reduction in mm of observer 1 and observer 2. Y-axis: difference in percentage reduction of both observers.The dotted horizontal lines are mean ± 1.96 (SD).

Table 4Intra-class correlation coefficient (ICC) for inter-observer diameter measurements of PAVM structures.

ICCs (95% CI)a

Feeding artery Aneurismal sac Draining vein

Pre-embolization diameter 0.77 (0.57–0.88) p < 0.0001 0.93 (0.87–0.96) p < 0.0001 0.89 (0.78–0.94) p < 0.0001

aefsba

4

tWoatodw5(fioioo

Post-embolization diameter 0.49 (0.02–0.73) p = 0.02

Percentage reduction 0.58 (0.19–0.78) p = 0.005

a 95% CI = 95% confidence interval.

nd post-embolization (p = 0.006). Post-embolization ICCs of diam-ter measurements were statistically significantly lower for theeeding artery than for the draining vein (p = 0.002) and aneurysmalac (p = 0.006). ICCs for percentage of diameter reduction were alsoetter for the aneurysmal sac than for the feeding artery (p = 0.002)nd draining vein (p = 0.020).

. Discussion

Our results show wide overlap in percentage of diameter reduc-ion between occluded and reperfused PAVMs after embolization.

hichever structure is considered (feeding artery, aneurysmal sacr draining vein), any cut-off of diameter reduction will lead ton important compromise between sensitivity and specificity forhe detection of PAVM reperfusion. According to our data, cut-ffs of <30% and <70% diameter reduction of PAVM structures toetect PAVM reperfusion reported in the literature [10,11,13,14,15]ould respectively lead to low reperfusion detection sensitivities of

1–56% and specificities of 20–47%, depending on PAVM structurefeeding artery, aneurysmal sac or draining vein). The low speci-city cut-off of <70% diameter reduction will lead to greater risk

f falsely suspecting reperfusion, resulting in unnecessary further

nvestigation which could offset the cost- and risk-saving attitudef performing contrast-enhanced CT as first examination. On thether hand, the low sensitivity cut-off of <30% diameter reduction

0.83 (0.72–0.92) p < 0.0001 0.85 (0.68–0.91) p < 0.00010.88 (0.77–0.94) p < 0.0001 0.68 (0.39–0.83) p = 0.0003

will bring greater risk of falsely considering PAVMs as not reper-fused (Fig. 4).

Areas under the ROC curve were better for the PAVM aneurys-mal sac than for the feeding artery or draining vein although thedifferences were not statistically significant. In addition, no ROCcurve would be statistically significantly better should sensitivityor specificity to detect PAVM reperfusion be considered the mostimportant endpoint.

Inter-observer concordances on the diameter of all 3 PAVMstructures were acceptable before embolization but decreased onpost-embolization CT. This is due, in large part, to embolizationmaterial artifacts, as inter-observer agreement deterioration wasworst for the feeding artery and least for the draining vein. In gen-eral, artifacts were more pronounced (and therefore more likely tointerfere with PAVM reperfusion detection), with coils than withplugs, most likely due to the higher density of platinum (coils) incomparison to nitinol (plugs). Modern CT technologies such as dualenergy scanning and iterative reconstruction algorithms could helpin reducing image artifacts and possibly improving diagnostic qual-ity while keeping radiation dose within acceptable levels. However,so far no studies have demonstrated whether and to what extent

diagnostic improvement could be reached with embolization mate-rial in the pulmonary arteries. After embolization, inter-observerconcordance on the percentage of diameter reduction was unac-ceptably low for the feeding artery, fair to good for the draining

156 C. Bélanger et al. / European Journal of Radiology 85 (2016) 150–157

Fig. 4. (a) Non-enhanced chest CT before embolization demonstrates the feeding artery (white arrow) and aneurysmal sac (black arrow) of a middle lobe PAVM. Despitem n (b)

i rial fl

vrt

ptrot

scbreaeareta

kdtr

ncCdas[

ore than 50% decrease in PAVM structure diameters beyond embolization coils odentified with black arrow) reveals a reperfused malformation, with contrast mate

ein, and remained excellent for the aneurysmal sac. This bettereproducibility of aneurysmal sac diameter reduction may be dueo its larger size, which makes it easier and more precise to measure.

Areas under ROC curves and inter-observer agreement on PAVMercentage diameter reduction were better for the aneurysmal sachan for the feeding artery and draining vein. Therefore, PAVMeperfusion assessment based on percentage diameter reductionf the PAVM aneurysmal sac should probably be preferred unlesshere is a better visualization of other PAVMS structures.

Follow-up CT may highlight PAVM reperfusion, indirectly byhowing incomplete PAVM retraction, and directly by using I.V.ontrast injection [22]. Whether contrast-enhanced CT shoulde preferred over non-enhanced CT to assess PAVM reperfusionemains difficult to answer. Sensitivity and specificity of contrast-nhanced CT to detect PAVM recanalization are not reportedlthough they may be expected to be better than with non-nhanced CT because reperfusion is based on PAVM enhancementnd not on diameter reduction. However, since the complicationates of reperfused PAVM and contrast-enhanced (versus non-nhanced) CT in patients with PAVMs are not reported, it is difficulto assess whether better reperfusion detection would counterbal-nce the risks associated with contrast injection.

Beyond the clinical implications for patient management,nowledge of sensitivity and specificity of PAVM percentageiameter reduction cut-off to detect recanalization is importanto facilitate uniform scientific communication, especially wheneporting recanalization rates for various embolization devices.

Our study has some limitations, including its retrospectiveature and the relatively small sample size, limiting the statisticalomparison of ROC curves. The fact that two completely differentT scanners were used is another potential limitation of the study,

espite the fact that both scanners yielded very similar images. Inddition, because some PAVMs were referred to PA because CT wasuggestive of reperfusion, some verification bias was introduced23]. Sensitivities and specificities were adjusted for verification

post-embolization CT, (c) supra-selective digital pulmonary angiography (catheterowing into the feeding artery through the coils.

bias, but this bias and the <70% percentage diameter reductioncut-off used to detect reperfusion may explain the relatively highprevalence of PAVM reperfusion in our study. Another limitation ofthe study is that we cannot rule out that the sensitivity and speci-ficity of noncontrast CT in detectiong PAVM reperfusion may havebeen higher using a thinner slice collimation. However given thehigh intrinsic contrast resolution of chest CT, it is unlikely thatvolume averaging would have changed diameter measurementssignificantly in comparison with thinner slices. Finally, PA, ourreference standard for PAVM recanalization, is not able to showPAVM reperfusion by bronchial or other systemic arteries. How-ever, reperfusion by systemic arteries is much less frequent thanfrom pulmonary arteries and does not lead to right-left shuntingcomplications [11,12].

In conclusion, evaluation of PAVM reperfusion after emboliza-tion on non-enhanced CT should be based on aneurysmal sacdiameter reduction because of its better diagnostic accuracy andbetter level of agreement between observers than for the PAVMfeeding artery and draining vein. Our results show wide variationsof sensitivity and specificity of various PAVM diameter reductioncut-offs reported in previous publications. Selection of PAVM diam-eter reduction cut-off to detect recanalization on non-enhanced CTimplies important compromises between sensitivity and specificityand should be standardized, or at least reported, to allow compar-ison of different studies in the literature.

References

[1] S.O. Trerotola, R.E. Pyeritz, PAVM embolization: an update, Am J. Roentgenol.195 (2010) 837–845.

[2] C.L. Shovlin, J.E. Jackson, K.B. Bamford, et al., Primary determinants ofischaemic stroke/brain abscess risks are independant of severity of

pulmonary arteriovenous malformations in hereditary haemorrhagictelangiectasia, Thorax 63 (2008) 259–266.

[3] T. Haitjema, C.J. Westermann, T.T. Overtoom, et al., Hereditary hemorrhagictelangiectasia (Osler-Weber-Rendu Disease), Arch. Intern. Med. 156 (1996)714–719.

rnal o

[

[

[

[

[

[

[

[

[

[

[

[

[

Ozanne, S. Chagnon, M. El Hajjam, Diagnosis and treatment of pulmonary

C. Bélanger et al. / European Jou

[4] V. Cottin, S. Dupuis-Girod, G. Lesca, J.F. Cordier, Pulmonary vascularmalformations of hereditary hemorrhagic telangiectasia (Rendu-OslerDisease), Respiration 74 (2007) 361–378.

[5] B.A. Ference, T.M. Shannon, R.I. White Jr., M. Zawin, C.M. Burdge,Life-threatening pulmonary hemorrhage with pulmonary arteriovenousmalformations and hereditary hemorrhagic telangiectasia, Chest 106 (1994)1387–1390.

[6] A.D. Kjeldsen, P. Vase, A. Green, Hereditary haemorrhagic telangiectasia: apopulation-based study of prevalence and mortality in Danish patients, J.Intern. Med. 245 (1999) 31–39.

[7] M.E. Faughnan, V.A. Palda, G. Garcia-Tsao, et al., International guidelines forthe diagnosis and management of hereditary haemorrhagic telangiectasia, J.Med. Genet. 48 (2011) 73–87.

[8] J.S. Pollak, R.I. White Jr., Distal cross-sectional occlusion is the key to treatingpulmonary arteriovenous malformations, J. Vasc. Interv. Radiol. 23 (2012)1578–1580.

[9] J.S. Pollak, S. Saluja, A. Thabet, et al., Clinical and anatomic outcomes afterembolotherapy of pulmonary arteriovenous malformations, J. Vasc. Interv.Radiol. 17 (2006) 35–45.

10] M. Remy-Jardin, P. Dumont, P.Y. Brillet, et al., Pulmonary arteriovenousmalformations treated with embolotherapy: helical CT evaluation oflong-term effectiveness after 2–21-year follow-up, Radiology 239 (2) (2006)576–585.

11] A. Milic, R.P. Chan, J.H. Cohen, M.E. Faughnan, Reperfusion of pulmonaryarteriovenous malformations after embolotherapy, J. Vasc. Interv. Radiol. 16(2005) 1675–1683.

12] C.S. Woodward, R.E. Pyeritz, J.L. Chittams, S.O. Trerotola, Treated pulmonary

arteriovenous malformations: patterns of persistence and associatedretreatment success, Radiology 269 (2013) 919–926.

13] J. Remy, M. Remy-Jardin, L. Wattinne, C. Deffontaines, Pulmonaryarteriovenous malformations: evaluation with CT of the chest before and aftertreatment, Radiology 182 (1992) 809–816.

[

f Radiology 85 (2016) 150–157 157

14] V. Prasad, R.P. Chan, M.E. Faughnan, Embolotherapy of pulmonaryarteriovenous malformations: efficacy of platinum versus stainless steel coils,J. Vasc. Interv. Radiol. 15 (2004) 153–160.

15] D.W. Lee, R.I. White Jr., T.K. Egglin, et al., Embolotherapy of large pulmonaryarteriovenous malformations: long-term results, Ann. Thorac. Surg. 64 (1997)930–939.

16] L. Letourneau-Guillon, M.E. Faughnan, G. Soulez, et al., Embolization ofpulmonary arteriovenous malformations with Amplatzer vascular plugs:safety and midterm effectiveness, J. Vasc. Interv. Radiol. 21 (2010) 649–656.

17] K. Sagara, N. Miyazono, H. Inoue, et al., Recanalization after coilembolotherapy of pulmonary arteriovenous malformations: study oflong-term outcome and mechanism for recanalization, Am. J. Roentgenol. 170(1998) 727–730.

18] K.L. Pham, A.J. Cohen, Iatrogenic venous air embolism during contrastenhanced computed tomography: a report of two cases, Emerg. Radiol. 10(2003) 147–151.

19] R. Groell, G.J. Schaffler, R. Rienmueller, R. Kern, Vascular air embolism:location, frequency, and cause on electron-beam CT studies of the chest,Radiology 202 (1997) 459–462.

20] R. Fluss, B. Reiser, D. Faraggi, A. Rotnitzky, Estimation of the ROC curve underverification bias, Biomed. J. 3 (2009) 475–490.

21] X. Robin, N. Turck, A. Hainard, et al., pROC: an open-source package for R andS+ to analyze and compare ROC curves, BMC Bioinf. 12 (2011) 77, http://dx.doi.org/10.1186/1471-2105-12-77.

22] P. Lacombe, A. Lacout, P.Y. Marcy, S. Binsse, J. Sellier, M. Bensalah, T. Chinet, I.Bourgault-Villada, S. Blivet, J. Roume, G. Lesur, J.H. Blondel, C. Fagnou, A.

arteriovenous malformations in hereditary hemorrhagic telangiectasia: anoverview, Diagn. Interventional Imaging 94 (2013) 835–848.

23] C.B. Begg, R.A. Greenes, Assessment of diagnostic tests when diseaseverification is subject to selection bias, Biometrics 39 (1983) 207–215.