**3. Results**

Fifty-five consecutive HHT patients with a pulmonary shunt (grade > 1) on TTCE were asked to participate. Two patients declined and therefore 53 patients were included in this study representing 105 PAVMs. Forty-five PAVMs had a feeding artery diameter of ≥2 mm (42.9%). Out of these, 11 PAVMs had a feeding artery diameter of ≥3 mm. The 45 PAVMs with a FAD ≥ 2 mm were present in 24 patients (45.3%). Patients' demographics are shown in Table 1.


**Table 1.** Baseline patient characteristics.

#### *3.1. Per-Patient Analysis*

With the threshold set at a diameter of the feeding artery of ≥2 mm, R1 correctly diagnosed PAVMs in 22 patients based on CE-MRA. The absence of PAVMs was correctly diagnosed in 28 patients; there were two false negative cases and one false positive case (sensitivity 92% (73–99%) and specificity 97% (82–100%)). Reader 2 also correctly diagnosed PAVMs in 22 patients. The absence of PAVMs was correctly diagnosed in 18. Similar to R1, R2 had two false negative cases, but 11 false positive cases (sensitivity 92% (73–99%) and specificity 62% (42–79%)). Both readers correctly diagnosed PAVMs in all patients when the threshold of ≥ 3 mm was applied. Results are presented in Tables 2 and 3.


**Table 2.** Contingency table of Reader 1 and 2 per-patient analysis.

#### *3.2. Per-PAVM Analysis*

One hundred and five PAVMs were identified in a total of 954 segments (53 patients × 18 segments per patient) on CT, of which 45 PAVMs had a feeding artery diameter of ≥2 mm. R1 correctly diagnosed 43 of 45 PAVMs and had seven false positives (sensitivity 96% (85–99%) and specificity 99% (98–100%)) on CE-MRA. R2 identified 42 of 45 PAVMs and had 33 false positives (sensitivity 93% (82–99%) and specificity 96% (95–97%)). As in the per-patient analysis, all PAVMs were identified by both readers when the threshold was set at ≥3 mm. Results are presented in Tables 3 and 4.

Data are presented as number, (%) and mean ± SD; HHT = hereditary hemorrhagic telangiectasia; PAVM = pulmonary arteriovenous malformation; \* diagnosis only based on Curacao criteria.


**Table 3.** Sensitivity, Specificity, NPV and PPV of CE-MRA.

Data are in percentages, data in parenthesis are 95% confidence intervals. PAVM = pulmonary arteriovenous malformation, CE-MRA = contrast enhanced magnetic resonance angiography, NPV = negative predictive value, PPV = positive predictive value.

**Table 4.** Contingency table of Reader 1 and 2 per-PAVM analysis (18-segment model).


## **4. Discussion**

In this study, CE-MRA showed a high sensitivity and negative predictive value for the diagnosis of PAVMs with a FAD ≥ 2 mm in patients with PAVMs and a TTCE confirmed significant pulmonary shunt. PAVMs with a FAD ≥ 2 mm were diagnosed by chest CT in 45% of patients with at least a moderate pulmonary shunt on TTCE. PAVMs with a FAD ≥ 2 mm were found in 43% using CE-MRA, with a sensitivity of 92% and a NPV between 90 and 93%.

Earlier, we have shown that, in the presence of a moderate or severe shunt on TTCE, the chest CT is negative for PAVMs in 55% and 8%, respectively [13]. Chest CT has an important role in the detection and follow-up of PAVMs in HHT. Its high spatial resolution enables imaging of feeding arteries, sacs and draining veins and thereby is able to accurately diagnose and assess treatment effect of PAVMs. However, the use of chest CT comes at the expense of ionizing radiation, which is an important disadvantage in this cohort of HHT patients, who are generally young and are likely to require several follow-up CT scans during their lifetime. In addition, metal artifacts hamper evaluation of PAVMs with CT after coil embolization. MRA is not limited by radiation burden and also seems to perform well with respect to the evaluation of PAVM persistence.

In the last two decades, a handful of studies have been published reporting on the feasibility of MRI in the managemen<sup>t</sup> of PAVMs. In 2008, Schneider et al. reported a study including 203 HHT patients evaluating CE-MRA as a screening procedure for the detection of PAVMs [16]. They showed that CE-MRA is more sensitive in detecting PAVMs compared to pulmonary angiography (PA), which was the reference standard. Overall, there were 40 patients identified in whom 119 PAVMs were found with screening and follow-up using CE-MRA. Interestingly, with PA only 92 PAVMs were found (77%). As pointed out by the authors, there were important limitations in this study. First, patients with a negative CE-MRA were not compared to the reference standard. Second, there was no comparison to chest CT as the most important diagnostic tool in current algorithms. A third limitation was that the lower threshold for detection was defined as a sac diameter of at least 5 mm instead of feeding artery diameter ≥ 2 mm which is advocated in the guidelines [3].

Although not the subject in our study, MRA could also have a role in the follow-up of embolized PAVMs. Several studies have demonstrated the feasibility of (CE)-MRA for detecting PAVM persistence [17–20]. Kawai et al. used time-resolved CE-MRA to assess PAVMs and correlated with the recanalization rate on chest CT and PA [18]. The sensitivity (96%) and specificity (96%) were excellent, with good interobserver agreement. Important limitations of this study were the low number of included PAVMs (*n* = 28) and that 3 PAVMs could not be imaged by CE-MRA because of field of view limitation. For the detection of PAVM perfusion after embolotherapy, Hamamoto et al. investigated the feasibility of time-spatial labeling inversion pulse MRA (time-SLIP MRA or arterial spin labeling), a cinematic MRI method in which the inversion pulse facilitates the determination of blood flow without a contrast medium [20]. In 11 patients, 38 PAVMs were imaged with the time-SLIP MRA technique that was compared to PA as reference standard. The sensitivity and specificity of time-SLIP MRA for detecting perfusion after embolization were both 100%, with an excellent inter-observer agreement. Although very promising, time-SLIP MRA is also limited by the relatively small field of view. In this small study, four PAVMs could not be imaged because of this.

MRA is a very promising imaging method for PAVMs but there are limitations. Our study and previous studies have shown that (CE)-MRA is a feasible method to detect PAVMs. It performs well compared to chest CT in the detection of PAVMs with a FAD ≥ 2 mm. However, the number of false positives needs to be improved. The experience of the operator seems to play an important role in this, as there is a large di fference between both operators in the number of false positives. With respect to the detection of perfusion of embolized PAVMs, MRA seems to perform superior compared to CT because it is less influenced by the presence of micro coils [17,18]. This is attributable to the fact that micro coils are of platinum material which has few paramagnetic e ffects. MRA could therefore be feasible for the postembolization follow-up as well. As mentioned above, with CE-MRA and time-SLIP MRA, it is possible that the field of view is not large enough to cover the whole lung. The time-SLIP MRA technique often requires targeted imaging and thus knowledge of the location of the PAVM of interest. This means that either extra contrast administration during the MRA investigation or a chest CT scan prior to the MRA investigation will be necessary. Considering these limitations, the relative high costs and limited availability of MR imaging, it may be too early to use (CE)-MRA for routine evaluation of PAVMs.

Our study has several limitations. MRA acquisitions were performed on a 1.5 T system. With a 3 T system, higher signal-to-noise ratios can be achieved. This could result in superior image quality based on a higher spatial and temporal resolution. In addition, it enables a broader coverage of the lung reducing the risk of false negative readings. Unfortunately, a 3 T system was not available during the study. Furthermore, there was a di fference in experience in reading CE-MRA investigations between both readers. This study showed that image interpretation is reader-dependent, mainly resulting in false positives by the less experienced reader. Reader 1 has both a large experience in chest radiology and PAVM embolotherapy. He represents the level of experience that can be expected in an expert center. Reader 2, although being trained in an expert center, has less experience and could be regarded as representing nonexpert centers. HHT guidelines recommend that PAVM screening, diagnosis and treatment should be performed in expert centers. Furthermore, both readers were aware of the fact that the subjects had a positive TTCE and thus a high chance of having PAVMs with a FAD ≥ 2 mm. This could have induced a review bias lowering the threshold for diagnosing PAVMs. Finally, a limitation was the relatively small number of patients in this study. However, it is largest cohort of consecutive patients investigated so far representing a reasonably number of PAVMs with a FAD ≥ 2 mm.
