Discussion

SDAVFs can pose a challenge for both diagnosis and specific localization. On MRI, SDAVFs are characterized by an increased T2 signal within the spinal cord and conus medullaris which is thought to be related to venous hypertension and resultant cord swelling.3 ,5 This can be associated with enhancement following contrast administration, reflecting breakdown of the blood/spinal cord barrier.14 ,15 The MRI appearance can mimic various other etiologies including inflammatory or infectious myelitis and cord edema from spinal canal stenosis. The location of T2 hyperintensity within the spinal cord does not necessarily localize the site of the SDAVF since the conus is frequently involved due to orthostasis.3 ,6 Flow voids are frequently seen on T2-weighted images within the subarachnoid space in the setting of SDAVFs, representing enlarged and arterialized perimedullary venous channels. These are best seen on heavily T2-weighted myelographic sequences,16 but have also been demonstrated on CT myelography as filling defects in the intrathecal contrast pool.17 These prominent subarachnoid space venous structures are frequently extensively present longitudinally, and thus also do not localize the level of the SDAVF and arterial supply.6

Given the inability of MRI to localize SDAVFs prior to DSA, various non-invasive techniques using both CT angiography (CTA) and MRA have been attempted for this purpose. The large craniocaudal coverage of the spine and small caliber of spinal vessels makes this challenging. Studies using CTA with newer generation CT scanners have shown a modest ability to detect SDAVFs, but with a high dependence on contrast bolus timing and frequently poor differentiation of arterial and venous structures at the expense of considerable radiation exposure.18 Early attempts with non-contrast MRA techniques including phase contrast and time-of-flight showed poor ability to detect SDAVFs.19 ,20 Initial CE MRA studies offered a single phase or limited phases that could not differentiate between arteries and veins and were not able to detect SDAVF arterial feeders with confidence.10

Vargas et al used a MRA protocol with timing determined by an initial test bolus of contrast agent followed by arterial and venous phase acquisitions. The study group consisted of a variety of spinal vascular malformations but included four SDAVFs that were localized accurately.21 More recent studies using elliptic centered reordering of k-space to generate high-resolution CE MRA has shown improved performance. Farb et al used an auto-triggered technique based on a set signal intensity threshold within the aorta in nine patients with suspected DAVF to initiate optimal arterial acquisition from which a baseline acquisition was subtracted. While this technique was able to localize the DAVF in all nine patients, five were initially negative and required repeat MRA to determine the site of the fistula.7 Luetmer et al8 examined 31 patients with a suspected SDAVF using a similar technique, detecting 20 of 22 SDAVFs found on DSA and accurately determining the level of the fistula in 14. Mull et al also used a similar CE MRA technique in 34 patients, of which DSA revealed 20 SDAVFs. The MRA was concordant with DSA for all cases with the feeding artery identified correctly in 14 cases, within one segmental level in five cases, and outside the field of view in one case. The post-processing in this technique, however, was fairly time-consuming and required specific expertise to perform.9

While techniques based on elliptic centric CE MRA are useful for pre-angiographic evaluation, they lack information on contrast kinetics from multiple phases, are reliant on a test bolus or fluoroscopic triggering and may require extensive post-processing. CE-TR techniques including vendor-specific products such as TRICKS use a combination of undersampled k-space trajectories and view sharing to accelerate acquisition, allowing for much higher temporal resolution than traditional bolus-chase MRA techniques while maintaining a high spatial resolution.12 ,22 This provides temporal information and the ability to achieve an uncontaminated arterial phase without dependence on exact contrast bolus timing due to variations in cardiac output and injection rates. In the current study a 7 s temporal resolution was obtained while maintaining sufficient spatial resolution to detect shunting at the level of the radiculomeningeal arteries. The post-processing for this technique was all performed in-line at the scanner using vendor-configured settings.

Early in our experience with the CE-TR MRA technique four cases were non-diagnostic. Two were due to improper initiation of the contrast bolus prior to scanning the pre-contrast mask which resulted in a delayed and entirely venous phase study and oversubtraction of the vasculature since the pre-contrast mask contained vascular enhancement. The other two were related to excessive patient motion resulting from the sensation of contrast injection. Technologist training with regard to the proper administration of contrast and scan initiation and better patient preparation concerning the sensation during injection of the contrast bolus can help to limit these problems. The fact that one SDAVF identified on DSA was not found on CE-TR MRA underscores the importance of optimal technique.

We found a high concordance with DSA for the presence or absence of SDAVFs. All SDAVFs were detected on diagnostic quality CE-TR MRAs; however, one case with a prominent right L1 segmental artery but no definite shunting on CE-TR MRA was still felt to be suspicious for an underlying SDAVF and a DSA was performed demonstrating no abnormality. This was the only false positive case. In six of the seven cases where a SDAVF was correctly detected, CE-TR MRA detected the arterial supply within one segmental level and always on the correct side. The slight variability in levels is probably a limitation of the temporal resolution of the technique, with adjacent vertebral artery levels demonstrating similar enhancement profiles such that a fistula between two levels may be attributed to an adjacent level instead. Because of this limitation, we recommend targeting the level in question by DSA first, but also thorough examination of the segmental artery from the level above and below on the ipsilateral side. In one case a feeding artery on CE-TR MRA was not identified and was ultimately found to arise from the left internal iliac artery far outside our standard field of view. While SDAVFs commonly arise from the thoracic and lumbar segmental arteries, other arterial sources are possible and an exhaustive search by DSA or repeat CE-TR MRA should be performed with a different field of view prior to DSA.

When we were able to identify a SDAVF on CE-TR MRA, the number of vascular pedicles catheterized on DSA could be limited and the contrast dose minimized to far less than that used when no localizing information was present prior to DSA. Our analysis excluded two of the eight CE-TR MRAs positive for SDAVF; a combined diagnostic and therapeutic endovascular procedure was performed or attempted, resulting in a generally longer procedure with more contrast and fluoroscopic time than a solely diagnostic DSA would generate. The ability to plan and perform a simultaneous diagnostic and therapeutic endovascular procedure in selected cases is, in fact, a major potential benefit of performing CE-TR MRA prior to DSA.

There are several limitations to our findings. First, the retrospective nature of this study incorporates a learning curve in both the performance of the technique by technologists and the experience of the readers. During the study period an additional 18 patients did not undergo DSA due to a very low risk of SDAVF or other reasons and had normal MRAs. It is unclear how the technique would fare with all patients or whether false negative cases would arise; however, in the 3-year follow-up period of these patients there has been no indication that a SDAVF was missed. Additionally, our institution is a regional referral center for suspected vascular malformations, and it is unclear whether or not CE-TR MRA would perform similarly in practices with a lower frequency and lower pretest probability of finding a SDAVF. The detection of low-flow vascular pathology may also be limited by the CE-TR MRA technique given the 7 s temporal resolution of the protocol. Newer iterations of the acquisition employing further increases in k-space undersampling and newer reconstruction techniques promise to further improve the temporal resolution. Finally, a disadvantage of this technique is the inability to identify the Adamkievicz artery (AKA) due to limitations in spatial resolution; however, for the purpose of determining the location of the arterial supply of a SDAVF, this becomes less important as the arteries in question will undergo catheterization by DSA to exclude the AKA arising from it to determine the risk of endovascular and surgical treatment.

In conclusion, CE-TR MRA of the spine is a reliable technique for the initial evaluation of suspected SDAVFs, with high concordance with DSA. For cases in which a SDAVF is demonstrated on CE-TR MRA but a definite arterial source is not identified, the source may be outside the field of view. In cases where the CE-TR MRA is able to localize an arterial feeder, the number of catheterized vessels and total contrast dose can be decreased during DSA, with attention to the suspected level ±1 segmental levels. However, if the CE-TR MRA study does not confidently identify an arterial feeder or is of suboptimal quality, a full spinal DSA should be performed.