Recently, diagnostic imaging methods for spinal

Transcription

1 J Neurosurg Spine 18: , 2013 AANS, 2013 Three-dimensional angioarchitecture of spinal dural arteriovenous fistulas, with special reference to the intradural retrograde venous drainage system Clinical article Keisuke Takai, M.D., Ph.D., 1 Taichi Kin, M.D., Ph.D., 1 Hiroshi Oyama, M.D., Ph.D., 2 Masaaki Shojima, M.D., Ph.D., 1 and Nobuhito Saito, M.D., Ph.D. 1 1 Department of Neurosurgery, The University of Tokyo Hospital; and 2 Department of Clinical Information Engineering, Health Science Services, School of Public Health, Graduate School of Medicine, The University of Tokyo, Japan Object. There have been significant advances in understanding the angioarchitecture of spinal dural arteriovenous fistulas (AVFs). However, the major intradural retrograde venous drainage system has not been investigated in detail, including the most proximal sites of intradural radiculomedullary veins as they connect to the dura mater, which are the final targets of interruption in both microsurgical and endovascular treatments. Methods. Between April 1984 and March 2011, 27 patients with 28 AVFs were treated for spinal dural AVFs at the authors university hospital. The authors assessed vertebral levels of feeding arteries and dural AVFs by using conventional digital subtraction angiography. They also assessed 3D locations of the most proximal sites of intradural radiculomedullary veins and the 3D positional relationship between the major intradural retrograde venous drainage system and intradural neural structures, including the spinal cord, spinal nerves, and the artery of Adamkiewicz, by using operative video recordings plus 3D rotational angiography and/or 3D computer graphics. In addition, they statistically assessed the clinical results of 27 cases. Of these lesions, 23 were treated with open microsurgery and the rest were treated with endovascular methods. Results. Feeding arteries consisted of T2 10 intercostal arteries with 19 lesions, T-12 subcostal arteries with 3 lesions, and L1 3 lumbar arteries with 6 lesions. The 3D locations of the targets of interruption (the most proximal sites of intradural radiculomedullary veins as they connect to the dura mater) were identified at the dorsolateral portion of the dura mater adjacent to dorsal roots in all 19 thoracic lesions, whereas they were identified at the ventrolateral portion of the dura mater adjacent to ventral roots in 7 (78%) of 9 cases of conus medullaris/lumbar lesions (p < 0.001). The major intradural retrograde venous drainage system was located dorsal to the spinal cord in all 19 thoracic lesions, whereas it was located ventral to the spinal cord in 4 (44%) of 9 cases of conus/lumbar lesions (p = 0.006). In 3 (11%) of 27 cases, AVFs had a common origin of the artery of Adamkiewicz. In 2 lumbar lesions, the artery of Adamkiewicz ascended very close to the vein because of its ventral location. Although all lesions were successfully obliterated without major complications and both gait and micturition status significantly improved (p = and p = 0.015, respectively), conus/lumbar lesions needed careful differential diagnosis from ventral intradural perimedullary AVFs, because the ventral location of these lesions contradicted the Spetzler classification system. Conclusions. The angioarchitecture of spinal dural AVFs in the thoracic region is strikingly different from that in conus/lumbar regions with regard to the intradural retrograde venous drainage system. One should keep in mind that spinal dural AVFs are not always dorsal types, especially in conus/lumbar regions. ( Key Words spinal dural arteriovenous fistula spinal cord blood supply classification diagnosis surgical outcome complications Abbreviations used in this paper: AVF = arteriovenous fistula; CG = computer graphics; DSA = digital subtraction angiography; RA = rotational angiography. Recently, diagnostic imaging methods for spinal dural AVFs, the most common type of spinal vascular malformation, have been developed, 9,10,13,18, 22,24,29 and there have been significant advances in understanding the angioarchitecture of AVFs, which are extremely small 6,20 and have various anatomical features. 16,23,25 Spinal dural AVFs are fed by small dural arterial branches of intercostal, subcostal, lumbar, and sacral arteries. Dural arterial branches of these arteries connect directly with intradural spinal veins, piercing the spinal dura mater. 15,21,23 This means that dural AVFs exist inside the dura mater itself

2 Angioarchitecture on 3D images of spinal dural AVFs Inadequacies in intradural venous drainage play a major part in the clinical symptoms of spinal dural AVFs, such as sensory disorders, muscle weakness, and micturition/defecation problems. 17 Abnormal direct arteriovenous connections lead to spinal venous outflow obstructions due to retrograde arterial blood reflux into the intradural venous system, resulting in venous hypertension. The final goal in the treatment of spinal dural AVFs is to interrupt intradural retrograde venous drainage. As a curative and long-lasting treatment of these lesions, permanent interruption of the most proximal site of the intradural radiculomedullary vein as it connects to the dura mater is essential in both microsurgical and endovascular treatments. 1,7,14,21,32 Occlusion of only part of the arterial feeding vessels will generally lead to recurrence the development of new abnormal arteriovenous connections because arterial feeders often have many collateral vessels on the dura mater. 19,20 If permanent obliteration of all pathways of intradural venous drainage is obtained, progressive deterioration is prevented and outcomes after microsurgical or endovascular treatment are similar. 8 Although the intradural retrograde venous drainage system is the most significant target for the treatment of spinal dural AVFs, a large number of earlier studies focused mostly on the arterial system, such as the vertebral levels of feeding arteries. 4,23,27,32 To the best of our knowledge, the major intradural retrograde venous drainage system of spinal dural AVFs has not been investigated in detail. Localization of the most proximal sites of intradural radiculomedullary veins as they connect to the dura mater has not been attempted, although they are final targets of interruption in both microsurgical and endovascular treatments. Also, anatomical variations at each spinal level of the angioarchitecture of major intradural draining veins remain obscure. The purpose of this study was to analyze the angioarchitecture of spinal dural AVFs stereoscopically, with special reference to the intradural retrograde venous drainage system, by means of conventional examination including DSA and operative video recordings as the gold standards, with advanced technologies consisting of 3D RA and 3D CG. Methods Between April 1984 and March 2011 at the University of Tokyo Hospital, 27 patients whose ages ranged from 37 to 88 years (Table 1) were treated for spinal dural AVFs. Because 1 of the 27 patients had 2 independent AVFs, AVFs in 27 patients were included in this study. Dural AVFs at the craniocervical junction and those in the sacral region were excluded. Spinal extradural AVFs with intradural retrograde venous drainage, which are generally similar to dural AVFs in the lumbar region, 30 were also excluded. The 28 AVFs consisted of 19 in the thoracic region (T2 10) and 9 in the conus medullaris/lumbar regions (T12 L3). All 28 AVFs were diagnosed using spinal DSA. In 11 of the 28 AVFs 3D RA was performed, and 9 AVFs were assessed using 3D CG based on the fusion of 3D RA and postmyelographic CT, which we had recently developed. The 3D CG technology provides accurate, simultaneous, and stereoscopic visualization of the spinal vascular system, spinal cord, dura mater, and bone. 29 In our institution, the first-line choice of treatment for spinal AVFs was direct microsurgical ligation of the lesions and the second-line treatment was endovascular embolization, because direct surgery is superior to endovascular treatment, with few clinical or radiographic recurrences. 26,32 Of the 28 AVFs, 23 were treated with surgery and the rest were treated with endovascular embolization because of the advanced age of patients or their own treatment choices. First, we analyzed the angioarchitecture of both arterial and venous systems of spinal dural AVFs. We assessed the vertebral levels of feeding arteries and dural AVFs and the number and flow directions of major intradural draining veins by using DSA as the gold standard. Also, we assessed 3D locations of the most proximal sites of intradural radiculomedullary veins as they connect to the dura mater, and the 3D positional relationship between the major intradural retrograde venous drainage system and intradural neural structures (including the spinal cord, spinal nerves, and artery of Adamkiewicz), by means of intraoperative video recordings as the gold standard, plus 3D RA and/or 3D CG. We statistically compared the intradural locations of the major intradural retrograde venous drainage system of the 19 AVFs in the thoracic region (T2 10) and the 9 in the conus/lumbar regions (T12 L3). In addition, we assessed radiological and clinical results in the 27 cases, including complications. To assess radiological results, we compared pre- and postoperative DSA and MRI findings. To assess clinical results, we statistically compared pre- and postoperative Aminoff- Logue disability grades for gait and micturition. 3 In 23 AVFs treated with open surgery, we statistically compared the range of laminectomy between cases diagnosed by 2D DSA and those diagnosed by 3D RA and/or 3D CG. The Fisher exact test and Wilcoxon signed-rank test were used as indicated for statistical analysis of the data. We used SPSS II for Windows (SPSS Japan, Inc.) statistical software for statistical analysis. A p value < 0.05 was considered significant. This study protocol (#2231) was approved by the institutional review board at the University of Tokyo Hospital. Because this was a retrospective and noninvasive study, written patient informed consent was not obtained. Instead, public notice of information about this study was given through the University of Tokyo Hospital s website. Results Detailed information on the angioarchitecture of spinal dural AVFs is provided in Table 1 and summarized in Table 2. Spinal Levels of Feeding Arteries on 2D Imaging Feeding arteries consisted of dural branches of T2 10 intercostal arteries (19 lesions), T-12 subcostal arteries (3 lesions), and L1 3 lumbar arteries (6 lesions) (Table 1). Twenty-eight AVFs were divided into 2 groups according to the vertebral levels of feeding arteries: 19 thoracic 399

3 K. Takai et al. TABLE 1: Detailed information on the angioarchitecture of spinal dural arteriovenous fistulas in 27 patients* Case No. Age (yrs), Sex Imaging Treatment Laminectomy Feeder Most Proximal Sites of Draining Veins Intradural Draining Venous System (flow direction) Artery of Adamkiewicz 1 55, M DSA open T1 2 rt supreme IA dorsolat portion at T-2 dorsal RMV, dorsal SCV ( ) lt T , M DSA, 3D CG open T-4 rt T-4 IA dorsolat portion at T-4 dorsal RMV, dorsal SCV ( ) ND 3 55, F DSA open T4 8 rt T-5 IA dorsolat portion at T-6 dorsal RMV, dorsal SCV ( ) ND 4 37, M DSA open T5 7 rt T-6 IA dorsolat portion at T-6 dorsal RMV, dorsal SCV ( ) lt T , F DSA open T5 7 rt T-6 IA dorsolat portion at T-6 dorsal RMV, dorsal SCV ( ) lt T , M DSA open T5 7 rt T-6 IA dorsolat portion at T-6 dorsal RMV, dorsal SCV ( ) lt T , M DSA open T4 6 lt T-6 IA dorsolat portion at T-6 dorsal RMV, dorsal SCV ( ) ND 8 57, M DSA endo lt T-6 IA dorsolat portion at T-6 dorsal RMV, dorsal SCV ( ) lt T , M DSA, 3D CG open T6 7 lt T-6 IA dorsolat portion at T-6 dorsal RMV, dorsal SCV ( ) ND 10 61, M DSA, 3D CG open T6 7 rt T-6 & T-7 IA dorsolat portion at T-6 dorsal RMV, dorsal SCV ( ) lt T , M DSA open T7 10 lt T-7 IA dorsolat portion at T-7 dorsal RMV, dorsal SCV ( ) lt T , M DSA, 3D CG open T-7 rt T-7 & T-8 IA dorsolat portion at T-7 dorsal RMV, dorsal SCV ( ) ND 57, M DSA, 3D CG open T11 12 lt T-12 SA dorsolat portion at T-12 dorsal RMV, dorsal SCV ( ) ND 13 65, M DSA open T7 8 rt T-8 IA dorsolat portion btwn T-7 & T-8 dorsal RMV, dorsal SCV ( ) rt T , M DSA, 3D CG open T-8 lt T-8 IA dorsolat portion btwn T-8 & T-9 dorsal RMV, dorsal SCV ( ) ND 15 53, M DSA open T7 9 rt T-8 IA dorsolat portion at T-8 dorsal RMV, dorsal SCV ( ) lt L , F DSA, 3D CG endo lt T-8 IA dorsolat portion at T-8 dorsal RMV, dorsal SCV ( ) lt L , M DSA endo rt T-8 IA dorsolat portion at T-7 dorsal RMV, dorsal SCV ( ) lt T , M DSA, 3D CG open T-10 lt T-10 IA dorsolat portion at T-10 dorsal RMV, dorsal SCV ( ) lt T , M DSA open T9 10 rt T-10 IA dorsolat portion at T-10 dorsal RMV, dorsal SCV ( ) lt T , M DSA, 3D RA open T12 L1 lt T-12 SA ventrolat portion at T-12 ventral RMV, ventral SCV ( ) ND 21 59, F DSA open T12 L1 rt T-12 SA ventrolat portion at L-1 ventral RMV, dorsal SCV ( ) ND 22 69, M DSA open T12 L1 rt L-1 LA ventrolat portion at L-1 ventral RMV, dorsal SCV ( ) lt T , F DSA open L1 2 lt L-1 LA ventrolat portion at L-1 ventral RMV, ventral SCV ( ) lt L , F DSA open L1 3 lt L-1 LA ventrolat portion at L-1 ventral RMV, ventral SCV ( ) lt L , M DSA open L1 2 lt L-1 LA dorsolat portion at L-2 dorsal RMV, dorsal SCV ( ) lt L , M DSA, 3D RA endo rt L-3 LA ventrolat portion at L-3 ventral RMV, dorsal SCV ( ) lt L , F DSA, 3D CG endo lt L-3 LA ventrolat portion at L-3 ventral RMV, ventral SCV ( ) lt T-11 * endo = endovascular; IA = intercostal artery; LA = lumbar artery; ND = not detected; RMV = radiculomedullary vein; SA = subcostal artery; SCV = spinal cord vein. As they connect to dura mater. Lesions with a discrepancy between the vertebral levels of feeding arteries and those of the most proximal sites of intradural RMVs. This patient had 2 AVFs. The artery of Adamkiewicz had a common origin with lesions. 400

4 Angioarchitecture on 3D images of spinal dural AVFs TABLE 2: Summary of the angioarchitecture of spinal dural AVFs* Arrangement & Distribution of Vessels No. of Thoracic AVFs (%) No. of Conus & Lumbar AVFs (%) total no. of AVFs 19 9 vertebral levels of feeders T2 10 T12 L3 feeding arteries dural branch of the IA 19 (100) 0 dural branch of the SA 0 3 (33) dural branch of the LA 0 6 (67) locations of the most proximal sites of intradural RMVs dorsolat portion of the dura mater 19 (100) 2 (22) ventrolat portion of the dura mater 0 7 (78) major intradural draining venous system (flow direction) dorsal RMV, dorsal SCV 15 (79) 0 dorsal RMV, dorsal SCV 4 (21) 0 dorsal RMV, dorsal SCV 0 2 (22) ventral RMV, dorsal SCV 0 3 (33) ventral RMV, ventral SCV 0 4 (44) AVFs having a common origin w/ the artery of Adamkiewicz 1 (5) 2 (22) * Values are the number of dural AVFs, with percentages in parentheses. Ventral locations of the most proximal sites of intradural RMVs as they connect to the dura mater were more commonly observed in patients with AVFs in conus/lumbar regions (p < 0.001, Fisher exact test). The major intradural draining venous system was commonly located ventral to the spinal cord in patients with AVFs in conus/ lumbar regions (p = 0.006, Fisher exact test). AVFs fed by T2 10 intercostal arteries and 9 conus/lumbar AVFs fed by T-12 subcostal and L1 3 lumbar arteries, respectively (Table 2). Locations of the Most Proximal Sites of Intradural Radiculomedullary Veins on 3D Imaging The 3D locations of the most proximal sites of intradural radiculomedullary veins as they connect to the dura mater were identified at dorsolateral portions of the dura mater adjacent to dorsal roots in all 19 lesions in the thoracic region (Fig. 1A). On the other hand, they were identified at ventrolateral portions of the dura mater adjacent to ventral roots in 7 (78%) of 9 lesions in the conus/ lumbar regions (Fig. 1B). In only 2 of 9 lesions in the conus/lumbar regions were these veins identified at dorsolateral portions of the dura mater adjacent to dorsal roots. Ventral locations of the most proximal sites of intradural radiculomedullary veins were more commonly observed in patients with AVFs in the conus/lumbar regions than in the thoracic region (p < 0.001, Fisher exact test). There were some lesions with a discrepancy between vertebral levels of feeding arteries and those of the proximal site of intradural radiculomedullary veins as they connect to the dura mater; in 4 (14%) of the 28 AVFs, the most proximal sites of intradural radiculomedullary veins were 1 vertebral level apart from their feeding arteries through collateral vessels on the dura mater (Fig. 2B and D). Positional Relationship Between the Major Intradural Retrograde Venous Drainage System and Intradural Neural Structures on 3D Imaging The major intradural retrograde venous drainage system was located dorsal to the spinal cord in all 19 lesions in the thoracic region (Fig. 1C). On the other hand, in 4 (44%) of 9 lesions in the conus/lumbar regions, the major intradural radiculomedullary veins were located ventral to the spinal cord (Fig. 1D). The major intradural draining venous system was commonly located ventral to the spinal cord in patients with AVFs in the conus/lumbar regions compared with the thoracic region (p = 0.006, Fisher exact test). In anteroposterior angiograms, major intradural draining veins showed 2 kinds of retrograde draining flow patterns in thoracic lesions and a single kind of flow pattern in conus/lumbar lesions (Fig. 2). In 15 (79%) of 19 lesions in the thoracic region, the most proximal site of a vein was single and the vein descended retrogradely into the single dorsal spinal cord vein (Fig. 2A). In the other 4 (21%) of the 19 lesions in the thoracic region, the most proximal site of a vein was single, and the vein divided into 2 dorsal spinal cord veins, one ascending and the other descending. In all conus/lumbar lesions, the most proximal site of a vein was single and the radiculomedullary vein ascended retrogradely into spinal cord veins. The radiculomedullary vein was long in proportion to the nerve root length of the cauda equina (Fig. 2C). In lateral angiograms obtained in conus/lumbar lesions, major intradural draining veins showed 3 kinds of retrograde draining flow patterns (Fig. 3). As shown in Table 2, in 4 of the 7 AVFs in which the most proximal sites of radiculomedullary veins were identified at ventrolateral portions of the dura mater, veins ascended retrogradely into ventral spinal cord veins (Fig. 3A). In the other 3 of these 7 AVFs, veins ventral to the cauda equi- 401

5 K. Takai et al. Fig. 1. A and C: Case 18. Three-dimensional CG images of T-10 thoracic AVF showing the most proximal site of the intradural radiculomedullary vein located at the dorsolateral portion of the dura mater (arrow) and the major intradural retrograde venous drainage system located dorsal to the spinal cord (arrowheads). B and D: Case 27. Three-dimensional CG images of L-3 lumbar AVF showing the most proximal site of the intradural radiculomedullary vein located at the ventrolateral portion of the dura mater (arrow) and the major intradural retrograde venous drainage system located ventral to the spinal cord and cauda equina (arrowheads). Feeding artery (red), draining vein (purple), dura mater (green), spinal cord (yellow), and bone (translucent). na ascended retrogradely into dorsal spinal cord veins through extrinsic and intrinsic anastomoses (Fig. 3B). In only 2 AVFs in which the most proximal sites of draining veins were identified at dorsolateral portions of the dura mater, the vein ascended retrogradely into dorsal spinal cord veins (Fig. 3C). Feeders of the artery of Adamkiewicz were identified in 19 of 27 cases by using DSA (Table 1). In the thoracic region, 1 (5%) of 19 AVFs had a common origin with the artery of Adamkiewicz (Case 18). In this case, because the most proximal site of the intradural radiculomedullary vein was located at the dorsolateral portion of the dura mater adjacent to the dorsal root, the artery of Adamkiewicz running adjacent to the ventral root was apart from the vein, which was successfully interrupted in open microsurgery. On the other hand, in conus/lumbar regions, 2 (22%) of 9 AVFs had a common origin with 402 Fig. 2. A: Case 4. A DS angiogram of the T-6 intercostal artery showing the descending intradural spinal cord vein. The most proximal portion of the radiculomedullary vein as it connects to the dura mater (arrow) is at the same level as the feeding artery. B: Case 17. A DS angiogram of the T-8 intercostal artery showing the ascending dural collateral vessel between T-8 and T-7 (arrowheads). Note a discrepancy between the proximal site of the radiculomedullary vein as it connects to the dura mater (arrow) and the feeding artery. C: Case 22. A DS angiogram of the L-1 lumbar artery showing the ascending radiculomedullary vein draining into the spinal cord vein. The most proximal portion of the radiculomedullary vein as it connects to the dura mater (arrow) is at the same level as the feeding arteries. D: Case 21. A DS angiogram of the T-12 subcostal artery showing the descending dural collateral vessel between T-12 and L-1 (arrowheads). Note a discrepancy between the proximal site of the radiculomedullary vein as it connects to the dura mater (arrow) and the feeding artery. the artery of Adamkiewicz (Case 23 and 24). In these 2 cases, because the most proximal sites of intradural radiculomedullary veins were located at ventrolateral portions of the dura mater adjacent to the ventral roots, the artery of Adamkiewicz adjacent to the ventral roots ascended very close to the vein (Fig. 3D F). Radiological and Clinical Results Twenty-seven cases were evaluated in the postoperative period with angiography and/or MRI. Postoperative angiography was performed within 1 month in 19 (68%)

6 Angioarchitecture on 3D images of spinal dural AVFs Fig. 3. A C: Two-dimensional lateral DS angiograms showing 3 kinds of flow patterns of the intradural retrograde venous drainage system in conus medullaris/lumbar AVFs. D F: Case 23. Dural AVF at the lumbar level. A: A DS angiogram of lumbar AVF (L-1) showing the most proximal site of the draining vein located ventral to the cauda equina (black arrow) and the ventral radiculomedullary vein (arrowheads) ascending into ventral spinal cord veins (white arrows). Note that 4 (44%) of 9 cases of conus/lumbar AVFs have this drainage pattern. B: A DS angiogram of lumbar AVF (L-1) showing the most proximal site of the draining vein located ventrally (black arrow) and the ventral radicular vein (arrowheads) ascending into the dorsal spinal cord vein through extrinsic anastomoses (white arrows). Note that 3 (33%) of 9 cases of conus/lumbar AVFs have this drainage pattern. C: A DS angiogram of lumbar AVF (L-2) showing the most proximal site of the draining vein located dorsally (black arrow) and the dorsal radiculomedullary vein (arrowheads) ascending into the dorsal spinal cord vein (white arrows). Note that 2 (22%) of 9 cases of conus/lumbar AVFs have this drainage pattern. D: An intraoperative photograph showing the most proximal site of the intradural radiculomedullary vein (black arrow) located at the ventrolateral portion of the dura mater, ascending adjacent to the ventral roots (white arrowheads). Note that the artery of Adamkiewicz ascends very close to the vein because of its ventral location (black arrowheads). E: Preoperative 2D DS angiogram showing the AVF (black arrow) between the left L-1 lumbar artery and intradural veins, draining into ascending spinal cord veins (white arrowheads). F: Postoperative 2D DS angiogram showing the artery of Adamkiewicz (arrowheads). of the 28 AVFs, and angiographic obliteration was verified in all 19 lesions. Postoperative MRI was performed in 20 (74%) of 27 cases, and the extent of hyperintensity of the spinal cord on T2-weighted images decreased 5 to 0.7 vertebral levels on average in all 20 cases. Detailed information on clinical results and complications are provided in Table 3 and Fig. 4. During the postoperative follow-up period (mean 29 months, range months), Aminoff-Logue grades for gait improved in 14 cases (52%), were unchanged in 12 cases (44%), and got worse in 1 case (4%). Aminoff-Logue micturition grades improved in 7 cases (26%) and were unchanged in 20 cases (74%). Both gait and micturition grades significantly improved (p = and p = 0.015, respectively; 403

7 K. Takai et al. TABLE 3: Summary of neurological status at presentation, follow-up duration, and operative complications* Aminoff-Logue Gait Grade Aminoff-Logue Micturition Grade Presentation FU Presentation FU Case No. Preop Sx (mos) FU (mos) Op Complications CSF fistula requiring 2nd op none none venous thrombosis suspected none none none none none none wrong-level laminectomy none none none none none none none none none none none none none none none none * Both Aminoff-Logue gait and micturition grades significantly improved (p = for gait and p = for micturition; Wilcoxon signed-rank test). Abbreviations: FU = follow-up; Sx = symptoms. Wilcoxon signed-rank test). Gait status tended to improve more than micturition status after the treatment, even in some wheelchair-bound patients with more than 2 years symptom duration (Fig. 4). Three perioperative surgical complications occurred in patients with thoracic lesions (Table 3). The patient in Case 1 developed a CSF fistula, which was resolved after the second repair surgery. The patient in Case 4 developed worsening of paraparesis due to suspected idiopathic venous thrombosis. In Case 11 laminectomy was unsuccessful, although the spinal levels of the feeding artery and intradural radiculomedullary vein were the same. This patient underwent multiple-level laminectomy (a total of 4 spinal levels), and the radiculomedullary vein was successfully interrupted. In 9 cases of conus/lumbar AVFs, 7 were treated with open microsurgery and 2 were treated with endovascular embolization. In 5 of the 7 cases treated with open surgery, the most proximal sites of the intradural radiculomedullary vein were identified at the ventrolateral portion of the dura mater. During surgery, localization of these veins required more scrupulous attention in conus/lumbar AVFs than in thoracic AVFs because they were located under a number of roots of the cauda equina. Furthermore, in 2 of these 5 cases, the artery of Adamkiewicz was identified very close to the vein (Fig. 3D F). After careful localization of intradural radiculomedullary veins and differential diagnosis between them and the artery of Adamkiewicz, we successfully interrupted these veins without any complications. In 2 cases treated with endovascular embolization, ventral localization and the ascending flow pattern of abnormal vessels became an issue because the differential diagnosis of these vessels was that they were either arterial feeders of ventral intradural perimedullary AVFs or veins of dural AVFs with retrograde drainage. Before embolization, we administered a small amount of 2% lidocaine into the vessels to confirm that they were veins and that this would not cause paraparesis. After careful differential diagnosis between these vessels and the artery of Adamkiewicz, we successfully obliterated them with N-butyl 2-cyanoacrylate without any complications. 404

8 Angioarchitecture on 3D images of spinal dural AVFs Fig. 4. Graphs showing postoperative changes in Aminoff-Logue gait (upper) and micturition (lower) grades. Both gait and micturition grades significantly improved (p = 0.005, p = 0.015, respectively; Wilcoxon signed-rank test). Each pair of bars represents 1 patient. In 4 (14%) of 28 AVFs, the most proximal sites of intradural radiculomedullary veins were 1 vertebral level apart from their feeding arteries through collateral vessels on the dura mater. Of these 4 AVFs, 3 were treated with open microsurgery and the last was treated with endovascular embolization. After careful localization of the proximal sites of intradural radiculomedullary veins, we successfully interrupted them without any complications. In the 23 cases treated with open surgery, the range of laminectomy was significantly fewer levels in patients diagnosed by 3D RA and/or 3D CG than in those diagnosed by 2D DSA only (p = 0.008, Fisher exact test; Table 4). Illustrative Case Case 23 This 68-year-old woman presented with numbness in both feet, gait disturbance, and difficulty in urinating that had progressed over 38 months. The Aminoff-Logue gait grade was 5 and the micturition grade was 3. A DSA study showed an AVF between the left L-1 lumbar artery and the intradural radiculomedullary vein, draining retrogradely into ascending spinal cord veins with a common origin to the artery of Adamkiewicz (Fig. 3E). During surgery, the most proximal site of the intradural radiculomedullary vein as it connected to the dura mater was identified at the ventrolateral portion of the dura mater, and the major intradural retrograde venous drainage system was located ventral to the spinal cord and cauda equina (Fig. 3D). In this case, the artery of Adamkiewicz was identified very close to the vein because of its ventral location adjacent to the ventral roots. The radiculomedullary vein was successfully interrupted by direct surgery without any complications (Fig. 3F). The patient gradu- 405

9 K. Takai et al. TABLE 4: Range of laminectomy in patients treated with open microsurgery* Diagnosis Range of Laminectomy 1 Level 2 Levels >2 Levels Total 2D DSA D CG/RA total * Values are the number of dural AVFs. The range of laminectomy levels was significantly fewer in patients diagnosed using 3D RA and/or 3D CG than in those diagnosed using 2D DSA only (p = 0.008, Fisher exact test). ally regained neurological function after surgery, improving from Aminoff-Logue gait Grade 5 to 3 and from micturition Grade 3 to 2, despite a long symptom duration preoperatively. Discussion The major finding of our study was that the angioarchitecture of spinal dural AVFs in the thoracic region is strikingly different from that in the conus/lumbar regions with regard to the intradural retrograde venous drainage system. Differences were that the venous drainage system included intradural radiculomedullary veins that were located dorsally (dorsolaterally) in all thoracic AVFs, whereas they were usually located ventrally (ventrolaterally) in conus/lumbar AVFs (p < 0.001; Fig. 1 and Table 2). Also, 22% of cases of conus/lumbar AVFs had a common origin with the artery of Adamkiewicz, adjacent to the ventral roots and ascending very close to the vein (Table 2). In addition, we found that 14% of all thoracic and conus/lumbar lesions had a discrepancy between vertebral levels of feeding arteries and those of the most proximal site of intradural radiculomedullary veins (Fig. 2 and Table 1). To the best of our knowledge these anatomical differences have yet to be described in the literature. 4,23,27,32 We previously introduced a new diagnostic method for spinal vascular malformations: 3D CG. 17 In this stereotactic visualization system, we noticed that not all of the intradural retrograde venous drainage system of spinal dural AVFs was located at the dorsal (dorsolateral) region. Therefore, we developed the present study, in which the angioarchitecture of spinal dural AVFs was stereoscopically analyzed in detail. Based on the morphological findings of dural AVFs, we considered the following 3 issues: 1) the apparent discrepancy with the contemporary classification system; 2) insights into morphological differences between thoracic and conus/lumbar AVFs; and 3) clinical implications in the treatment of spinal dural AVFs. Critical Comparisons to the Contemporary Classification System Although we have shown in the present study that the venous retrograde drainage system is commonly located dorsally (dorsolaterally) in thoracic AVFs and ventrally (ventrolaterally) in conus/lumbar AVFs, recent studies published by Spetzler and colleagues claimed that spinal dural AVFs are intradural dorsal lesions. 16,25 Spinal dural AVFs were formerly known as Type I AVFs, and have recently been modified as dorsal intradural AVFs according to the Spetzler classification system because they are commonly considered to exist dorsally to the spinal cord and nerves. 16,25 In this classification system, lesions formerly known as perimedullary AVFs or Type IV AVFs with high-flow direct AVFs on the surface of the spinal cord, 5,31 not inside the dura mater, are classified as ventral intradural AVFs. This classification system is not always consistent with our findings. We believe our study to be superior to the aforementioned ones for the following reasons. To our knowledge, these investigators have never substantiated the angioarchitecture of spinal dural AVFs by statistical analysis of angiographic or objective operative data. Our study is supported by a great deal of objective and statistical evidence not only from conventional examinations such as DSA and intraoperative video recordings, but also from advanced technologies such as 3D RA and 3D CG. To analyze the vascular system in detail, we performed interactive manipulations such as rotation, enlargement, and reduction of these 3D images, in which any tissue could be made transparent and virtual-reality operations including bone elimination were available. 29 In our conus/lumbar cases, the ventral location of intradural vascular lesions became an issue in the differential diagnosis between these lesions and perimedullary AVFs (ventral intradural AVFs in the Spetzler classification) because spinal dural AVFs were considered to be dorsal lesions, whereas perimedullary AVFs were ventral lesions. Ventral spinal cord and radiculomedullary veins of spinal dural AVFs are sometimes similar to anterior spinal and radiculomedullary arteries in terms of intradural location and blood flow direction. Knowledge of the common ventral location of the intradural retrograde venous system of conus/lumbar AVFs yields important information on the differential diagnosis between these lesions and other spinal arteriovenous entities, which require different treatment strategies. The classification system of spinal arteriovenous lesions should be updated according to the latest scientific evidence. Insights Into Morphological Differences Between Thoracic and Conus/Lumbar AVFs Morphological differences in the proximal site of perimedullary venous drainage between thoracic and conus/lumbar AVFs may be explained by anatomical features of the normal spinal venous system, because spinal dural AVFs are considered to be acquired lesions; thus, they may reflect the angioarchitecture of the normal spinal venous system. 19,28 Morphological features of normal spinal veins vary according to their level in the spine. In the thoracic region from T-2 to T-10, normal dorsal and ventral spinal cord veins ascend to carry venous outflow from the spinal cord and drain laterally toward the dura mater via short dorsal and ventral radiculomedullary veins, respectively. Dorsal and ventral radiculomedullary veins meet to form a single vessel, just before they reach the dura mater. 28 There, they pierce the dorsolateral por- 406

10 Angioarchitecture on 3D images of spinal dural AVFs tion of the dura mater and drain into the internal vertebral plexus. The dorsal location of thoracic spinal AVFs is most likely related to these normal thoracic venous arrangements. In conus/lumbar regions from T-10 to the sacrum, normal dorsal and ventral spinal cord veins ascend to carry venous outflow from the spinal cord up to T-10. There, they do an inverted U-turn and descend obliquely toward the dura mater via long dorsal and ventral radiculomedullary veins, respectively. Dorsal radiculomedullary veins reach and pierce the dorsolateral portion of the dura mater, whereas ventral radiculomedullary veins pierce the ventrolateral portion of the dura mater and drain into the internal vertebral plexus. 11 Especially in conus/lumbar regions, normal dorsal and ventral spinal cord veins have rich extra- and intramedullary vascular anastomoses; 19 thus, dorsal or ventral radiculomedullary veins may be missing, in which case they are compensated by each other. 28 The ventral location of conus/lumbar AVFs may be related to the predominance of ventral over dorsal locations for normal radiculomedullary venous drainage. In conus/lumbar AVFs, the blood flow pattern fits the angioarchitecture of normal spinal veins in conus/lumbar regions. Normal radiculomedullary veins have longer structures than those in thoracic regions in proportion to the nerve root length of the cauda equina (Fig. 2C). The ascending flow direction of the intradural retrograde venous drainage system of spinal dural AVFs is directly opposite to that of normal spinal venous outflow because of abnormal arterial blood reflux, resulting in venous hypertension. Clinical Implications in the Treatment of Spinal Dural AVFs Although many cerebral dural AVFs are now best treated with endovascular embolization, we believe that spinal dural AVFs are better suited to open microsurgery because of fewer clinical or radiographic recurrences than with endovascular embolization. 26,32 The present study raises some important points from a surgical perspective. In microsurgical treatment, we first need to recognize that spinal dural AVFs are not always dorsal types, especially in conus/lumbar regions (Figs. 1 and 3, Table 2). Also, we need to recognize that the artery of Adamkiewicz sometimes has a common origin with AVFs in lower thoracic, conus, and lumbar regions. 27 In our case series of 27 patients, we successfully obliterated intradural radiculomedullary veins as they connect to the dura mater, without major complications. During the follow-up period, both gait and micturition grades improved significantly (Table 3). However, although it is not extremely difficult, we should carefully localize the proximal sites of intradural radiculomedullary veins in conus/lumbar lesions more than in thoracic lesions, because the former are commonly located ventrolaterally to the spinal cord and/or cauda equina. Furthermore, more than 20% of cases of conus/ lumbar AVFs have a common origin with the artery of Adamkiewicz, ascending very close to veins because of the ventral location (Fig. 3D). During surgery, a relevant differential diagnosis of these vessels should be carefully made because such lesions have a potentially high risk of spinal cord infarction due to arterial occlusion. 2 Last, we should focus attention on some lesions with a discrepancy between the vertebral levels of feeding arteries and those of intradural radiculomedullary veins as they connect to the dura mater (Fig. 2, Table 1). Accurate spinal laminectomy is essential in these lesions. More than 3-level laminectomy has been used to treat spinal dural AVFs; 1,21 however, our study shows that the proximal portion of radiculomedullary veins can be exposed and interrupted by limited laminectomy in patients in whom 3D RA and/or 3D CG are used for diagnosis, relative to those in whom 2D DSA only is used (p = 0.008). In the proper diagnosis and treatment of spinal dural AVFs it is important to focus on the 3D angioarchitecture, with special reference to the intradural retrograde venous drainage system. This study has some limitations; it was limited to a small series of patients, and assessment of the angioarchitecture of the major intradural retrograde venous drainage system was based on the judgments of our own investigators. Therefore, further studies are needed in a larger series of patients. Conclusions The angioarchitecture of spinal dural AVFs is strikingly different according to their level in the spine, particularly for the major intradural retrograde venous drainage system. Disclosure The authors declare that they received no financial or material support for this research. Author contributions to the study and manuscript preparation include the following. Conception and design: Takai. Acquisition of data: Takai, Kin, Shojima. Analysis and interpretation of data: Takai, Kin, Shojima. Drafting the article: Takai. Critically revising the article: all authors. Reviewed submitted version of manuscript: all authors. Approved the final version of the manuscript on behalf of all authors: Takai. Statistical analysis: Takai. Study supervision: Saito. References 1. Afshar JK, Doppman JL, Oldfield EH: Surgical interruption of intradural draining vein as curative treatment of spinal dural arteriovenous fistulas. J Neurosurg 82: , Aggarwal S, Willinsky R, Montanera W, Terbrugge K, Wallace MC: Superselective angiography of a spinal dural arteriovenous fistula having a common segmental origin with the artery of Adamkiewicz. Neuroradiology 34: , Aminoff MJ, Logue V: The prognosis of patients with spinal vascular malformations. Brain 97: , Atkinson JL, Miller GM, Krauss WE, Marsh WR, Piepgras DG, Atkinson PP, et al: Clinical and radiographic features of dural arteriovenous fistula, a treatable cause of myelopathy. Mayo Clin Proc 76: , Barrow DL, Colohan AR, Dawson R: Intradural perimedullary arteriovenous fistulas (type IV spinal cord arteriovenous malformations). J Neurosurg 81: , Benhaiem N, Poirier J, Hurth M: Arteriovenous fistulae of the meninges draining into the spinal veins. A histological study of 28 cases. Acta Neuropathol 62: , Britz GW, Lazar D, Eskridge J, Winn HR: Accurate intraop- 407

11 K. Takai et al. erative localization of spinal dural arteriovenous fistulae with embolization coil: technical note. Neurosurgery 55: , Dehdashti AR, Da Costa LB, terbrugge KG, Willinsky RA, Tymianski M, Wallace MC: Overview of the current role of endovascular and surgical treatment in spinal dural arteriovenous fistulas. Neurosurg Focus 26(1):E8, Eddleman CS, Jeong H, Cashen TA, Walker M, Bendok BR, Batjer HH, et al: Advanced noninvasive imaging of spinal vascular malformations. Neurosurg Focus 26(1):E9, Gilbertson JR, Miller GM, Goldman MS, Marsh WR: Spinal dural arteriovenous fistulas: MR and myelographic findings. AJNR Am J Neuroradiol 16: , Gillilan LA: Veins of the spinal cord. Anatomic details; suggested clinical applications. Neurology 20: , Hanakita S, Takai K, Kin T, Shojima M, Saito N: Double independent spinal dural arteriovenous fistulas at the thoracic spine. Acta Neurochir (Wien) 154: , Hanel RA, Nakaji P, Spetzler RF: Use of microscope-integrated near-infrared indocyanine green videoangiography in the surgical treatment of spinal dural arteriovenous fistulae. Neurosurgery 66: , Jellema K, Sluzewski M, van Rooij WJ, Tijssen CC, Beute GN: Embolization of spinal dural arteriovenous fistulas: importance of occlusion of the draining vein. J Neurosurg Spine 2: , Jellema K, Tijssen CC, van Gijn J: Spinal dural arteriovenous fistulas: a congestive myelopathy that initially mimics a peripheral nerve disorder. Brain 129: , Kim LJ, Spetzler RF: Classification and surgical management of spinal arteriovenous lesions: arteriovenous fistulae and arteriovenous malformations. Neurosurgery 59 (5 Suppl 3): S195 S201, Koenig E, Thron A, Schrader V, Dichgans J: Spinal arteriovenous malformations and fistulae: clinical, neuroradiological and neurophysiological findings. J Neurol 236: , Lai PH, Weng MJ, Lee KW, Pan HB: Multidetector CT angiography in diagnosing type I and type IVA spinal vascular malformations. AJNR Am J Neuroradiol 27: , Lasjaunias P, Berenstein A, Raybaud C: Surgical Neuroangiography Volume 3 Functional Vascular Anatomy of Brain, Spinal Cord and Spine. Berlin Heidelberg: Springer- Verlag, 1990, pp McCutcheon IE, Doppman JL, Oldfield EH: Microvascular anatomy of dural arteriovenous abnormalities of the spine: a microangiographic study. J Neurosurg 84: , Oldfield EH, Di Chiro G, Quindlen EA, Rieth KG, Doppman JL: Successful treatment of a group of spinal cord arteriovenous malformations by interruption of dural fistula. J Neurosurg 59: , Prestigiacomo CJ, Niimi Y, Setton A, Berenstein A: Threedimensional rotational spinal angiography in the evaluation and treatment of vascular malformations. AJNR Am J Neuroradiol 24: , Rosenblum B, Oldfield EH, Doppman JL, Di Chiro G: Spinal arteriovenous malformations: a comparison of dural arteriovenous fistulas and intradural AVM s in 81 patients. J Neurosurg 67: , Saraf-Lavi E, Bowen BC, Quencer RM, Sklar EM, Holz A, Falcone S, et al: Detection of spinal dural arteriovenous fistulae with MR imaging and contrast-enhanced MR angiography: sensitivity, specificity, and prediction of vertebral level. AJNR Am J Neuroradiol 23: , Spetzler RF, Detwiler PW, Riina HA, Porter RW: Modified classification of spinal cord vascular lesions. J Neurosurg 96 (2 Suppl): , Steinmetz MP, Chow MM, Krishnaney AA, Andrews-Hinders D, Benzel EC, Masaryk TJ, et al: Outcome after the treatment of spinal dural arteriovenous fistulae: a contemporary singleinstitution series and meta-analysis. Neurosurgery 55:77 88, Symon L, Kuyama H, Kendall B: Dural arteriovenous malformations of the spine. Clinical features and surgical results in 55 cases. J Neurosurg 60: , Tadié M, Hemet J, Freger P, Clavier E, Creissard P: Morpho logical and functional anatomy of spinal cord veins. J Neuroradiol 12:3 20, Takai K, Kin T, Oyama H, Iijima A, Shojima M, Nishido H, et al: The use of 3D computer graphics in the diagnosis and treatment of spinal vascular malformations. Clinical article. J Neurosurg Spine 15: , Takai K, Taniguchi M: Comparative analysis of spinal extradural arteriovenous fistulas with or without intradural venous drainage: a systematic literature review. Neurosurg Focus 32(5):E8, Takai K, Usui M: Spontaneous thrombosis of a spinal conus perimedullary arteriovenous fistula. Case report. Neurol Med Chir (Tokyo) 52: , Van Dijk JM, TerBrugge KG, Willinsky RA, Farb RI, Wallace MC: Multidisciplinary management of spinal dural arteriovenous fistulas: clinical presentation and long-term follow-up in 49 patients. Stroke 33: , 2002 Manuscript submitted March 26, Accepted January 15, Please include this information when citing this paper: published online February 22, 2013; DOI: / SPINE Address correspondence to: Keisuke Takai, M.D., Ph.D., Department of Neurosurgery, The University of Tokyo Hospital, Hongo, Bunkyo-ku, Tokyo , Japan