Article Text

Download PDFPDF

original research
Supernova hemorrhage: obliterative hemorrhage of brain arteriovenous malformations following gamma knife radiosurgery
  1. Matthew D Alexander1,
  2. Steven W Hetts2,
  3. William L Young3,
  4. Van V Halbach2,
  5. Christopher F Dowd2,
  6. Randall T Higashida2,
  7. Joey D English4
  1. 1Santa Clara Valley Medical Center, Department of Radiology, San Jose, California, USA
  2. 2Department of Radiology, UCSF, San Francisco, California, USA
  3. 3Department of Anesthesiology, Neurological Surgery and Neurology, UCSF, San Francisco, California, USA
  4. 4Department of Neurology and Radiology, UCSF, San Francisco, California, USA
  1. Correspondence to Dr J D English, Department of Neurology, San Francisco General Hospital, 1001 Potrero Ave, 1X55, San Francisco, CA 94110, USA; joey.english{at}


Hemorrhage represents the most feared complication of cerebral arteriovenous malformations (AVMs) in both untreated patients and those treated with gamma knife radiosurgery. Radiosurgery does not immediately lead to obliteration of the malformation, which often does not occur until years following treatment. Post-obliteration hemorrhage is rare, occurring months to years after radiosurgery, and has been associated with residual or recurrent AVM despite prior apparent nidus elimination. Three cases are reported of delayed intracranial hemorrhage in patients with cerebral AVMs treated with radiosurgery in which no residual AVM was found on catheter angiography at the time of delayed post-treatment hemorrhage. That the pathophysiology of these hemorrhages involves progressive venous outflow occlusion is speculated and the possible mechanistic link to subsequent vascular rupture is discussed.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


The natural history of cerebral AVMs above all includes the risk of intracranial hemorrhage (ICH), and seizure and focal neurological deficit may also occur.1 Stereotactic radiosurgery has gained popularity in recent decades to ablate AVMs and reduce these risks.2 Such treatment does not immediately lead to AVM occlusion; before obliteration there is a latency period often lasting years during which the patient remains at risk for ICH.

ICH, the most feared outcome from AVMs, is possible both prior to intervention and in the latency period following irradiation. Therapeutic levels of radiation cause an ill defined process of endothelial or myointimal proliferation and intravascular thrombosis within the vascular structures within the targeted volume leading to eventual obliteration of the nidus. Endothelial damage from the treatment can also predispose to vascular rupture and bleeding.3 ,4 Such obliteration usually requires 1–3 years to occur, and bleeding rates approximate or exceed those prior to treatment during the latency period between radiosurgery and complete obliteration (2–15% vs 1.6–16%).3 ,5 Obliteration constitutes treatment success and must be confirmed angiographically. Success rates are highly variable, ranging from 54% to 92%.3 Persistent hemorrhage risk is the major drawback of radiosurgery compared with surgical resection.6–8

Radiosurgical obliteration is more likely to occur, and sooner, with smaller and lower grade AVMs, those with only one draining vein and after higher radiation doses.9 Traditionally, complete obliteration has been considered curative, with the risk of post-obliteration hemorrhage exceedingly low.6 However, several cases of ICH have been reported following demonstrated obliteration. In general, these hemorrhages occurred months to years after radiosurgery, in brain tissue previously containing or adjacent to the AVM nidus in settings of confirmed residual AVM or lesion recurrence recognized on imaging, and in patients 18 years of age or younger.9–15

We report three cases of delayed ICH in patients with cerebral AVMs previously treated with radiosurgery. Two of these patients had hemorrhage without evidence of residual or recurrent AVM at the time of ictus. The third case involved a post-radiosurgical hemorrhage with features that may shed light on the pathophysiology of the hemorrhages in the other two patients, which we describe as supernova hemorrhages and represent a previously undescribed complication of AVMs treated with gamma knife therapy.

Case reports

Patient A

Patient A is a man in his early 50s with a Spetzler–Martin grade 3 right frontal AVM incidentally found during participation in a research study (figure 1A,B). At the time he was asymptomatic and declined treatment. The patient presented 8 years later with headache and left hemiparesis and was found to have a small intracerebral hemorrhage adjacent to the AVM (figure 1C). Catheter angiography confirmed a peri-Sylvian AVM nidus with arterial supply from the right middle cerebral artery cortical and lateral lenticulostriate branches (figure1D–G). Intra-nidal aneurysms were identified within the AVM, and it had deep venous drainage through an intraventricular varix. Three months after the intracerebral hemorrhage the patient underwent the first stage of gamma knife radiosurgery, followed by a second stage 3 months later, receiving 17 Gy in both sessions. Seventeen months after the second intervention, he presented with generalized tonic clonic seizures. CT imaging demonstrated subtle acute subarachnoid hemorrhage within the Sylvian fissure, anterior to the location of the previously identified AVM, and CT angiography demonstrated a patent residual AVM nidus and deep venous drainage (figure 1H,I). Two days later the patient's left hemiparesis worsened, and a follow-up CT demonstrated increased peri-nidal edema and associated mass effect (figure 1J). Conventional cerebral angiography at this time demonstrated neither residual AVM nidus nor arteriovenous shunting (figure 1K,L). The previously identified venous varix draining the AVM was not seen, suggesting thrombosis of the venous outflow and the AVM nidus. Follow-up catheter angiography 6 months later demonstrated no residual AVM (figure 1M,N).

Figure 1

Axial unenhanced (A) and contrast enhanced (B) CTs demonstrated a right frontal arteriovenous malformation (AVM) with large draining varix. Unenhanced CT (C) demonstrated intracranial hemorrhage (ICH) after the patient presented with headache several years after being lost to follow-up. Diagnostic angiography characterized the lesion prior to gamma knife radiosurgery, with early (D) and late (E) arterial phase anteroposterior (AP) views, as well as early (F) and late (G) arterial phase lateral views. Unenhanced (H) and contrast enhanced (I) CTs demonstrated ICH and residual AVM with deep venous drainage in the setting of new seizures. Unenhanced CT (J) 2 days later demonstrated no residual AVM or residual varix. Angiography, with early (K) and late (L) arterial phase AP views. Early (M) and late (N) arterial phase lateral views confirmed no residual malformation or varix at the 6 month follow-up.

Patient B

Patient B is a woman in her late 30s on chronic warfarin therapy for atrial flutter who experienced aneurysmal subarachnoid hemorrhage. Angiography at that time demonstrated a right posterior communicating artery aneurysm as well as a high flow right parietal Spetzler–Martin grade 3 AVM receiving both anterior and posterior circulation arterial supply and draining through deep veins with large varices. The right posterior communicating artery aneurysm was surgically clipped, and the AVM underwent partial embolization intended to reduce nidal supply in preparation for subsequent gamma knife radiosurgery. Eleven months after embolization the nidus appeared slightly smaller angiographically with reduced arterial inflow and no outflow stenosis. Single stage gamma knife radiosurgery with a dose of 18 Gy was performed 28 months after initial detection and embolization. Angiography at this time demonstrated similar arterial supply but venous drainage occurred predominantly through one large vein of Galen varix. Fifty-five months after treatment, she presented with several days of severe headache and confusion. Intraventricular hemorrhage was seen on CT and MR imaging with no clear source of bleeding. Conventional cerebral angiography showed no arteriovenous shunt and no other source of the intraventricular hemorrhage.

Patient C

Patient C is a woman in her mid-teens diagnosed with a Spetzler–Martin grade 5 left frontotemporal AVM discovered during workup for seizures she had been experiencing since the age of 5 years (figure 2A–C). The patient underwent three courses of gamma knife radiosurgery at 0, 3 and 13 months after diagnosis, receiving 18 Gy, 17 Gy and 17.5 Gy, respectively. Fifty-one months following her final treatment, she re-presented with worsening left frontotemporal headaches in addition to more frequent seizures. MR imaging (figure 2D–F) and angiography (figure 2G,H) at this time showed the AVM was now smaller and drained through both superficial and deep veins. Angiogram at 57 months demonstrated a new large left deep draining venous varix with venous restriction. At 64 months she presented with a new, severe headache with acute onset. CT showed bilateral intraventricular hemorrhage and partial thrombosis of a large venous varix associated with the primary venous outflow of the AVM (figure 2I–K). At no point prior to this had she demonstrated hemorrhage. Cerebral angiography demonstrated a generally unchanged arterial supply but progressive restriction of the draining veins (figure 2I–M). In comparison with prior angiograms, the large varix associated with the major venous drainage was notably decreased in size and had a halo consistent with the intraluminal thrombus detected on CT. Another vein draining into the superior sagittal sinus demonstrated markedly sluggish flow with stasis of contrast in the distal vein itself.

Figure 2

Axial T2-weighted MR (A) demonstrated left frontotemporal arteriovenous malformation (AVM). Diagnostic angiography, with lateral venous (B) and mid-arterial phase (C) images, characterized the AVM prior to gamma knife radiosurgery. T2 weighted (D), T1 weighted (E) and contrast enhanced T1 weighted (F) MR images obtained 5 years following treatment demonstrated new deep drainage through a large varix. Venous phase anteroposterior (AP) (G) and lateral (H) angiography images further characterized the new deep drainage. In the setting of new severe headache, unenhanced (I,J) and contrast enhanced (K) CT images demonstrated intraventricular hemorrhage, decreased deep drainage and thrombosis of the varix with decrease in size. AP (L) and lateral (M) venous phase angiographic images confirmed the reduced flow through and size of the varix.


Here we have reported a previously undescribed complication of AVMs treated with gamma knife radiosurgery—the supernova hemorrhage. As discussed above, hemorrhage is the most feared complication of AVMs during their natural history and during the latent period between radiosurgery and obliteration. Indeed, patient A's course involved a typical latency period bleed. However, rupture in the setting of lesion occlusion is exceedingly rare and typically associated with residual AVM. Patients A and B had atypical courses in that their hemorrhages could not be attributed to residual or recurrent malformation. Previously reported cases of hemorrhage in patients with prior demonstration of lesion occlusion had evidence of bleeding at or immediately adjacent to the previous nidi with residual AVM demonstrated on imaging.9–15 The cases of patients A and B are also the first in which AVM occlusion was first demonstrated on imaging performed for an acute hemorrhage.

We speculate that the ICHs experienced by patients A and B resulted from the progressive occlusion of nidal venous outflow, with a dramatic, supernova-like vascular rupture at the culmination of complete obliteration of the nidus as a result of an acute increase in pressure in residual AVM feeding arteries, nidal components and the proximal venous outflow. The progressive venous restriction and thrombosis demonstrated on sequential imaging in patients A and B support this hypothesis. Furthermore, the progressive venous occlusion and subsequent rupture prior to complete AVM obliteration seen in patient C suggest the same pattern.

Venous hypertension is a powerful angiogenic stimulus in the cerebral circulation and can induce arteriovenous fistulae formation in rodent models.16 ,17 Importantly, even at levels not associated with decreased perfusion pressure and therefore non-ischemic, venous hypertension results in a pro-angiogenic state in the brain by upregulating HIF-1 and its downstream targets, VEGF and SDF-1a. This increases leucocyte infiltration and MMP-9 activity.18 Increased proteolytic activity may well predispose to rupture. Such pathophysiologic theorization deserves further basic science investigation.

Patients A and B are older than any patients previously reported to have post-occlusion hemorrhage.9–15 Some have argued that younger patients with AVMs are more prone to recurrence, as previously reported cases involved bleeds secondary to recurrence in young patients, although this has been disputed.2 ,19 In theory, the pathologic vasculature in older patients has presumably undergone a longer period of stress. This could make vasculature in older patients more likely to rupture during the stress of increased backpressure on dysplastic nidal vessels during venous outflow thrombosis. Other possible explanations for currently reported hemorrhages are that malformations were not truly obliterated, the lesions had recurred but were not seen on angiography or the lesion was compressed or obscured by adjacent hemorrhage, although these scenarios are unlikely given the negative delayed follow-up angiography for patient A.

Supernova hemorrhage—AVM rupture at the moment of venous outflow thrombosis and nidus obliteration—is a late complication that should be considered in patients treated for cerebral AVMs with gamma knife radiosurgery. This phenomenon warrants further investigation to elucidate which patients and lesions are most susceptible to such hemorrhage. For example, steady state free precession MR imaging has been used in the evaluation of partially thrombosed cerebral aneurysms.20 Such non-invasive longitudinal imaging of thrombus within venous varices draining AVMs may provide better insight into the mechanisms involved in spontaneous rupture and the likelihood of supernova hemorrhage.



  • Competing interests None.

  • Patient consent Obtained.

  • Provenance and peer review Not commissioned; externally peer reviewed.