Unruptured Arteriovenous Malformations in the Multidetector Computed Tomography Era: Frequency of Detection and Predictable Failures
How to cite this article: Mattay RR, Miner L, Copelan AZ, Davtyan K, Schmitt JE, Church EW, et al. Unruptured arteriovenous malformations in the multidetector CT era: Frequency of detection and predictable failures. J Clin Imaging Sci 2022;12:5.
While hemorrhage arising from ruptured arteriovenous malformations (AVMs) is usually evident on multidetector non-contrast computed tomography (NCCT), unruptured AVMs can be below the limits of detection. We performed a retrospective review of NCCT of patients with a proven diagnosis of unruptured AVM to determine if advances in CT technology have made them more apparent and what features predict their detection.
Material and Methods:
Twenty-five NCCTs met inclusion criteria of having angiography or MR proven AVM without hemorrhage, prior surgery, or other CNS disease. Demographic variables, clinical symptoms at presentation, abnormal CT imaging findings, attenuation of the superior sagittal sinus (SSS), and Spetzler-Martin grade of each AVM were recorded. We examined the relationship between AVM detection and SSS attenuation through Kruskal–Wallis test. Exploratory serial logistic principal components analysis was performed including demographics, symptoms, and CT features in the multivariate model.
About 80% of the NCCTs showed an abnormality while 20% were normal. All those with an identifiable abnormality showed hyperdensity (80%). Logistic regression models indicate that clustered associations between several CT features, primarily calcifications, hyperdensity, and vascular prominence significantly predicted Spetzler-Martin grade (likelihood ratio 7.7, P = 0.006). SSS attenuation was significantly lower in subjects with occult AVMs when compared to those with CT abnormalities (median 47 vs. 55 HU, P < 0.04).
Abnormal hyperdensity was evident in all detectable cases (80%) and multiple CT features were predictive of a higher Spetzler-Martin AVM grade. Moreover, SSS attenuation less than 50 HU was significantly correlated with a false-negative NCCT.
While cerebral arteriovenous malformations (AVMs) occur in only 0.1% of the population, they underlie 1–2% of all strokes, 3% of strokes in young adults, and 9% of subarachnoid hemorrhages. Patients with AVMs can present with wide variations in symptoms that contribute to the difficulty with their diagnosis. Acute symptoms usually arise from a new hemorrhage, but seizures, headache, and focal neurologic deficits[2,3] can occur without hemorrhage. A recent meta-analysis has shown that AVMs have an overall annual hemorrhage rate of 3% and prior retrospective studies demonstrated that the average annual mortality in patients with untreated brain AVMs ranged from 0.7 to 1.0%.[5,6] While AVMs that previously ruptured have a higher annual rupture rate (4.5%), unruptured AVMs have an annual rupture rate of 1–2%.[4,7]
It is for these reasons that the detection of unruptured AVMs is essential since there are currently many available treatment options that include medical management, microsurgical resection, endovascular embolization, stereotactic radiosurgery, or some combination of these strategies. Early detection is valuable before hemorrhage since AVMs frequently appear in young patients, and in most cases, treatment will provide a complete cure or diminish the likelihood of hemorrhage and its associated morbidity and mortality.
Approximately half of AVMs present with intracranial hemorrhage,[4,7] which is usually first detected with a multidetector non-contrast computed tomography (NCCT) of the head, as this modality has a greater than 90% sensitivity for acute subarachnoid hemorrhage and hemorrhagic stroke.[9-11] Although Almandoz et al. have shown that NCCT features along with clinical features in patients presenting with intraparenchymal hemorrhage can establish the probability of an underlying vascular anomaly, the diagnosis of an AVM itself is usually not established on the basis of imaging findings on the non-contrast scan. Whenever there is a reasonably high probability of an underlying vascular lesion, based on age, history, and location of the hemorrhage, the patient should have an MR, MRA, CTA, or digital subtraction cerebral angiography (DSA) for diagnosis.
AVMs usually have a moderately well-defined cluster of closely spaced vascular structures, called the nidus, that, in some cases, will be visible on cross-sectional imaging. While modern CT and MR provide exceptional spatial resolution, the definitive diagnosis of an AVM usually requires the temporal resolution of DSA since the critical diagnostic finding of an AVM is not only a nidus but also the appearance of a contrast within a vein during the arterial phase of injection. This is referred to as the “early vein.”
While imagers have come to expect that unruptured AVMs will be difficult or impossible to detect on NCCT, there are much older studies (>30 years ago) that describe some imaging characteristics and very little in the recent literature that quantifies the likelihood of their detection [Table 1]. Since the utilization of NCCTs in the emergency setting has been increasing in the past several decades,[14,15] approximately doubling over a 10-year period, and advances in CT technology during this time have significantly improved; it seems essential for clinicians to be very familiar with the expected NCCT findings of unruptured AVMs. The dearth of recent literature describing and characterizing the NCCT characteristics of this entity and their infrequent occurrence in the course of routine clinical imaging represents a barrier to their recognition without hemorrhage. This study was initiated after one author’s experience with an AVM that was prospectively identified on NCCT in a young adult with new seizures, but only by its subtle high attenuation and associated parenchymal calcifications [Figure 1].
|Year||Author||Patients with NCCT (n)||NCCT abnormality detected (n)||No NCCT abnormality detected (n)||Details||Additional information|
|22||16||6||13 with high or low attenuation lesions
3 with other nonspecific abnormalities
|54 total studies evaluated included ruptured AVMs.|
|60||50||10||35 with calcifications
36 with intrinsic hypodensities
2 with extrinsic surrounding hypodensity (edema)
33 with mass effect
15 with ectasia of ipsilateral cistern
12 with compression of high convexity sulci
|All large unruptured AVMs
No specific mention of frequency of cases with NCCT evidence of hyperdensity other than calcifications
Vascular prominence was evaluated on contrast enhanced CT
|13||6||7||1 with calcification
3 with mixed, low, or high attenuation abnormalities
|MRI detected AVM in all cases|
On that basis, we performed a retrospective review of patients at two academic hospitals with DSA, CTA, or MR proven unruptured AVMs along with NCCT to determine the likelihood of detection during this modern era of multidetector CT. Our goals in collecting this information were to determine: (1) If advances in CT technology have improved the likelihood of visualization of AVMs on NCCT and (2) establish what specific imaging features contribute to making unruptured AVMs apparent or inapparent on NCCT.
MATERIAL AND METHODS
After IRB approval, a retrospective search was performed through this institution’s database of reports of cervicocerebral DSA from January 2002 until December 2017. We searched for reports with keywords “AVM,” “arteriovenous malformation,” and “head” and also without terms “ICH,” “rupture,” and “hemorrhage.” This revealed a total of 106 separate examination reports. Of these 106 examination reports, only patients with available NCCT predating the cerebral angiogram were included in the study. All patients with diagnoses other than AVM (e.g., dural AVF or cavernous-carotid fistula), prior neurosurgery, or intracranial hemorrhage in the general region of the AVM were excluded from the study. From that initial search, 16 patients with unruptured AVMs and NCCT were included in our analysis from Center 1. The diagnosis of AVM was established with DSA in 15 cases and CTA in one.
After IRB approval, a retrospective search through this institution’s database from March 2019 to November 2019 was performed. We searched reports for “AVM,” which resulted in 230 separate reports of initial and follow-up studies performed on patients during the specified time period. These 230 reports and images were reviewed for initial NCCT of the head performed on patients with unruptured AVMs, which spanned a time period of January 2003–March 2019. All studies with alternate diagnosis (e.g., dural AVF or cavernous-carotid fistula), prior neurosurgery, or intracranial hemorrhage in the location of the AVM were also excluded from the study. This search resulted in nine unruptured AVMs with CT. The diagnosis of AVM was established with DSA in six and typical findings on MR and CTA/MRA in the other three.
Numerous clinical variables for each patient were recorded, including demographic variables of age and gender as well as presenting symptomology including presence of headache, seizure, vertigo, visual disturbances, or sensorimotor disturbances [Table 2].
|Demographics||Center 1||Center 2||Total|
|Age (years)||47.4 (16.3)||44.0 (23.5)||46.2 (18.8)|
|Gender||12 M||(75%)||7 M||(78%)||19 M||(76%)|
|4 F||(25%)||2 F||(22%)||6 F||(24%)|
The NCCT images along with subsequent diagnostic vascular imaging of each case were reviewed independently by two investigators to rate the appearance of typical AVM imaging characteristics which included focal high attenuation, calcifications, vascular prominence, parenchymal edema, and adjacent bony change. In addition, the density of the superior sagittal sinus (SSS) in Hounsfield units (HU) was measured in the mid-sagittal sinus using available PACS tools for each case. The investigators then corroborated the location of the abnormalities on each CT with the location of the AVM based on vascular imaging (DSA, CTA, or MRI/MRA) and recorded the Spetzler-Martin grade of each AVM. The independent assessments of the two investigators were recorded and any differences were then reviewed by a third, more senior investigator (CAQ neuroradiologist with 27 years of experience) and resolved [Table 3]. Evidence of abnormality detection in the original interpreting radiologist’s report was also recorded.
|Non-contrast CT features||Advanced imaging features|
|Any abnormality||20 (80%)||Spetzler-Martin grade|
|Calcifications||9 (36%)||Grade 1||9 (36%)|
|Hyperdensity||20 (80%)||Grade 2||8 (32%)|
|Vascular prominence||14 (56%)||Grade 3||7 (28%)|
|Parenchymal edema||3 (12%)||Grade 4||1 (4%)|
|Adjacent bony change||1 (4%)|
|SSS density (HU)||57.8 (11.0)|
Data were imported into the R statistical environment for cleaning and analysis. Following assessment of data integrity, descriptive statistics characterizing frequencies of individual clinical and imaging features were first performed. To identify relationships between clinical symptoms, tetrachoric correlations were then calculated. Because we hypothesized that decreased attenuation of the blood pool may influence AVM detectability, we examined the relationship between AVM-positive CT and SSS attenuation through Kruskal–Wallis test.
We also wished to explore the predictive relationships of both clinical and CT imaging data on advanced imaging features; however, traditional analytic options were statistically limited by sample size and associated concerns for multiple testing. Thus, we performed an exploratory logistic principal components analysis (PCA) including demographic (sex and age), presenting symptoms (headache, seizure, vertigo, visual, and sensorimotor), and CT findings (calcifications, hyperdensity, vascular prominence, bony changes, edema, and SSS density) as features in the multivariate model. PCA reduces the dimensionality of the data, maximizing variance explained by the fewest number of orthogonal factors. Logistic PCA is similar to traditional PCA but is specifically designed for the analysis of dichotomous measures; this was ideal given that 11 of our 13 variables were binary. For this analysis and based on natural divisions in the data, age was binarized with a threshold of 45 years, and SSS density was set at a threshold of 70 HU.
Serial logistic PCA was subsequently performed with a range of 1–10 components; a three-factor solution was selected based on a natural elbow in the scree plot. This three-component model explained >70% of the deviance in the data. Factor loadings from the best fit model were examined, and natural groupings in the data were identified. Each of the three principal components was tested for its ability to predict Spetzler-Martin grade through ordinal logistic regression. Specifically, we included all three principal components in the full model and serially attempted to remove individual components. Statistical significance was assessed through likelihood ratio test.
Finally, we assessed the associations between our two continuous measures (age and SSS density) and AVM grade through ordinal logistic regression, after adjusting for other demographic and clinical features. An a of 0.05 (after Bonferroni correction) was selected as the threshold for statistical significance.
[Table 2] summarizes the clinical features of the sample. Our two separate centers’ retrospective analyses identified 25 patients with NCCT of brain and proven unruptured AVMs. Of the patients identified, 19 (76%) were male and 6 (24%) were female with a mean age of 46.2 ± 18.8 years. The most common presenting symptoms were seizure and headache (80% of all cases), with seizure more common at Center 1 and headache more common at Center 2. This may be a consequence of the relatively small sample size from each institution. Less frequent symptoms included vertigo, visual symptoms, and sensorimotor symptoms. [Table 4] summarizes correlational patterns between clinical features. Seizure at presentation was more common in males, while headaches were more common in females. Most other presenting symptoms were uncorrelated, although there were modest positive associations between visual symptoms, vertigo, and sensorimotor symptoms.
Of the patients with AVMs that met our inclusion criteria, 80% (20/25) of the AVMs showed some type of abnormality on CT [Figures 2-5], while 20% (5/25) were completely unidentifiable [Figure 6]. We then confirmed that these five were completely unidentifiable on NCCT through additional corroboration with the original interpreting radiologists’ report. The attenuation of the SSS was significantly lower in subjects with CT occult AVM (median 47 vs. 55 HU, c2 = 4.3, df = 1, P = 0.038). Of those with evident abnormalities on CT, all had some variation of high attenuation in the region of the AVM (20/20) ranging from ill-defined to serpiginous density, while 70% of these also had abnormal vascular prominence (14/20). About 45% of cases (9/20) also demonstrated abnormal calcifications in the region of high attenuation. Low attenuation suggesting edema or encephalomalacia was only seen in 20% (4/20) of the cases, and adjacent bony changes were seen in a single case (5%). In the five cases with no abnormalities to suggest an AVM, there were no other abnormalities elsewhere in the brain or skull.
Only one case demonstrated four CT features of an AVM, that is, focal high attenuation, vascular prominence, calcifications, and adjacent bony change. This case proved to be the only Spetzler-Martin Grade 4 AVM in the sample. There were no Spetzler-Martin Grade 5 AVMs nor micro-AVMs among our cases.
Nine cases demonstrated three abnormal CT features. Seven of these nine showed hyperdensity, vascular prominence, and calcifications [Figure 5]. All seven were SpetzlerMartin Grade 2 and above. One among the nine showed hyperdensity, vascular prominence, and edema and another showed hyperdensity, calcifications, and edema. Seven cases demonstrated two abnormal CT features, five of which showed hyperdensity and vascular prominence, and the other two showed hyperdensity and edema. Only three cases showed focal high attenuation as the only abnormality, and all of these proved to be Grade 1 AVMs.
Of the five AVMs that were unidentifiable on NCCT, two were Grade 1, two were Grade 2, and one was Grade 3. However, in four of the five cases, the measured attenuation of intravascular blood within the SSS measured 50 HU or less. In the one case with an unidentifiable AVM with attenuation of greater than 50 HU in the SSS, the AVM nidus measured about 1 cm based on dimensions from MR.
Results from logistic PCA are summarized in [Figure 7a-c]. The first component (PC1) again identified the relationship between sex and clinical presentation, with males associated with seizures and females with headaches. The second component (PC2) clustered associations between several CT features, primarily calcifications, hyperdensity, and vascular prominence, with a weaker loading on bony changes. The third component primarily identified associations between age and SSS density. Logistic regression models found that PC2 significantly predicted Spetzler-Martin grade (likelihood ratio 7.7, P = 0.006), with a progressive increase in PC2 score with increasing AVM grade [Figure 7c]. Associations between AVM grade and the other two principal components were not statistically significant.
We observed an inverse association between SSS density and Spetzler-Martin grade, with lower grades associated with higher density. However, this association was not statistically significant (likelihood ratio 2.1, P = 0.15). Patient age was inversely associated with AVM grade (likelihood ratio 8.9, P = 0.003), with younger patients generally having higher grade AVMs in our sample.
There are a few papers in the imaging literature that mention the conspicuity of AVMs on CT, but these prior studies appeared in the pre-helical CT era. In one review of the CT findings of 54 patients at the Montreal Neurological Institute with AVMs published in 1979, over half (30/54) presented with hemorrhage. Among the 24 cases with unruptured AVMs, 22 cases had CT available for review, six of which were interpreted as entirely normal. 13/22 CT scans demonstrated a high or low attenuation lesions, and the remaining 3/22 scans showed other non-specific abnormalities, such as infarct or dilatation of an ipsilateral ventricle. It is of interest that in six cases included in that study, no AVM was demonstrated on conventional angiography. On that basis, we can assume that some of the lesions detected on CT were cavernomas.
In 1987, the superior sensitivity of MRI for the diagnosis of AVM when compared with CT was reported in 15 patients. Among the 13 patients with NCCT, only a single patient demonstrated visible calcifications, and three had mixed, low, or high attenuation abnormalities. Seven of the 13 scans (53%) were completely normal. An additional, larger study by Kumar et al. in 1984 described the non-contrast and contrast-enhanced CT features in 60 unruptured AVMs, and showed a substantially higher NCCT detection rate of abnormalities in 50/60 scans (83%), although this is likely due to inclusion of only large AVMs in this series [Table 1 for additional details].
The paucity of information in the recent literature on the subject of the detection of AVMs using modern multidetector CT, which has high spatial resolution, may reflect the relatively infrequent detection of unruptured AVMs on NCCTs as well as the tacit acceptance that NCCT is not reliable for their detection. Berman et al. suggested in a paper from 2000 that the true prevalence of cerebral AVMs is unknown because they are rare, and the asymptomatic cases may be undercounted. Based on incidence of detection, they predict the prevalence of 10.3 per 100,000. To put this in perspective, one should consider the prevalence of a more common vascular abnormality, cerebral aneurysm, which has a prevalence that is estimated to be 2000 per 100,000 or 2%, that is, 200x higher.
The diagnostic difficulties in identifying AVMs on CT along with their low prevalence, especially those that present without rupture, very likely account for our difficulties in identifying a large number of cases for this study. In a careful search for documented AVM cases with CT from two institutions, we screened 336 radiology reports but found only 25 cases with proven AVMs without clinical or radiologic evidence of hemorrhage.
Based on this carefully selected series, the AVM was visible in some fashion on multidetector helical NCCT in 20/25 cases, with only 20% of scans rated as completely normal. This represents a substantial improvement compared to previously reported CT detection rates and likely reflects the improvement in CT hardware and reconstruction software over the intervening years.
The most common finding, seen in all positive scans, was abnormal high attenuation in the brain reflecting some combination of the nidus and associated supply and draining vessels. In about half of those cases, abnormal brain parenchymal calcifications were also evident. The most common set of findings, evident in about a third of our cases, was parenchymal focal high attenuation, prominent vasculature, and abnormal calcifications.
We did not expect to see evidence of an AVM on NCCT in all of these cases since the detection of small AVMs can be difficult even on DSA. However, we found it of interest that in four of the five cases with no detectable abnormalities on NCCT, the attenuation of intravascular blood, measured in the center of the mid-SSS, was lower than that of blood in this venous sinus in the cases with CT imaging abnormalities. This reached statistical significance even in this relatively small group of patients.
This emphasizes the reason why AVMs are apparent on NCCT. With the exception of calcifications, visualization of the nidus and associated enlarged vessels is entirely due to the fact that normally blood in the intravascular compartment is of higher attenuation than the brain parenchyma. The X-ray attenuation of flowing, intravascular blood has been demonstrated to correlate with the patient’s hematocrit, and the ratio of the measured attenuation of blood in the venous sinuses on NCCT with the patient’s hematocrit can be used to diagnose venous thrombosis on NCCT.[24,25] Since the detection of abnormal vessels and the AVM nidus on NCCT depends on the intravascular compartment having a higher attenuation than the brain parenchyma, it follows that in patients with low attenuation intravascular blood either from low hemoglobin or hemodilution, the AVM will not be visible on NCCT.
A central limitation of this study is that only cases with proven AVMs were included in this retrospective study of findings on CT. As a result, we can only report the findings in patients with AVMs that were eventually diagnosed. Since only about half of AVMs will lead to hemorrhage and some patients with AVMs are asymptomatic, it is entirely possible that patients with AVMs had NCCT scans during this time period that were read as normal. In addition, while we did not perform a blinded retrospective review of all 25 scans, our data regarding lack of detection of AVM were also corroborated with the original interpreting radiologists’ reports, who in essence were likely “blinded” to the patient’s eventual diagnosis of AVM.
It is accepted that CT features in patients with intraparenchymal hemorrhage should be considered along with age and clinical history to predict the likelihood of an underlying vascular anomaly as the cause of hemorrhage and on that basis order vascular imaging. We would add that in patients with the absence of intracranial hemorrhage and an attenuation of intravascular compartment blood similar to brain attenuation, it is likely that an unruptured AVM will be overlooked on NCCT. Just as with the patients with intraparenchymal hemorrhage, these patients may benefit from additional imaging, such as MR with or without MRA, especially when their symptoms could be on the basis of an underlying, unruptured AVM.
We found that in 80% of our patients with unruptured AVMs, abnormal high attenuation in the brain, with or without calcification, was evident on CT. This is a higher percentage than reported in early imaging studies with similar inclusion criteria[19,20] and very likely reflects improvements in CT imaging technology. The presence of multiple CT findings including hyperdensity, vascular prominence, and calcifications was significantly predictive of a higher Spetzler-Martin AVM grade. In the five patients with AVMs and scans rated as normal, four of the five had a relatively low attenuation of blood in the SSS (50 HU or less) and the fifth had a small 1 cm AVM. It seems likely that in patients with low contrast between brain parenchyma and the intravascular blood, the AVM nidus and associated vascular structures will be inapparent on NCCT and additional imaging is necessary for detection when an AVM is suspected as the source of symptoms.
Declaration of patient consent
Patient’s consent not required as patients identity is not disclosed or compromised.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
- R Foundation for Statistical Computing. Vienna, Austria: R Core Team; 2020.R: A Language and Environment for Statistical Computing In:
- [Google Scholar]