Vertebral Hemangiomas: Prevalence, new classification and natural history. magnetic resonance imaging-based retrospective longitudinal study

Abstract

Background and purpose:

To determine the prevalence of vertebral hemangiomas (VHs), establish a new classification of VHs based on their MRI-signal pattern, and study their natural history.

Methods

MRI of 1000 consecutive patients who underwent at least two MRI with an interval of at least 3 years. Growth rate and change of MRI-signal pattern during the follow-up period were the parameters included in studying the natural history of VHs.

Results

The prevalence of VHs was 41%. VHs were classified as type I–IV with fat-rich VHs (type I), constituted 79% of all VHs. VHs were more common among females 43% versus males 39%, p = .22. The most affected vertebra was L1. Occurrence rates for cervical (1%), thoracic (7%), and lumbar spine (10%) differed significantly (p < .001). The prevalence of VHs increased with age regardless of gender or spinal part involved (p < .001). Only 26% of VHs changed their size and 4% changed their signal during the average follow-up of 7 years. All VHs were slowly growing lesions (average expected growth of <3 mm/10 years). No significant difference between growth rate of VHs type I (0.25 mm/year) and other types of VHs. None of the VHs that were initially reported as “metastases cannot be rule out” showed alarming change in signal or size.

Conclusions

VH can be classified into four types based on their MRI-signal pattern. Regardless of their type, VHs are slowly growing lesions. The presence of typical morphological pattern should enable radiologists to confidently differentiate them from vertebral metastases.

Introduction

Skeletal hemangiomas are commonly affecting the spine and the skull. Vertebral hemangiomas (VHs) are frequent incidental findings on spinal confidence interval (CT) and MRI. Reported prevalence of VH in studies based on autopsy, ¹ CT, ² and MRI ³ was 11%, 26%, and 27%, respectively. A majority 89–99% of VHs are asymptomatic. ⁴

VHs are benign vascular tumors classified into four basic histopathological types: capillary, cavernous, arteriovenous, and venous with capillary and cavernous being the most encountered types of VHs. VHs usually include non-vascular components, for example, fat and smooth muscles, ⁵ which may explain diverse signal patterns of VHs. Typical VH show high signal intensity on both T1- and T2-weighted images (without fat saturation) reflecting their fatty component. The balance between fat and vascular components is reflected in high or low signal on Short Tau Inversion Recovery (STIR). VHs with muscle components may exhibit low signal on T1-weighted images (T1W). Hemosiderin deposits and thrombosis may contribute to signal inhomogeneity of VHs. High fluid contents might to some extent contribute to the high signal intensity on STIR.

Radiologists usually address VHs with high signal intensity on T1W- and T2W-images as typical VHs while those exhibiting different signal pattern usually are addressed as atypical VHs. Among patients with known neoplasm, this may sometimes cause diagnostic dilemma as atypical VHs (especially those with low signal on T1W) can mimic metastatic lesions and occasionally are reported as suspected metastases. This may give rise to repeated CT or MRI, scintigraphy, positron emission tomography (PET), and biopsy as well as generating unnecessary outpatient visits and causing anxiety for the patients involved. On CT, VHs usually exhibit the polka-dot sign. ⁶ Gaudino et al. ⁷ recently presented a comprehensive systematic illustration of different types of hemangiomas. However, VHs in this systematic approach were still classified as either typical or atypical.

There are numerous studies and case reports on e.g. large and symptomatic VHs and their management. ⁸ Aggressive hemangiomas that cause myelopathy or radiculopathy might be treated with embolization of the afferent vessels of the lesion followed by radiotherapy. ⁸ However, neither true prevalence nor natural history is well known. The high sensitivity of MRI in the assessment of focal bone marrow lesions makes it well suited for this purpose.

The aims of this longitudinal retrospective study were to: (1) determine the prevalence of VHs using MRI, (2) establish a new and clinically feasible classification of different types of VHs based on their MRI-signal pattern, and (3) study the natural history of VHs with special focus on the so called atypical VHs to evaluate the growth rate and the temporal evolution on MRI.

Methods

This retrospective longitudinal study included 1000 consecutive patients who fulfilled the following inclusion criteria: (1) Patients in our regional county council who underwent at least two spine MRI with a time interval of at least 3 years, and (2) MRI exams of whole, cervical, thoracic, and lumbar spine, that include T1W, T2 STIR, and T2W sagittal images or T2W isotropic 3D imaging (SPACE, VISTA, or CUBE). Exclusion criteria were: (1) Patients with diffuse malignant infiltration of bone marrow, for example, diffuse bone marrow metastases and multiple myeloma and (2) patients with metal implants. Sacrum was not included as heterogeneous fatty infiltration of the pelvis and sacrum is a common MRI-finding and thus expected to cause false positive results. Of 4725 patients examined during a period of about 2.5 months, 1000 met the inclusion criteria.

This study was approved by the regional ethical committee (Dnr 2016/300).

MRI exams

MRI exams were performed at 18 different scanners (Siemens, Philips and GE 1.5, and 3 Tesla). All MRI included were re-read by an experienced neuroradiologist with special profile, interest and long experience in spine radiology. All VHs were assessed with regard to: (1) T1-, T2-, and STIR-signal intensity pattern at the initial MRI exam, (2) presence of signal pattern change at the later MRI exam/s, (3) size at baseline MRI and later follow-up, (4) pedicular involvement, (5) extension into epidural fat, and (6) internal morphological features typical of VHs, namely: (a) punctate and/or coarse trabeculated pattern and (b) serpentine morphology typical of small vessel/vascular channels.

Classification of different types of vertebral hemangiomas

Based on our previous clinical experience, we have noticed four different MRI-signal patterns among VHs. Hereby, we proposed the following new classification of VH types based on their signal pattern. VHs in this study were classified into four types: The so-called typical VHs were classified as type I (high signal on T1- and T2-weighted images). The remaining VHs exhibited three different signal patterns and subsequently classified as type II, III, and IV. Table 1 and Figure 1.

Table 1.

Signal pattern of different types of vertebral hemangiomas.

Type of VH	T1-signal	T2-signal	T2-STIR signal
Type I	High	High	Low a
Type II	High	High	High
Type III	Low signal with serpentine and/or peripheral rim of high signal	High	High
Type IV	Low or indistinct	High	High or indistinct

^aVHs with predominantly low STIR-signal but with peripheral rim of high STIR-signal were also classified as type I.

Figure 1.

Different types of VHs. 1 (a–c): VH type I: high signal on T1W and T2W (a–b) and low signal on STIR-images (c). 2 (a–c): VH type II: high signal on all three sequences T1W, T2W, and STIR-images. 3 (a–c): VH type III: high signal on STIR-images (c) and predominantly high T2-signal (b) and alternating punctate low and high T1-signal with peripheral rim of high signal intensity (a). 4 (a–c): VH type IV: low signal on T1W-image (a) and high signal T2W and STIR-images (b–c).

Reliability of the proposed classification of vertebral hemangiomas

The reliability of the proposed classification of VH types was tested by Kohen’s kappa statistics for the degree of inter-reader and intra-reader agreement. A total of 100 randomly selected patients were included in this analysis (75 patients with VHs and 25 cases with no VH). Reader 1 performed the classification at two separate occasions with 1.5-year interval and Reader 2 performed the classification independently at one occasion, both readers are experienced neuroradiologist.

Natural history of vertebral hemangiomas

The natural history of VHs was analyzed with respect to change of VH size or/and change of signal pattern during the follow-up period of at least 3 years. A lesion that did not grow or change signal pattern in an alarming way during the ≥3 years follow-up period was considered benign. Furthermore, atypical VHs described in the original radiological report as suspected malignant lesions were thoroughly analyzed. Assessment of growth rate and occurrence of change of MRI-signal pattern were the parameters included in studying the natural history of VHs. The size of all VHs at baseline and at the last follow-up was recorded to calculate their growth rate. The growth rate of VHs has been calculated as: total increase or decrease in the size of VH during the whole follow-up period given in mm/follow-up period given in years. The MRI-signal pattern at baseline and at follow-up was compared to evaluate changes in MRI-signal pattern. The mean follow-up interval was 7 years (range 3–18).

Statistical analysis

All statistical analyses were done in R-statistics. ⁹ Data are presented as proportions (%) or as mean ± standard deviations (SD). Uncertainty in estimates is presented as the 95% confidence interval (CI). Statistical significance was set to a p value ≤0.05. Chi square test was performed for studying association between categorical variables. Cohen´s kappa statistics of inter-reader and intra-reader agreement were interpreted according to grading proposed by Landis and Koch. ¹⁰

Results

Included in the analysis of this study were 1000 patients, 609 (61%) were females. Mean age (±SD) was 58 ± 16 years (median: 59; range: 6–93). Indications for MRI: Back pain with or without neurological deficit (71%), known malignancy to exclude of metastasis (5.7%), multiple sclerosis (5.2%), rheumatic disorder (2.8%), trauma (2.5%), spinal deformity (2%), and other indications (10.8%).

Reliability analysis of the proposed classification of vertebral hemangiomas

Inter-reader agreement for the presence of VH was perfect (k 0.92, 95% CI 0.82–1). Inter-reader agreement for classification of the four different types of VH was substantial (k 0.80, 95% CI 0.71–0.89).

Intra-reader agreement for the presence of VH was perfect (k 0.97, 95% CI 0.92–1). Intra-reader agreement for classification of the four different types of VH was perfect (k 0.87, 95% CI 0.80–0.95).

Prevalence of vertebral hemangiomas

Of 1000 patients included, a total of 653 VHs were found in 414 patients (41%). Prevalence of VHs among males and females was 39% and 43%, respectively, p = .22. Of 414 patients with VHs, 260 (63%) had one VH, 92 (22%) had two VHs, and 62 (15%) had 3–7 VHs. Of 653 VHs found, 519 (79%) exhibited signal characteristics consistent with type I, 77 (12%) with type II, 26 (4%) with type III, and 31 (5%) with type IV. Of 653 VHs, 163 (25%) were ≥1 cm in diameter, nine (1.4%) involved ≥50% of the vertebral body, and 15 (4%) showed pedicular involvement. Two VHs showed extension into epidural fat (Figure 2) and one into paravertebral space (Figure 3). One VH type I has been treated with vertebroplasty. The diameter of VH varied between 5 and 31 mm with an average of 9 ± 5 mm at the first MR exam and 12 ± 6 mm at the last follow-up.

Figure 2.

(a) T2 SPACE sagittal image and (b) T1W-axial image at the level of L3. The vertebral body is totally infiltrated with a VH type I, with epidural extension (arrows).

Figure 3.

Sagittal T1W (a), sagittal T2 SPACE (b), and coronal T2 STIR (c) show a large VH type II involving almost the whole L3-vertebral body with extension into the paravertebral fat medial to right psoas muscle (c). Notice the generally high signal in all three sequences and the typical pattern of trabeculation and punctate signal loss in this small vessel rich lesion.

Among patients ≤20 years, the prevalence of VH was 13%, increasing with age to 16%, 46% and 46% in age groups 21–40, 41–60, and ≥60 years respectively. Logistic regression showed that the increasing prevalence of VHs with age was independent of gender or spinal part involved (Coefficient 0.02, p < .001).

Topographical distribution of vertebral hemangiomas

Totally 9733 vertebrae were included in the analysis. The proportion of VHs in the 9733 evaluated vertebrae was 6.7%. The distribution of VHs at different vertebral levels is shown on Figure 4. The most commonly involved vertebra was L1 followed by L4 and T12 with occurrence rate of 12.5, 11.9, and 10.6%, respectively. Occurrence rates for cervical, thoracic, and lumbar spine differed significantly (p < .001), Table 2.

Figure 4.

Rate of occurrence (proportion) of VH at different vertebral levels. The rate of occurrence of VHs at different vertebral levels was calculated as: 100*number of VHs/number of vertebrae included at every specific vertebral level.

Table 2.

Proportion of vertebral hemangiomas at different vertebral levels. (Chi square, pairwise comparison of proportions with Bonferroni adjustment, p < .001 between all pairs.)

Spinal part	Number of vertebra included	Number (%) of VH
Cervical spine	2334	33 (1%)
Thoracic spine	3734	263 (7%)
Lumbar spine	3665	357 (10%)
Total	9733	p < 0.001

Natural history of vertebral hemangiomas

(A) Growth of vertebral hemangiomas

Of 653 detected VHs, 172 (26%) showed measurable change of size: 159 VHs increased in size and 13 VHs decreased in size during the follow-up period. The growth rate for 172 VHs was 0.3 ± 0.4 mm/year (minimum growth rate −1.2 mm/year and maximum growth rate 2.25 mm/year). Separate estimation of the rate of growth of VH > 1 cm was performed. Regardless of their type, the growth rate of VH exceeding 1 cm in size was 0.8 ± 0.7 mm/year compared with 0.2 ± 0.2 mm/year for VH < 1 cm (p = 0.068). The growth rate of VHs type I was 0.25 ± 0.5 mm/year compared with 0.23 ± 0.3 mm/year (type II), 0.21 ± 0.2 mm/year (type III), and 0.3 ± 0.4 mm/year (type IV) (p = .68, 0.50, and 0.22, respectively).

Nine VHs showed growth rate ≥1 mm/year, that is, expected increase of 1 cm/10 years. All of these nine VHs showed morphological characteristics that remind of benign VHs, namely, punctate and/or coarse trabeculated pattern and/or small vessel rich lesions, often seen best on T2W-images, Table 3.

Table 3.

Occurrence of typical signal characteristics among vertebral hemangiomas type II–IV that increased ≥1 mm/year (i.e. expected increase of ≥1 cm in 10 years).

Type of VH	Size first exam-last exam (mm)	Follow-up period (year)	Rate of growth (mm/year)	Exhibited typical morphological characteristics of VHs
IV	22–31	4	2.25	Yes
II	15–25	6	1.67	Yes
II	9–14	4	1.25	Yes
IV	6–11	4	1.25	Yes
II	15–20	4	1.25	Yes
IV	12–16	4	1	Yes
II	26–29	3	1	Yes
IV	7–11	4	1	Yes
II	12–15	3	1	Yes

(B) Change of MRI-signal pattern

Of 653 VHs, only 28 (4%) showed change of signal characteristics, that is, changed their hemangioma type during the follow-up period. Different constellations of signal pattern change are tabulated, Table 4. The most common type of change of signal pattern was from VH type II to VH type I, observed in 13 (46%) out of 28 VHs that changed their signal pattern. However, the vast majority of VHs (96% of all VHs) retained their hemangioma type. No statistically significant difference in the length of follow-up period of VHs that change their hemangioma type (6.8 ± 1.5 years) compared with those that retained their hemangioma type (7.3 ± 0.2 years), p = .56. Figure 5 shows two different examples of VHs that changed their signal pattern and increased in size during the follow-up.

Table 4.

Vertebral hemangiomas that changed their signal pattern (n = 28, 4%) and thus their type during the follow-up period.

Type of VH: first MRI-exam	Type of VH: last MRI-exam	Number of VHs that changed their type/signal pattern
I	III	2
II	I	13
III	I	4
III	IV	2
IV	I	3
IV	II	2
IV	III	2

Figure 5.

(a–b) T1W- and T2W-images show VH type IV (low T1-signal, a) that changed its signal pattern to type I (high T1-signal, c) at the follow-up MRI 6 years later (c–d). (E–H) T1W-images and T2 STIR-images of another patient with VH type I. VH increased in size from 17 mm to 23 mm during the follow-up period of 13 years (0.5 mm/year) but retained its fat-rich signal pattern.

Out of the 57 patients with a known malignancy examined to exclude vertebral metastases, 10 had vertebral metastases. Among these patients, 23 VHs were found. These VHs could easily be distinguished from metastases.

Vertebral hemangiomas type IV

As VHs type IV exhibit signal pattern similar to metastatic lesions (low T1-signal and high T2- and STIR-signal); further search for this type of VHs was done at the end of the study period. A total of 72 VHs type IV were found at MRI exams performed during 6-month period. The growth rate of VHs type IV was 0.3 ± 0.4 mm/year (range −0.4 mm/year – 2.25 mm/year). Of 72 VHs type IV, 11 were described as suspected metastases with different formulations in the original radiological reports, the most commonly used was: “Metastases cannot be ruled out”. This resulted in 9 CT exams, three skeletal scintigraphies, and one biopsy and generated many neuroradiological consultations and out-patient visits to the orthopedic or oncology clinics. No evidence of malignancy was found at follow-ups. Two VH type II and four VH type III were reported as suspected malignancy, resulted in repeated MRI follow-up which did not show any significant growth, nor signal changes supporting malignancy. Table 5 shows summary of different results of the current study.

Table 5.

Summary of different results of the current study.

• Prevalence of VHs: Whole study cohort	41%
• Proportion of different types of VHs
VHs type I	79%
VHs type II	12%
VHs type III	4%
VHs type IV	5%
• VHs >1 cm in size	25%
• VHs involved >50% of the vertebral body	1.4%
• VHs with pedicular involvement	4%
• Rate of growth of VHs	mm/year
Whole study cohort	0.3 ± 0.4
VHs ≥1 cm	0.8 ± 0.7
VHs <1 cm	0.2 ± 0.2
VHs type I	0.25 ± 0.5
VHs type II	0.23 ± 0.3
VHs type III	0.21 ± 0.2
VHs type IV	0.3 ± 0.4

Discussion

In this study, VHs were very common findings on spine MRI. The prevalence of VH was 41%, which is significantly higher than the prevalence of 11% reported in previous autopsy-based studies, ¹ and higher than the prevalence of 26% based on CT. ² The lower detection rate of VHs by CT can be explained of the superiority of MRI as clinical experience of 4 decades of using MRI as diagnostic modality and numerous previous reports have shown that MRI has better sensitivity than CT for detection of focal bone lesions.^11–13 Reported prevalence of VHs among Iranian population using MRI was 26.9%. ³ We believe this reported prevalence of VHs ³ is underestimated. Different population constellation and racial differences may however explain some of the discrepancy. Furthermore, inclusion of three sequences (T1W, T2W, and STIR) in our study may have improved the detection rate.

Although the imaging appearance cannot precisely predict the pathological type of VHs, it could help increasing understanding of the natural history of VHs. Pastushyn´s et al. reported differences in the clinical course, imaging characteristics, and outcome among different histopathological types of VHs. ¹⁴ This was one of the reasons why we adopted a classification based on signal pattern of VHs. However, for VHs type IV, we also recommend searching for morphological features reminding of VHs, namely, the occurrence of serpentine morphology, small flow void, and the typical coarse trabeculated pattern. Furthermore, we noticed that T1-signal in type IV hemangiomas is usually slightly reduced compared with metastatic lesions showing more pronounced reduction of signal intensity on T1W.

In accordance with previous reports, the prevalence of VHs among females was higher than among males with occurrence M:F ratio varying from 1:1.3 to 1:2.25 ² . In our study, the M:F ratio was 1:1.1. VHs can be found at all ages. Age distribution of VHs in our study (mean age 61 years) was almost the same as previously reported, with higher occurrence at later decades of life.^2,15 Although the occurrence of VHs among children was not specifically addressed in the current study, the increased prevalence of VHs with age indicates that VHs are at least to a large extent acquired lesions.

In our study, atypical VHs (VHs type II–IV) had almost the same behavior and natural history as the typical ones (VHs type I) with no significant difference in growth rate. No single VH type IV had a growth rate or change in signal pattern that gave rise to suspicion of malignancy at later MRI. Follow-up MRI of 17 cases with VH type II–IV that initially reported as suspected metastases showed neither evidence of rapid growth nor signal changes in favor of malignant lesions.

Loredo et al. suggested that a high fat content in VH type I was a sign of inactivity, while VHs with low fat content (VH type II, III, and IV) were believed to be more vascular and tend to enlarge and give rise to symptoms. ¹⁶ In our study, all types of VHs including those with alarming low T1-signal (VHs type IV) were as slow growing lesions as the typical VH type I (expected growth rate of 2–3 mm/decade). Furthermore, in our study, out of nine VHs involving ≥50% vertebral body, seven were fat-rich (VHs type I). Two VHs showed extension into the epidural space, both were type I. All nine VHs (type II and IV) with growth rate ≥1 mm/year (expected growth rate of about 1 cm/decade) exhibited morphological characteristics typical of VHs. One VH type IV showed growth rate of 2.25 mm/year. This VH was examined with a 4-year interval. The low growth rate (9 mm in 4 years) and the occurrence of typical hemangioma pattern indicated a benign nature.

Friedman et al. stated that high fat content and aggressiveness may coexist. What made the diagnosis straightforward was the appearance of these lesions with punctate hypointensity. ¹⁷ This is in line with our suggestion that morphological behavior is as important as the growth rate when evaluating the aggressiveness and nature of VHs and with our observation that large VHs mainly were type I.

In many institutions, shortage and low availability of MRI sometimes force cutting short MRI-protocols and consequently do not include all three sagittal images (T1-, T2-, and STIR). Therefore, not only VHs type IV but even type II and III (high STIR-signal) may in some occasions impose diagnostic challenge and may mimic metastases. We believe that the results of our study, partly our new classification and partly the results from the study of the natural history of VHs, may give some support for reporting radiologists faced with these lesions on daily basis. In our opinion, the morphological characteristics of VHs are the most important clue to make a correct diagnosis.

Included in this study was MRI of whole spine as well as MRI of cervical, thoracic, and lumbar spine. This may constitute a limitation of this study. However, the aim was to include as many vertebrae as possible when studying the natural history of VHs. Another limitation was the lack of pathological confirmation. A third limitation of this study was its retrospective nature which did not allow additional imaging with, for example, diffusion weighted images (DWIs) as these have recently been proved helpful to differentiate between metastatic deposits from VHs. ¹⁸ However, the latter study included only VHs with typical fat signal that seldom impose any diagnostic difficulties.

Huge number of case reports of single or few cases of VHs mimicking metastases, aggressive VHs, rapidly progressive VHs, symptomatic VHs, VHs with unusual appearance, VHs with unusual clinical presentation, VHs with epidural extension, and surgically treated VHs have been published. Furthermore, the diversity of inclusion criteria and criteria used to classify these lesions make it extremely difficult to have an overview, not even with a meta-analysis. We hereby chosen some of the reports^3,7,19–24 that focused on the diagnosis of VHs, classification of the VHs, and differentiation of them from metastases. Table 6. The old terminology of typical and atypical VHs has its drawbacks. Atypical VHs (type 2, 3, and 4 according to our classification) do not need to raise malignant suspicion only because they show high signal intensity on STIR. Internal structural architecture typical of VHs has according to our study cohort showed to be helpful to differentiate what previously called atypical VHs from metastases. We believe that our adopted classification of VHs will raise the threshold for suspecting vertebral metastases, help radiologists to be more confident in differentiating VHs from other bony lesions, especially skeletal metastases, and result hopefully in lesser dependency on the usage of modalities such as DWI and perfusion studies to differentiate the two entities. Consequently, awareness of different types of VHs despite the diversity of their signal changes reduces the need for unnecessary radiological follow-up exams, out-patient visit, bone biopsies, and above all less concern for the patients.

Table 6.

Some of previously published studies using the old terminology of typical and atypical vertebral hemangiomas.

Author	Type of study/study cohort	Terms used to describe VHs	Notes
Barzin et al 3	782 MRI	Non specified	Incidence 26.9%
N. L. N Moorthy et al 19	3200 MRI	Typical, atypical and aggressive	245 VHs, incidence of VHs 7.6%
Robert L et al 20	726 patients (683 MRI and 43 CT)	Only typical VHs included	998 VHs found. Incidence not studied
J S Ross et al 21	8 VHs	Typical and atypical	—
K A Morales 22	54 atypical VHs 42 metastases	Typical and atypical	MR perfusion for differentiation
Yan-Jie Shi et al 23	27 patients (33 VHs) 71 metastases	Typical and atypical	MR diffusion weighted images for differentiation. Accuracy 70.19% and 89.53%
S.Gaudino et al 7	Systemic approach	Typical and atypical	Provided a practical guide for the differential diagnosis from findings on radiography, CT, and MRI
L Nigro et al 24	Editorial	Typical and atypical	Focusing on the radiological appearance of VHs and treatment

Conclusions

In summary, VHs are very common incidental findings found among 41% of patients examined with MRI of spine. Instead of classifying VHs into typical and atypical, we recommend using the classification proposed in this study. Regardless of the signal characteristics (VH type), the presence of typical internal structural features that remind of VHs is the most important clue to confidently differentiate them from the focal, relatively sharply demarcated, and punched-out vertebral metastases. As this study showed that all VHs irrespective of their type are benign slow-growing lesions, we recommend that small VHs (<1m VHs type I–III) not necessarily need to be reported as they may cause unnecessary radiological follow-up exams, need for biopsy, out-patient visit, and above all concern for the patients.

ORCID iD

Kasim Abul-Kasim https://orcid.org/0000-0002-5323-149X