Vertebral marrow lesions are quite frequently detected in magnetic resonance imaging
(MRI) of the spine referred for various clinical scenarios. Although MRI is highly
sensitive for detecting marrow lesions, the specificity can be improved by understanding
and choosing appropriate sequences. Relative distribution of fat and water in the
vertebral marrow lesions, estimation of cell density, and analysis of edema or vascularity
using various MRI sequences are increasingly used to improve the specificity of lesion
characterization. MRI plays the main role in avoiding the biopsy of benign vertebral
lesions or tumor mimics. In this regard, conventional sequences like T1-weighted,
T2-weighted, and short tau inversion recovery (STIR) sequences can be combined with
problem-solving sequences like diffusion-weighted imaging (DWI), chemical shift imaging
(CSI), or Dixon-based fat quantification and dynamic contrast studies.
Vertebral lesions can be broadly categorized into three groups based on the most common
patterns we encounter in daily practice. Group I includes vertebral collapse, which
can be osteoporotic or neoplastic, and group II includes single or multiple focal
vertebral benign or malignant lesions, which can be detected incidentally or in patients
with known underlying malignancy. Group III lesions are diffusely heterogeneous marrow
signal changes that can be for various reasons like hematological, metabolic disorders,
marrow reconversion related, and chemoradiation induced. Although patterns are varied,
the principles of distinguishing the above groups of lesions remain the same.
Acute and subacute vertebral collapse poses diagnostic challenges to distinguish from
malignant causes mainly due to variable marrow signal changes. Although there are
many distinguishing features of benign and malignant collapse in conventional sequences,
the most specific feature to rule out malignancy is the preserved fat signal within
the collapsed vertebra. Whereas pedicle involvement; and soft-tissue components are
fairly specific for metastasis. The sensitivity, specificity, and accuracy of conventional
MRI sequences to detect metastatic compression fracture are 100, 93, and 95%, respectively.[1]
[2] The total predictive value to detect metastatic fracture is between 94 and 97.3%.[3]
[4]
Whenever there are equivocal findings, DWI, CSI, Dixon-based fat quantification, and
dynamic contrast can be used. CSI and Dixon imaging are based on the detection of
the amount of marrow fat replaced by the lesion. CSI with in-phase and out-phase sequences
can be used to quantify the fat signal drop. The cutoff value ratio of 0.8 resulted
in a sensitivity of 95% and a specificity of 89 to 95% for detection of neoplasm.[5]
[6] Ragab et al[6] used 35% as a cutoff and found the sensitivity and specificity of the out-of-phase
images was 95% and 100%, respectively. The multi-echo Dixon method is now very popular
for rapid and more accurate quantitative assessment of the fat content as compared
with CSI. The fracture fat fraction cutoff value is 5.2% with a sensitivity of 95.8%
and a specificity of 95%. However, several cutoff values are suggested by various
studies with a sensitivity of 93 to 95% and specificity of 82 to 100%.[7] DWI can quantify the water mobility in tissues with ADC value as a substitute providing
indirect estimates of cellularity. Malignant lesions show low ADC values due to high
cellularity. Various studies in the literature have mentioned cutoff values of less
than 1.7 to 0.8 and collective sensitivity and specificity for osteoporotic and metastatic
vertebral fractures of 0.92 (95% confidence interval [CI]: 0.82–0.97) and 0.91 (95%
CI: 0.87–0.94), respectively.[8] We should be aware that the DWI acquisitions are not uniform across the world; however,
high b value and thin sections are recommended. When compared with fat fraction by CSI and
ADC, fat fraction had excellent repeatability and was a superior discriminator than
ADC as reported by a study by Donners et al.[9] There can be overlap in the values, especially in sclerotic lesions, which has to
be considered when relaying on CSI. In completely collapsed vertebrae, chronic infection
may also give false-negative results for metastasis in both CSI and DWI. Dynamic contrast-enhanced
MRI[10] is another problem-solving tool that is reported to be of use. The increased cost,
acquisition time involved, and multiple flow parameters to be analyzed make it a less
popular sequence to be included in routine clinical use.
Both vertebral marrow metastatic lesions and heterogeneous marrow pose different diagnostic
challenges since the imaging features in routine MRI overlap with many benign lesions
including red marrow islands. T1-weighted sequences are quite sensitive to detect
the marrow lesions with lesions having a signal intensity lower than disks or muscles,
indicating the malignant nature.[11] These lesions are less bright on fat-saturated images and do not show contrast enhancement.
However, when conventional sequences are indeterminate, CSI or Dixon and DWI may be
useful. Studies have shown CSI provides a sensitivity of 91.7%, specificity of 73.3%,
and accuracy of up to 82%. A signal drop of more than 20% almost rules out malignancy
with up to 91% sensitivity and high negative predictive value; however, the lesion
that shows a signal drop of ≤20% has to be followed up or biopsied.[12] DWI with quantitative ADC value estimation is more accurate than visual estimation
of restricted diffusion. Pooled sensitivity and specificity of quantitative ADC values
(1.01 ± 0.22 mm2/s) for marrow lesions are 89% and 87%, respectively.[8] Hypercellular focal red marrow may have overlapping values with metastasis on ADC.
CSI is ideal for diagnosing atypical hemangioma and focal red marrow hyperplasia with
good certainty, reducing the need for biopsies. Sclerotic metastasis, lesions with
small foci of metastasis, and infiltrative pathologies with preserved trabeculations
especially myeloma or lymphoma might show greater signal drop-off, falsely indicating
benignity. Lesions with hypercellular hematopoietic marrow like spindle cell hemangioma
and extensive fibrosis in chronic osteomyelitis might show false-positive results.
Predicting vertebral lesions as benign or malignant using conventional imaging sequences
can be inconclusive in a small number of cases. Problem-solving sequences greatly
enhance the confidence of radiologists to categorize these lesions. These are now
routinely included in scanning protocols, of which Dixon sequences are more popular
due to ease, good repeatability, and rapid acquisition. Biopsy or close follow-up
with clinical evaluation is still standard practice in lesions with overlapping imaging
features in spite of problem-solving sequences.
