Differentiation of Acute Osteoporotic and Malignant Vertebral Fractures by Quantification of Fat Fraction With a Dixon MRI Sequence

Transcription

1 Musculoskeletal Imaging Original Research Kim et al. MRI of Vertebral Fractures Musculoskeletal Imaging Original Research Dong Hyun Kim 1 Hye Jin Yoo 2,3 Sung Hwan Hong 2,3,4 Ja-Young Choi 2,3 Hee Dong Chae 2,3 o Mi Chung 5 Kim DH, Yoo HJ, Hong SH, Choi JY, Chae HD, Chung M Keywords: Dixon sequence, fat fraction, malignant compression fracture, osteoporotic compression fracture DOI: /JR Received March 1, 2017; accepted after revision May 1, Supported by Seoul National University Hospital Research Fund grant Department of Radiology, Seoul Metropolitan Government Seoul National University, oramae Medical Center, Seoul, Korea. 2 Department of Radiology, Seoul National University College of Medicine, 103 Daehak-ro, Jongno-gu, Seoul, 03080, Korea. ddress correspondence to H. J. Yoo (dalnara3@gmail.com). 3 Department of Radiology, Seoul National University College of Medicine, Seoul, Korea. 4 Institute of Radiation Medicine, Seoul National University Medical Research Center, Seoul, Korea. 5 Department of Radiology, Veterans Health Service Medical Center, Seoul, Korea. JR 2017; 209: X/17/ merican Roentgen Ray Society Differentiation of cute Osteoporotic and Malignant Vertebral Fractures by Quantification of Fat Fraction With a Dixon MRI Sequence OJECTIVE. The purpose of this study was to differentiate malignant compression fractures from acute osteoporotic compression fractures of the spine by use of a Dixon MRI sequence to quantify fat fraction (FF). MTERILS ND METHODS. Forty-four vertebral compression fractures were assessed with turbo spin-echo T1-weighted and six-echo Dixon sequences for FF quantification at 3-T MRI. The fractures were divided into malignant compression fractures (n = 24) and acute osteoporotic compression fractures (n = 20). Two radiologists independently measured quantitative parameters from ROIs in the fractures, including the T1 signal intensity of the fracture, the FF of the fracture, and the FF ratio (fracture FF divided by normal marrow FF). The mean values of the parameters were compared between the two groups, interobserver reliability between two radiologists was assessed, ROC curves were analyzed, and logistic regression analysis was performed. RESULTS. The fracture FF and FF ratio of malignant compression fractures were significantly lower than those of acute osteoporotic compression fractures (fracture FF, 2.73% vs 14.36% [p < 0.001]; FF ratio, 0.05 vs 0.22 [p < 0.001]). There was no difference in T1 signal intensity of the fracture. The ROC UC of fracture FF was 0.98 and of FF ratio was In logistic regression analysis, fracture FF remained a significant variable that could be used to independently differentiate malignant from acute osteoporotic compression fractures (odds ratio, 0.33; p < 0.005). CONCLUSION. FF and FF ratio obtained from FF maps obtained with a six-echo Dixon MRI sequence may be useful for differentiating acute osteoporotic compression fractures from malignant compression fractures. R adiologists must often seek to differentiate malignant compression fractures from acute osteoporotic compression fractures during interpretation of conventional MR images of the spine [1]. Particularly in elderly patients, malignant compression fractures must be excluded because the spine is a common site of metastasis, even though osteoporosis is the most common cause of compression fractures in patients in this age group [2]. In general, acute osteoporotic fractures can be adequately differentiated from malignant compression fractures with only conventional MRI. However, confident differentiation is not always possible. Sometimes at conventional MRI, malignant compression fractures exhibit signal intensities and morphologic changes similar to those of acute osteoporotic compression fractures, necessitating invasive bone biopsy to confirm the nature of the vertebral compression fracture [1, 3, 4]. This may affect treatment strategy and the prognosis. Many investigators have sought to differentiate the two types of fracture using dual-echo in-phase and opposed-phase chemical shift encoded imaging by the two-point Dixon method. The rationale is that fatty marrow remains within acute osteoporotic compression fractures but is replaced or infiltrated by tumor tissue in malignant compression fractures [5 9]. However, the two-point Dixon method has several limitations when used for semiquantitative assessment of bone marrow fat content. First, it is difficult to correct for the T2*-shortening effects of the bony trabeculae unless additional mapping sequences are performed. Second, the two-point Dixon method is limited in terms of fat quantification, allowing only fat fractions (FFs) of 0 50% to be estimated. The sequence cannot be used to determine whether the dominant component JR:209, December

2 Kim et al. is fat or water; an additional sequence is required to resolve this problem. Moreover, calculation of decreases in signal intensity during dual-echo imaging is time-consuming and inconvenient in daily clinical practice [5 7, 10]. Multiecho chemical shift encoded imaging by the multiecho Dixon method has been used for rapid assessment of fat content. The technique is also called quantitative water-fat imaging and FF MRI [11, 12]. Of the Dixon variants, the six-echo acquisition strategy more consistently separates fat and water signals, affording reliable R2* measurements and correcting for many confounding factors within a clinically acceptable acquisition time [11, 13 18]. Thus, of the various sequences available, measurement of the FF by the six-echo Dixon method with T2* correction and a multipeak fat model may be a powerful noninvasive and convenient method of quantitatively differentiating acute spinal osteoporotic fractures from malignant compression fractures. The aims of our study were to evaluate the diagnostic accuracy and reliability of T2*- corrected FF quantification using a six-echo Dixon method and to determine a cutoff that enables differentiation of malignant from acute osteoporotic compression fractures. Materials and Methods Patient Selection This retrospective study received institutional review board approval; the requirement for informed patient consent was waived. etween November 2014 and July 2016, 154 consecutively registered patients were referred for MRI because of low-back pain (n = 73) or for follow-up of bone metastases (n = 81). The imaging examinations were performed in a designated MRI machine (Ingenia 3 T, Philips Healthcare). Forty-one of the patients were included in the current study because they fulfilled the following inclusion criteria: age older than 19 years; availability of MR images, including images obtained with the six-echo Dixon method with subsequent automatic reconstruction of FF maps; and presence of an acute vertebral fracture (see later, Definition of an cute Vertebral Fracture). The exclusion criteria were metallic implant (n = 2) and a defect too small to allow placement of an ROI on the focal bone lesion of the FF map (n = 1). Therefore, 38 patients (17 men, 21 women; mean age, 63.1 years; range, years) were enrolled. Six patients had recent fractures of two vertebral body segments; therefore, a total of 44 lesions were included in our study. The fractures were divided into two groups: acute osteoporotic compression fractures (n = 20) and malignant compression fractures (n = 24). oth mean age (68.7 years [range, years] vs 58.4 years [range, years]; p = 0.003) and ratio of patients who underwent systemic chemotherapy (7/20 vs 19/24; p = 0.003) differed between acute osteoporotic compression fractures and malignant compression fractures, but no significant difference was evident in terms of sex ratio, ratio of patients who underwent radiotherapy, and proportions of single and multiple compression fractures. The patient demographic data are summarized in Table 1. Definition of an cute Vertebral Fracture vertebral fracture was diagnosed when a loss of height was evident in the anterior, middle, or posterior dimension of the vertebral body. nother fracture characteristic was the presence of a horizontal linear region of low signal intensity within a vertebral body; this represented the fracture line. However, not all fractures had such lines [4]. Fractures were considered to be acute if the interval between symptom onset and MRI was less than 1 month and if MRI showed marrow edema within a deformed vertebral body [19, 20]. Final Diagnosis ll fractures were diagnosed as acute osteoporotic or malignant compression fracture by a radiologist with 7 years of experience in musculoskeletal radiology. Twenty vertebrae had acute osteoporotic compression fractures. The final diagnoses were confirmed with biopsy (n = 1) or imaging follow-up (n = 19). Four patients underwent follow-up MRI 3 10 months after their initial examinations. If a lesion did not progress and the fatty marrow became redistributed within the lesion, the fracture was considered to be benign. In 15 patients, complete regression or a gradual reduction in pain, no evidence of radiographic progression of vertebral destruction (n = 17), and absence of progression as revealed by CT (n = 7) served to indicate the benign nature of the fracture [21]. Regression and pain reduction were assessed by physical examination and recorded in the electronic medical records. Follow-up radiography and CT were performed 2 9 months after the initial MRI. Twenty-four of the vertebral body fractures were malignant compression fractures, diagnosed with either biopsy (n = 1) or on the basis of unequivocal imaging findings (with or without progressive deterioration) at follow-up MRI (n = 23). Unequivocal imaging findings at conventional MRI included a paravertebral soft-tissue mass, abnormal signal intensity and thickening of the pedicle and posterior element, development of a convex posterior border with the vertebral body, and evidence of other spinal metastases [21]. Of the 23 vertebral fractures that were not biopsied, seven had all four unequivocal imaging findings, five had three (paravertebral mass [n = 5], abnormal signal intensity [n = 3], convex posterior border of vertebral body [n = 3], and other spinal metastases [n = 4]), and nine had two (paravertebral mass [n = 2], abnormal signal intensity [n = 6], convex posterior border of vertebral body [n = 1], and other spinal metastases [n = 9]). There were three cases of sclerotic bone metastasis; the others were osteolytic metastasis. Of the 24 malignant compression fractures, 16 were focal lesions, and eight TLE 1: Patient Demographic and Clinical Characteristics in cute Osteoporotic and Malignant Compression Fractures Parameter Osteoporotic (n = 20) Malignant (n = 24) p Sex (no.) Men 8 12 Women ge (y) 68.7 (28 79) 58.4 (28 80) a Time from symptom onset to MRI study (d) 16 (2 29) 14 (2 30) Previous chemotherapy a Previous radiation therapy No. of vertebral fractures Single Multiple 9 10 Location of vertebral fracture Thoracic 5 9 Lumbar Sacral 2 Note Values in parentheses are ranges. Dash ( ) indicates not applicable. a Statistically significant JR:209, December 2017

3 MRI of Vertebral Fractures diffusely involved the whole vertebral body. There was no significant difference in fracture FF between focal and diffuse lesions (p = 0.707). The 24 patients with malignant compression fractures had the following underlying malignancies: lung cancer (n = 5), breast cancer (n = 4), colon cancer (n = 3), lymphoma (n = 2), nasopharyngeal cancer (n = 2), prostate cancer (n = 2), cervical cancer (n = 2), stomach cancer (n = 2), hepatocellular carcinoma (n = 1), and cholangiocarcinoma (n = 1). Fourteen of the 33 acute osteoporotic compression fractures were combined with the following malignancies: lung cancer (n = 2), hepatocellular carcinoma (n = 2), colon cancer (n = 2), cholangiocarcinoma (n = 1), lymphoma (n = 1), glottic cancer (n = 1), pancreatic cancer (n = 1), prostate cancer (n = 1), and stomach cancer (n = 1). MR Image cquisition MRI was performed with a 3.0-T MRI system fitted with eight-element phased-array total spine coils (Ingenia 3 T, Philips Healthcare). Standard clinical MRI, including turbo spin-echo T1-weighted imaging (sagittal acquisition; TR/TE, 694/15; flip angle, 90 ; slice thickness, 3 mm; interslice gap, 0 mm; matrix size, ; FOV, mm; imaging time, 3 minutes 50 seconds), was performed. Sixecho Dixon sequences (sagittal acquisition; TR/TE, 7.2/1.21; echo spacing, 1.1 ms; flip angle, 3 ; slice thickness, 3 mm; interslice gap, 0 mm; matrix size, ; FOV, mm; imaging time, 50 seconds) allowing FF quantification were then performed. Care was taken to ensure that T1-weighted and six-echo Dixon slices were acquired under the same geometric conditions. R2* maps were estimated from the six-echo sequences and corrected for T2* effects, because trabecular bone shortens the T2* values of water and fatty components, inducing rapid decay of the measured gradient-echo signal with TE [11, 18]. fter T2* correction, FF maps were automatically reconstructed. efore this study, we performed a comparison study to validate the six-echo Dixon sequence for measurement of the FF of a vertebral body with single-voxel MR spectroscopy as the reference standard. In these unpublished data, the six-echo Dixon sequence had good correlation with singlevoxel MR spectroscopy (r 2 = 0.869; p < 0.001). MR Image nalysis Five parameters were measured on T1-weighted and automatically reconstructed FF images: T1 signal intensity of normal bone marrow, T1 signal intensity of the fracture, T1 signal intensity of a nondegenerated disk, FF of normal bone marrow, and FF of the fracture (Fig. 1). The T1 signal intensity of a nondegenerated disk served as the internal reference standard for assessment of bone marrow signal intensity on T1-weighted images. Nondegenerated disks were selected for evaluation of sagittal T2-weighted images on the basis of normal signal intensity and height evident on T2 images [22]. Each measurement of T1 signal intensity of a nondegenerated disk was obtained by drawing an ROI within the disk (inside the adjacent endplates) on a T1-weighted image. To measure the T1 signal intensity and FF of the facture, we drew freehand ROIs on the images of sagittal slices centered on the lesion and the adjacent two paramedian sagittal images. We adjusted these ROIs in terms of the size and shape of low-signalintensity alterations evident on T1-weighted images, excluding the far periphery and the cortex of the region. We then calculated the means of all values. The T1 signal intensity and FF of normal bone marrow of the unaffected vertebrae were measured above and below the fractured vertebra on both midline sagittal images and the adjacent two paramedian sagittal images. Each value was obtained by placing a circular ROI that was as large as possible within the vertebral body inside the endplate and cortex. We next calculated the means of all values. Each FF ratio was calculated as fracture FF divided by the FF of normal bone marrow. The fracture-to-normal T1 signal intensity ratio was calculated as the T1 signal intensity of the fracture divided by the T1 signal intensity of normal bone marrow. The lesion-to-disk ratio was calculated as the T1 signal intensity of the fracture divided by the T1 signal intensity of a nondegenerated disk. ll parameters were independently measured by two musculoskeletal radiologists (1 and 7 years of experience in spinal MR image interpretation). oth used PCS workstations (Infinitt, Infinitt Healthcare) after training in ROI measurement using sample image sets. Statistical nalysis Interclass correlation coefficients were calculated to assess the extents of agreement between Fig year-old woman with acute osteoporotic fracture. For measurement of T1 signal intensities and fat fraction (FF), ROI was placed on portion of lesion exhibiting low signal intensity on T1-weighted image. To avoid partial volume effect, far periphery and cortex were not included., Sagittal T1-weighted MR image shows locations of measurement of T1 signal intensity of normal bone marrow (yellow circles), T1 signal intensity of fracture (arrow, blue outline), and T1 signal intensity of nondegenerated disk (white circle)., FF map shows locations of measurement of FF of fracture (arrow, blue outline) and FF of normal bone marrow (circles). JR:209, December

4 Kim et al. the two readers in terms of measurement of four parameters (not T1 signal intensity of a nondegenerated disk) on T1-weighted images and the automatically reconstructed FF maps. The averages of the values of the two readers were used in further analyses. n interclass correlation coefficient or kappa value of 0 indicated poor agreement; , slight agreement; , fair agreement; , moderate agreement; , good agreement; and , excellent agreement. Various parameters of acute osteoporotic and malignant compression fractures were compared. Fisher exact test or the Mann-Whitney U test was used to compare demographic data and imaging parameters between the two groups. inary multiple logistic regression analysis was performed to determine the relative contribution of the different imaging findings and demographic data for differentiation of acute osteoporotic and malignant compression fractures. Characteristics with p < 0.20 at univariate analysis were used as independent input variables for multiple logistic regression analysis. In multiple logistic regression analysis, the forward stepwise selection mode was used with iterative entry of variables based on test results (p < 0.05). Removal of variables was based on likelihood ratio statistics with a probability of To evaluate the diagnostic utilities of various parameters, we performed ROC analysis to determine sensitivities, specificities, and cutoff values. ll statistical analyses were performed with one of two programs (SPSS version 22, SPSS; MedCalc version , MedCalc Software). Results Interobserver greement ll interobserver agreements were excellent. The interclass correlation coefficient for fracture FF was 0.97; T1 signal intensity of the fracture, 0.97; FF of normal bone marrow, 0.99; and T1 signal intensity of normal bone marrow, Comparisons of Mean Parameter Values Comparisons between the mean parameter values of acute osteoporotic and malignant compression fractures are summarized in Table 2. Of the various parameters, fracture FF, FF of normal bone marrow, T1 signal intensity of normal bone marrow, and FF ratio of malignant compression fractures were significantly lower than those of acute osteoporotic compression fractures (Figs. 2 and 3). Otherwise, we found no significant differences between the fracture types in T1 signal intensity of the fracture, T1 signal intensity of a nondegenerated disk, fracture-to- TLE 2: Comparisons of Means of Parameters etween Patients With cute Osteoporotic and Malignant Compression Fractures Parameter Osteoporotic Malignant p Fat fraction Fracture (%) ± ± 1.60 < a Normal bone marrow (%) ± ± a Ratio 0.22 ± ± 0.03 < 0.00 a T1 signal intensity Fracture ± ± Normal bone marrow ± ± a Nondegenerated disk ± ± Fracture-to-normal ratio 0.47 ± ± Lesion-to-disk ratio 0.98 ± ± Note Data are mean ± SD. a Difference between benign and malignant lesions was significantly significant for all b value combinations (Mann-Whitney U test). Fig year-old man with malignant compression fracture and known advanced gastric carcinoma., Sagittal T1-weighted MR image shows malignant compression fracture (arrow, green outline) of L3 vertebral body with epidural tumor extension. Yellow circles indicate normal bone marrow; white circle indicates nondegenerated disk., Fat fraction (FF) map shows fractured vertebral body (arrow, green outline) with FF of 4.62% and FF ratio of 0.10, allowing identification of malignant compression fracture at cutoff value of 5.26% of FF. Yellow circles indicate normal bone marrow JR:209, December 2017

5 MRI of Vertebral Fractures normal T1 signal intensity ratio, or lesion-todisk T1 signal intensity ratio (Table 2). Multivariable nalysis and Diagnostic Performance Multivariate analysis was performed with the parameters and demographic data found to be significantly different between the two groups in the univariate analysis: age, systemic chemotherapy, fracture FF, FF of normal bone marrow, T1 signal intensity of normal bone marrow, and FF ratio. Multivariate forward stepwise logistic regression analysis showed that fracture FF was the only variable that could be used to independently differentiate malignant compression fractures from acute osteoporotic compression fractures (odds ratio, 0.33; 95% CI, ; p < 0.005). The cutoff value and the ROC UC of fracture FF are shown in Table 3 and Figure 4. The fracture FF and FF ratio had high diagnostic performance (UC, 0.98 and 0.95; p < 0.001) in the diagnosis of fracture type. The fracture FF cutoff value of 5.26% was clinically applicable, affording excellent diagnostic accuracy (sensitivity, 95.83%; specificity, 95.00%). When the cutoff was 6.18%, sensitivity for differentiation between acute osteoporotic and malignant compression fractures was 100%; that is, all malignant compression fractures had fracture FF values lower than 6.18%. With a fracture FF cutoff of 2.71%, specificity was 100% for differentiation of fracture type; all acute osteoporotic compression fracture FF values were greater than 2.71%. False-Positive Results In our study, all malignant fractures (n = 24) were correctly diagnosed with only routine MRI sequences. However, two osteoporotic compression fractures were misdiagnosed as malignant fractures with routine MRI. When the six-echo Dixon sequence with a fracture FF cutoff value of 5.26% was added, one of the misdiagnosed cases was correctly diagnosed as osteoporotic compression fracture with a high fracture FF (10.6%). However, the other severely collapsed fracture remained false-positive with a low fracture FF (2.82%) (Fig. 5). False-Negative Results There were three osteosclerotic metastases among the 24 malignant compression fractures. Two of these lesions had fracture FF values greater than the cutoff of 5.26% and were misclassified as osteoporotic fracture (false-negative). The other osteosclerotic metastasis had a low fracture FF value (3.47%) and was correctly diagnosed as a malignant fracture. CT showed two falsenegative lesions with high attenuation (450 and 686 HU), but the other true-positive lesion had relatively low attenuation compared with two false-negative lesions (264 HU). Discussion We found that the FFs and FF ratios of malignant compression fractures derived with a six-echo Dixon method were significantly Fig year-old woman with acute osteoporotic compression fracture and known advanced gastric carcinoma., Sagittal T1-weighted MR image shows acute osteoporotic compression fracture of L2 vertebral body (arrow, blue outline). Yellow circles indicate normal bone marrow; white circle indicates nondegenerated disk., Fat fraction (FF) map shows fractured vertebral body (arrow, blue outline) with FF of 29.27% and FF ratio of 0.31, which suggest acute osteoporotic compression fracture at cutoff of 5.26% of FF. Yellow circles indicate normal bone marrow. lower than those of acute osteoporotic compression fractures, allowing differentiation of the two fracture types. FF at a cutoff of 5.26% had high diagnostic performance (UC, 0.98). In contrast, neither the fracture T1 signal intensity nor the lesion-to-disk ratio was useful for differentiating malignant from acute osteoporotic compression fractures. Previous investigators [5 8] used dualecho chemical shift encoded imaging to assess tissue fat content by measuring the decrease in signal intensity on opposed-phase TLE 3: Diagnostic Performance of Parameters at ROC Curve nalysis Parameter UC Cutoff Value Sensitivity (%) Specificity (%) p a Fracture fat fraction 0.98 ( ) < ( ) ( ) < Fat fraction ratio 0.95 ( ) < ( ) ( ) < Note Values in parentheses are 95% CI. a Of UC. JR:209, December

6 Sensitivity (%) Fat fraction ratio Fracture fat fraction Specificity (%) Fig. 4 Graph of results of ROC analysis shows fracture fat fraction UC of 0.98 (95% CI, ) and fat fraction ratio UC of 0.95 (95% CI, ). Fig year-old man with acute osteoporotic compression fracture., Sagittal T1-weighted MR image shows acute osteoporotic compression fracture of L1 vertebral body (arrow, blue outline) causing severe collapse. Circles indicate normal bone marrow and disk., Fat fraction (FF) map shows fractured vertebral body (arrow, blue outline) with low FF value of 2.82% and low FF ratio of 0.04, inviting misinterpretation as malignant compression fracture at cutoff of 5.26% for FF value. Circles indicate normal bone marrow. Kim et al. TLE 4: Summary of Previous Studies of Chemical-Shift Imaging for Differentiating cute Osteoporotic and Malignant Compression Fractures Study Zajick, Jr. et al. [5] No. of Patients and Lesions a Lesion Type Parameter 92 (215) enign lesion, 164; benign compression fracture, 8; malignant lesion, 51 Erly et al. [6] 21 (49) enign compression fracture, 29; malignant lesion, 20 Ragab et al. [7] 40 enign compression fracture, 20; malignant lesion, 20 Zidan et al. [8] 32 enign compression fracture, 17; malignant lesion, 15 Percentage decrease in opposed-phase compared with in-phase signal intensity SI ratio (opposed phase / in phase) Percentage decrease in opposed-phase compared with in-phase SI SI ratio (opposed phase / in phase) Suggested Cutoff Value b Sensitivity (%) Specificity (%) 20% N N % Note N = not available, SI = signal intensity. a Values in parentheses are numbers of lesions. b Lesions were defined as malignant when suggested cutoff value was satisfied for each parameter JR:209, December 2017

7 MRI of Vertebral Fractures images. Normal marrow fat in osteoporotic compression fractures suppresses signal in such sequences. However, when normal marrow fat is replaced with tumor in a malignant compression fracture, the signal is not suppressed. Several cutoffs have been suggested for differentiating acute osteoporotic from malignant compression fractures. These parameters afford sensitivities and specificities of 93 95% and %. The sensitivity (95.83%) and specificity (95.00%) of FF in our study are comparable to those in previous studies in which chemical-shift imaging was used (Table 4). However, dual-echo chemical shift encoded imaging is time-consuming and inconvenient. Moreover, errors are inevitable in the calculation of fat and water signal intensity owing to the many confounding factors, including the inhomogeneity effects of the principal magnetic field; the presence of multiple peaks in the fat spectrum; T2*, T1, and eddy current effects; and susceptibilityinduced shifts in fat resonance [23]. Several MRI vendors have adapted the six-echo Dixon method to quantitative assessment of bone marrow fat, correcting the confounding factors to allow more accurate fat measurements. In two previous studies [12, 23] the six-echo Dixon method was used to differentiate malignant from benign bone marrow lesions. Kim et al. [12] explored whether FF can be used to differentiate malignant marrow-replacing lesions from benign red marrow depositions by use of T2*-corrected FF maps obtained with a 3D volume-interpolated breath-hold gradientecho Dixon sequence. Those authors found that the lesions can be differentiated on the basis of FF. The diagnostic performance was high, similar to our observation (UC, 0.961; sensitivity, 85.7%; specificity, 100%). However, vertebral compression fractures were not included in the analysis. Yoo et al. [23] conducted a study with a modified sixecho Dixon sequence. They explored whether FF and FF ratio can be used to differentiate vertebral fractures of different types. Those authors found that both the FF and FF ratio of malignant vertebral fractures were significantly lower than those of benign vertebral fractures (p < for both comparisons). limitation of the study was that only a small number of vertebral compression fractures were included. Fig year-old man with malignant compression fracture and known breast cancer. MRI showed progression of extent of lesion over 6 months., Sagittal T1-weighted MR image shows compression fracture at superior endplate of T10 vertebral body (arrow, blue outline). Circles indicate normal bone marrow and disk., Fat fraction (FF) map shows fractured vertebral body (arrow, blue outline) with high FF of 5.75% and high FF ratio of 0.06, inviting misinterpretation as acute osteoporotic compression fracture at FF cutoff value of 5.26%. Circles indicate normal bone marrow. C, CT scan shows fractured segment (arrow) is osteosclerotic metastatic lesion. C JR:209, December

8 Kim et al. We found overlap in the FFs of malignant and acute osteoporotic compression fractures, as did previous investigators. One acute osteoporotic compression fracture (a false-positive finding) had an FF of 2.8%, below the 5.26% cutoff. This was an acute posttraumatic vertebral compression fracture associated with severe collapse of the vertebral body (Fig. 5). We suggest that the residual fat signal intensity may be obscured by noise-related bias [15] and the T2* effect [17] of severely collapsed bone. Of the malignant compression fractures, two sclerotic lesions had false-negative findings with FF values greater than the cutoff of 5.26% (Fig. 6). Zajick, Jr. et al. [5] found that some metastatic lesions had large decreases in signal intensity on opposed-phase images, mimicking benign bone sclerotic lesions (thus of lower signal intensity). Kim et al. [12] also reported that two sclerotic metastatic lesions with high FFs were erroneously diagnosed as benign lesions. lthough those authors did not seek to completely explain the reasons for such observations, we believe that the two sclerotic lesions exhibiting falsenegative results in our study mimic those described by Kim et al. However, another sclerotic lesion was still correctly diagnosed with low fracture FF (3.47%). The different results may have been caused by the degree of bone density. On CT scans two false-negative lesions had high attenuation (450 and 686 HU), but the other, true-positive lesion had relatively low attenuation compared with two falsenegative lesions (264 HU). In chemical-shift imaging, trabecular bone in general shortens the T2* of water and fat components, inducing rapid decay of the measured gradient-echo signal intensity with TE. Moreover, in osteosclerotic lesions, the T2* decay effect is more severe owing to the presence of dense trabecular bone, and noise-related bias may also be more severe because of the low fat content [11, 24]. However, our hypothesis should be evaluated in a further study with larger sample size. The signal intensity of focal lesions has commonly been compared with that of an intervertebral disk (internal standard) on conventional T1-weighted images to distinguish malignant from acute osteoporotic compression fractures. However, we found that neither T1 signal intensity nor lesionto-disk ratio was useful for this purpose. In contrast to our data, results of previous quantitative analytical studies showed that lesional T1 signal intensity afforded only marginal diagnostic performance (sensitivity, %; specificity, %) when used to differentiate benign from malignant lesions [12, 22, 25]. This may be because almost all of the fractures in our study had low T1 signal intensity attributable to prominent bone marrow edema in patients with acute osteoporotic compression fractures. ll previous relevant studies included heterogeneous patient groups with acute and chronic compression fracture, whereas our study included only patients with acute compression fracture. Limitations First, the small number of subjects may have biased our results. Second, not all assessed lesions were pathologically proven. It is clinically unethical to biopsy lesions that are very probably either malignant or benign. Third, we did not evaluate the diagnostic performances of multiple readers with varying levels of experience when they sought to distinguish acute osteoporotic from malignant compression fractures. Clinically, FF maps may aid less experienced radiologists in differentiating acute osteoporotic from malignant compression fractures. Fourth, although we were careful to draw ROIs in locations that best reflected the lesion characteristics, the readers made subjective decisions on where the outer margins of lesions lay and which intensity threshold to adopt. Moreover, the use of ROIs, as opposed to volumes of interest, may create bias. Fat is heterogeneously distributed within a bone lesion. However, we defined ROIs on three sagittal images of each lesion. Fifth, we did not compare the six-echo Dixon method with other novel techniques, such as the DWI sequences used to differentiate benign and malignant lesions. Sixth, we did not perform subgroup analysis by type of primary tumor or characteristics of the metastatic lesion (e.g., osteoblastic or osteolytic metastasis). Seventh, patients with acute osteoporotic fractures were significantly older than those with malignant compression fractures. Osteoporotic fractures are more common among the elderly as bone mineral density decreases. Studies [26, 27] have shown linear increases in vertebral marrow fat content with aging. The age difference in this study may thus have influenced our results. In addition, the FFs of normal bone marrow differed significantly between the two groups (68.33% vs 56.35%; p = 0.037). To overcome these problems, we also measured FF ratio (fracture FF divided by FF of normal bone marrow) to standardize the fat content of the lesion and performed the multivariate analysis. Eighth, the confounding effect of systemic chemotherapy was not corrected. fter chemotherapy or radiation therapy, fatty transformation generally occurs in the bone marrow [6, 28], and it may increase the FF value in the vertebral body. In terms of radiation therapy, there was no significant difference in the number of patients who underwent radiation therapy between malignant and osteoporotic fracture groups, removing the effect of radiotherapy in our study. The proportion of patients undergoing chemotherapy was statistically higher in malignant group, which may have increased the FF value in the malignant fracture group. However, our results showed that the FF was still lower in the malignant compression fracture group, because the effect of chemotherapy was not so strong as to affect the results. Finally, we did not explore whether the FF maps could also be used to improve the diagnostic accuracy of morphologic MRI. We used FF map evaluation as a stand-alone method. In the future, any additional or adjunct utility of FF mapping should be evaluated in parallel with morphologic or other novel MRI techniques. Conclusion The FFs and FF ratios obtained from sixecho Dixon sequencing can be used to differentiate acute osteoporotic compression fractures from malignant compression fractures. mong the various features of these fractures, fracture FF was the only statistically significant factor for distinguishing acute osteoporotic and malignant compression fractures in our multivariate analysis. References 1. Jung HS, Jee WH, McCauley TR, Ha KY, Choi KH. Discrimination of metastatic from acute osteoporotic compression spinal fractures with MR imaging. RadioGraphics 2003; 23: Maccauro G, Spinelli MS, Mauro S, Perisano C, Graci C, Rosa M. Physiopathology of spine metastasis. Int J Surg Oncol 2011; 2011: n HS, ndreshak TG, Nguyen C, Williams, Daniels D. Can we distinguish between benign versus malignant compression fractures of the spine by magnetic resonance imaging? Spine 1995; 20: aur, Stabler, rbogast S, Duerr HR, artl R, Reiser M. cute osteoporotic and neoplastic vertebral compression fractures: fluid sign at MR 1338 JR:209, December 2017

9 MRI of Vertebral Fractures imaging. Radiology 2002; 225: map using a three-echo volume interpolated zmann DR. enign versus pathologic compres- 5. Zajick DC Jr, Morrison W, Schweitzer ME, breath-hold gradient echo Dixon sequence. Kore- sion fractures of vertebral bodies: assessment with Parellada J, Carrino J. enign and malignant an J Radiol 2014; 15: conventional spin-echo, chemical-shift, and STIR processes: normal values and differentiation with 13. Fischer M, Nanz D, Shimakawa, et al. Quanti- MR imaging. Radiology 1990; 174: chemical shift MR imaging in vertebral marrow. fication of muscle fat in patients with low back 21. Sung JK, Jee WH, Jung JY, et al. Differentiation Radiology 2005; 237: pain: comparison of multi-echo MR imaging with of acute osteoporotic and malignant compression 6. Erly WK, Oh ES, Outwater EK. The utility of inphase/opposed-phase imaging in differentiating malignancy from acute benign compression fractures of the spine. JNR 2006; 27: Ragab Y, Emad Y, Gheita T, et al. Differentiation of osteoporotic and neoplastic vertebral fractures by chemical shift in-phase and out-of-phase MR imaging. Eur J Radiol 2009; 72: Zidan DZ, Habib L, Chalabi N. Quantitative chemical-shift MR imaging cutoff value: benign versus malignant vertebral compression: initial experience. Egypt J Radiol Nucl Med 2014; 45: Disler DG, McCauley TR, Ratner LM, Kesack CD, Cooper J. In-phase and out-of-phase MR imaging of bone marrow: prediction of neoplasia based on the detection of coexistent fat and water. JR 1997; 169: O Regan DP, Callaghan MF, Wylezinska-rridge M, et al. Liver fat content and T2*: simultaneous measurement by using breath-hold multiecho MR imaging at 3.0 T feasibility 1. Radiology 2008; 247: Karampinos DC, Melkus G, aum T, auer JS, Rummeny EJ, Krug R. one marrow fat quantification in the presence of trabecular bone: initial comparison between water-fat imaging and singlevoxel MRS. Magn Reson Med 2014; 71: Kim YP, Kannengiesser S, Paek MY, et al. Differentiation between focal malignant marrow-replacing lesions and benign red marrow deposition of the spine with T2*-corrected fat-signal fraction single-voxel MR spectroscopy. Radiology 2013; 266: Kühn JP, Jahn C, Hernando D, et al. T1 bias in chemical shift-encoded liver fat-fraction: role of the flip angle. J Magn Reson Imaging 2014; 40: Liu CY, McKenzie C, Yu H, rittain JH, Reeder S. Fat quantification with IDEL gradient echo imaging: correction of bias from T1 and noise. Magn Reson Med 2007; 58: Yu H, Shimakawa, McKenzie C, rodsky E, rittain JH, Reeder S. Multiecho water-fat separation and simultaneous R2* estimation with multifrequency fat spectrum modeling. Magn Reson Med 2008; 60: Hines CD, Yu H, Shimakawa, McKenzie C, rittain JH, Reeder S. T1 independent, T2* corrected MRI with accurate spectral modeling for quantification of fat: validation in a fat-water- SPIO phantom. J Magn Reson Imaging 2009; 30: Yoo YH, Kim HS, Lee YH, et al. Comparison of multi-echo Dixon methods with volume interpolated breath-hold gradient echo magnetic resonance imaging in fat-signal fraction quantification of paravertebral muscle. Korean J Radiol 2015; 16: Yuh WT, Zachar CK, arloon TJ, Sato Y, Sickels WJ, Hawes DR. Vertebral compression fractures: distinction between benign and malignant causes with MR imaging. Radiology 1989; 172: aker LL, Goodman S, Perkash I, Lane, En- fractures of the spine: use of additive qualitative and quantitative axial diffusion-weighted MR imaging to conventional MR imaging at 3.0 T. Radiology 2014; 271: Zhao J, Krug R, Xu D, Lu Y, Link TM. MRI of the spine: image quality and normal-neoplastic bone marrow contrast at 3 T versus 1.5 T. JR 2009; 192: Yoo HJ, Hong SH, Kim DH, et al. Measurement of fat content in vertebral marrow using a modified Dixon sequence to differentiate benign from malignant processes. J Magn Reson Imaging 2017; 45: Karampinos DC, Ruschke S, Dieckmeyer M, et al. Modeling of T2* decay in vertebral bone marrow fat quantification. NMR iomed 2015; 28: Carroll KW, Feller JF, Tirman PF. Useful internal standards for distinguishing infiltrative marrow pathology from hematopoietic marrow at MRI. J Magn Reson Imaging 1997; 7: Schellinger D, Lin CS, Fertikh D, et al. Normal lumbar vertebrae: anatomic, age, and sex variance in subjects at proton MR spectroscopy initial experience. Radiology 2000; 215: Kugel H, Jung C, Schulte O, Heindel W. ge-and sex-specific differences in the 1 H-spectrum of vertebral bone marrow. J Magn Reson Imaging 2001; 13: Daldrup-Link HE, Henning T, Link TM. MR imaging of therapy-induced changes of bone marrow. Eur Radiol 2007; 17: JR:209, December