- Email: [email protected]
Study of single voxel 1H MR spectroscopy of bone tumors: Differentiation of benign from malignant tumors
European Journal of Radiology 82 (2013) 2124–2128
Contents lists available at ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Study of single voxel 1 H MR spectroscopy of bone tumors: Differentiation of benign from malignant tumors Jing Zhang a , Kebin Cheng a , Yi Ding b , Wei Liang a , Yi Ding c , Daniel Vanel d , Xiaoguang Cheng a,∗ a
Department of Radiology, Beijing Jishuitan Hospital, Beijing, China Department of Orthopaedic Oncology, Beijing Jishuitan Hospital, Beijing, China Department of Pathology, Beijing Jishuitan Hospital, Beijing, China d Rizzoli Institute, Bologna, Italy b c
a r t i c l e
i n f o
Keywords: Bone tumor Proton magnetic resonance spectroscopy Single voxel spectroscopy
a b s t r a c t Objective: To evaluate the clinical application of single voxel 1 H MRS in the discrimination of benign and malignant bone tumors. Materials and methods: Eighty-three patients (64 male, 19 female), presenting with a bone tumor, were examined on a 1.5 T MRI scanner. Using pathological results as a gold standard, there were 34 benign and 49 malignant tumors. After plain MRI scans, a 3D fast SPGR sequence was used for dynamic contrastenhanced scanning. Dynamic images were transferred to the workstation, where the region of maximal enhancement was identified for prescription of the 1 H MRS sequence. Single-voxel 1 H MRS was then performed with the probe-p sequence, TR/TE = 1500/110 ms, VOI ranging from 14.4 mm × 7.3 mm × 20.2 mm to 27.9 mm × 25.5 mm × 20.1 mm, automatic shimming and water suppression, 15 min post-contrast. For control purposes, the 3rd lumbar spine vertebral body of six patients having lumbar disc herniation (LDH) without systemic disease was examined with 1 H MRS of normal bone marrow. The static contrast enhancement scan was used for these LDH patients. Conversion of raw MR signal to an MR spectrum was performed using SAGE 7. Cho/Lip (choline/lipids) peak height ratios were calculated. ROC curve analysis was used to determine the cut-off of Cho/Lip ratio for discrimination. Results: For malignant tumors, one resonance at 3.30–3.19 ppm attributed to choline and another at 1.14–1.55 ppm attributed to lipid were detected. With normal bone marrow and most benign tumors, no choline signal was detected. Choline was only found in six benign lesions. With a threshold for Cho/Lip peak height ratio of 0.2, the area under ROC curve was 0.819. The corresponding sensitivity and specificity of 1 H MRS were 76% and 88%. Conclusions: Single voxel 1 H MRS can help in discriminating benign and malignant bone tumors. © 2011 Published by Elsevier Ireland Ltd.
1. Introduction Proton (1 H) magnetic resonance spectroscopy (MRS) is an assay that uses magnetic resonance and chemical shift to quantitatively measure specific compounds in tissues. Its emergence has advanced imaging from gross morphology to metabolism. Normal bone tissues consist of a cortex and marrow, which contains lipids and water. In contrast, skeletal lesions display ossification, calcification, necrosis and hemorrhage. Difficulties in shimming of MRS occur due to the intensive nonlinear changes in the magnetic fields of these sites. The spectral peaks become widened, overlapped and undifferentiated, which interferes with the detection of metabolites. The characteristics of the skeletal system and its disorders
∗ Corresponding author. E-mail address: [email protected] (X. Cheng). 0720-048X/$ – see front matter © 2011 Published by Elsevier Ireland Ltd. doi:10.1016/j.ejrad.2011.11.033
thus restrict to some extent the use of MRS on the skeletal system. Recently, the feasibility of using 1 H MRS for the skeletal system has been demonstrated. Current skeletal 1 H MRS studies focus primarily on the diagnosis of tumors and osteoporosis [1–6]. In this study, we evaluated single-voxel 1 H MRS for the qualitative diagnosis of bone tumors, and further investigated the cut-off value for Cho/Lip peak height ratio for quantitative diagnosis.
2. Materials and methods Eighty-three subjects were enrolled in this study, which was performed at our Orthopedic Oncology Unit from May 2007 to July 2009. There were 64 males and 19 females, aged 7–76 years. None of the subjects had received biopsies or treatment before the MRI examination. The 83 lesions were verified with needle biopsy or surgical pathology (34 benign, 49 malignant). The pathological results of the patients are summarized in Table 1. To obtain control
J. Zhang et al. / European Journal of Radiology 82 (2013) 2124–2128 Table 1 Pathological results.
2125
Table 2 Qualitative diagnoses results of tumors using MRS.
Pathological results
Case number
Pathological results
CHO:LIP < 0.2
CHO:LIP ≥ 0.2
Benign Giant cell tumor of bone Chondroblastoma Chondromyxoma fibroma Eosinophilic granuloma Enchondroma Hemangioma Neurilemmoma Hyperosteogeny Malignant Osteosarcoma Chondrosarcoma Ewing’s sarcoma Chordoma Metastatic tumors Lymphoma Spindle cell sarcoma Malignant fibrous histiocytoma Myeloma Malignant non-small cell tumor Unclassified malignant tumor
34 20 3 1 2 1 1 1 1 49 17 5 4 3 9 2 2 2 1 1 3
Benign Malignant
30 11
4 38
data in normal osseous tissues, the L3 vertebral bodies of six subjects were examined by 1 H MRS. These patients were generally healthy and required contrast-enhanced scanning due to intervertebral disc herniations. Informed consent was obtained from all subjects. Scanning was performed with a 1.5 T MR scanner (Excite, General Electric, Milwaukee, USA). Routine MRI plain scans consisted of axial SE T1WI and FSE T2WI, sagittal or coronal T2WI with fat suppression. Dynamic contrast-enhanced scanning was performed with the fast acquisition with multiphase enhanced fast GRE (FAME) and the following parameters: TR = 6 ms, TE = 1.4 ms, FA = 15◦ , slice thickness = 3–10 mm and matrix = 288 × 160. Fifteen minutes after injection of contrast media, the singlevoxel 1 H MRS examination was initiated. The location and size of the voxel depended on the location and size of the maximum slope of increase (MSI) area in the dynamic enhancement. The voxel sizes for the patients ranged from 14.4 mm × 7.3 mm × 20.2 mm to 27.9 mm × 25.5 mm × 20.1 mm. PRESS sequences were used to allow for spatial localization of the 1 H MRS voxel, with the following parameters: TR = 1500 ms, TE = 110 ms, and NEX = 8. The sequences had a total of 128 acquisition times, with 4096 spectral data points at a frequency of 2500 Hz and a scanning time of 228 s. The automatic shimming and water suppression were performed before the MRS scanning. The bandwidths of all the patients were below 20 Hz, while the effectiveness of water suppression was all over 80%. The phases and baselines of the data were adjusted by using the GE SAGE 7 software on the workstation, and the peak heights and areas were automatically generated for the metabolites. A Chi square (Pearson) was used to assess the difference of CHO between benign and malignant tumors with SPSS v15.0. A p value less than 0.05 was considered statistically significant. ROC curve analysis was performed using the height ratio of the choline peak (CHO) to the lipid peak (LIP) to evaluate sensitivity and specificity for distinguishing benign from malignant bone tumors. A Chi square (McNemar test) was used to compare the diagnostic measure of MRS with pathology. A p value less than 0.05 was considered statistically significant. A kappa value was used to evaluate the agreement between MRS and pathology. For the subjects with enhanced MRI examinations due to intervertebral disc herniations (n = 6), the routine MRI plain scan was followed by a contrast-enhanced scan. The MRS examination and post-processing procedures were identical to those with bone tumors.
3. Results In the 1 H MRS spectra of osseous tissues in six healthy subjects, only LIP peaks were identified at 1.25–1.38 ppm. Tumor spectra showed CHO and LIP peaks or LIP peaks only at 3.30–3.19 ppm (CHO) and 1.14–1.55 ppm (lipid), with the ratio of signal to noise over 3. Among 34 benign tumors, CHO peaks were identified in 6 subjects (Fig. 1). Among 49 malignant tumors, CHO peaks were not identified in 11 subjects (Fig. 2). Pearson Chi square showed a significantly different presence of CHO between benign and malignant tumors (2 = 28.917, p < 0.05). A more quantitative analysis was critical for the differentiation between benign and malignant tumors because CHO peaks were also identified in the 1 H MRS spectra of benign tumors. The analysis of ROC curves showed that, for a threshold value of the height ratio of CHO peak to LIP peak of 0.2, giving an area under the ROC curve of 0.819, the sensitivity for diagnosis of bone tumors was 76% and specificity was 88% (Fig. 3). A McNemar test showed no significant difference between MRS (height ratio of CHO peak to LIP peak, threshold = 0.2) and pathology for discrimination of benign and malignant bone tumors (p = 0.118). The kappa value was 0.628, which demonstrated a good agreement between MRS and pathology. The diagnostic results with the height ratio of the CHO peak to LIP peak (threshold 0.2) are summarized in Table 2. In four patients histologically diagnosed with benign tumors, where CHO:LIP ≥ 0.2 on MRS, the pathological examinations identified giant cell tumors. All of giant cell tumor with CHO:LIP ≥ 0.2 had no aggressive appearance under microscope. The reason of which is not clear at present. However, in 11 patients diagnosed with malignant tumors from pathology, where CHO:LIP < 0.2 on MRS, the pathological examinations identified osteosarcoma (n = 7), lymphoma (n = 1), spindle cell sarcoma (n = 2), and Ewing’s sarcoma (n = 1). These cases were retrospectively reviewed. The ROI of MRS did not match precisely with the ROI defined by dynamic contrastenhanced scan in two patients with osteosarcoma. This result is likely due to a minimal shift in position during the MRS scanning. The positions of voxels were incorrectly selected in three lesions; specifically, the MRS voxel did not match the maximum slope of increased area in the dynamic enhancement. The misdiagnoses of six patients may be due to the ROI of MRS containing normal tissues surrounding the lesion. 4. Discussion Although previous studies have demonstrated the use of 1 H MRS on the skeletal system, it was still extremely difficult to generate MRS spectra of diagnostic quality. One reason is that the diagnoses of various disorders required variable MRS scanning parameters. The spatial localization techniques commonly used in 1 H MRS consist of the stimulated echo acquisition method (STEAM), point resolved selective spectroscopy (PRESS) and chemical shift imaging (CSI). The MRS methods include single voxel spectroscopy (SVS) and multiple voxel spectroscopy (MVS). Due to the intrinsic T1 and T2 relaxation times of metabolites, the spectral data are dependent on the scan parameters. Longer TRs are associated with higher spectral SNRs, although requiring longer acquisition times. Optimal TRs, therefore, range from 1000 to 1500 ms. The selection of TE depends on the 1 H MRS examination target. For tumors, the
2126
J. Zhang et al. / European Journal of Radiology 82 (2013) 2124–2128
Fig. 1. A 32-year-old male with a right femoral trochanteric chondroblastoma. (A) The lesion manifested as moderate signals on axial T1WI; and moderate signals and focal high signals on coronal T2WI with fat suppression (B); (C) in the dynamic contrast-enhanced slope map, the layer with the maximum slope of increase was selected to set the 1 H MRS voxel; and (D) only a LIP peak was identified in the 1 H MRS images of the lesion.
identification of CHO peaks is used to differentiate benign from malignant tumors. Moderate TEs, which favor the identification of CHO peaks, should therefore be selected for 1 H MRS of tumors. Based on our experience, and considering the skeletal characteristics, we used TR = 1500 ms and TE = 110 ms in the single voxel PRESS sequence for the 1 H MRS examination of bone tumors. Multiple voxel spectroscopy can locate multiple voxels simultaneously, requiring larger shimming volumes. The interference of magnetic susceptibility artifacts results in difficulties in shimming in addition to those caused by the skeletal characteristics. For the 1.5 T MR scanner, it has been technically difficult to perform multiple voxel 1 H MRS in skeletal tissue [7,8]. The single voxel MRS was used in skeletal system, but with limited coverage. Therefore, the accurate localization of the voxel is deemed critical for the accuracy of single voxel MRS. Since tumors, especially malignant ones, are heterogeneous (i.e., biological activity varies across different locations), neither MRI plain scans nor routine contrastenhanced scans can identify the areas where tumors grow most actively. It has been reported that the growth of tumors is dependent on the blood supply and that the most vascularized site should be the area where tumors grow most actively [9]. Therefore, with the dynamic contrast-enhanced scans, the post-processing of data identified the variable enhancement characteristics across different locations in the tumor and objectively identified the voxel for single voxel MRS. The ROI of MRS did not match precisely with the ROI defined by dynamic contrast-enhanced scanning in two patients with osteosarcoma, and by reviewing the DCE-MRI, we found that the ROI selected did not coincide with the maximal enhanced area
for two patients with osteosarcoma and one patient with spindle cell sarcoma, compromising the accuracy of MRS. The impact of the contrast agent Gd-DTPA on MRS remains controversial [10,11]. We started the injection of the contrast agent 15 min prior to the MRS scanning to minimize the impact of the contrast agent on MRS. Moreover, even if the effect of the contrast agent is significant, it should be consistent, ensuring the comparability among subjects. In addition to voxel position, voxel size was also critical for the single voxel 1 H MRS. For the 1.5 T scanner, a voxel of 2 cm × 2 cm × 2 cm usually provides MRS spectra of high SNR. Smaller voxels decrease SNR, often making the spectra unreadable. The bone tumors contained complex components, and the homogenous area (excluding hemorrhage, necrosis, ossification and calcification) was quite limited. In the present study, although the spectra exhibited stable baselines with diagnostic quality, the MRS voxel of 2 cm × 2 cm × 2 cm resulted in false negatives for six malignant lesions in comparison with the pathological examination. Previous studies have attributed this phenomenon to a partial volume effect from normal tissues surrounding the lesion. Referring to the dynamic contrast-enhanced images, the maximum enhanced area was smaller than the ROI of the MRS voxel in those six subjects. In our experience, if the volume of maximum slope of increase is smaller than the MRS voxel volume, the spectra have poor SNR and are unreadable. For these subjects, the voxels used were beyond the maximum slope of the increase area, resulting in false negative results even though the spectra exhibited stable baselines with diagnostic quality. With an MR scanner of higher magnetic field
J. Zhang et al. / European Journal of Radiology 82 (2013) 2124–2128
2127
Fig. 2. A 56-year-old male with a left scapular metastatic tumor. (A) The lesion manifested as moderate signals on axial T1WI; (B) and as high signals on axial T2WI with fat suppression; (C) in the dynamic contrast-enhanced slope map, the layer with the maximum slope of increase was selected as the location of the 1 H MRS voxel; and (D) LIP and CHO peaks were identified in the 1 H MRS images of the lesion. The patient also exhibited an enhanced mass in his left lung.
intensity, the MRS at smaller voxels would likely result in fewer false negatives [12,13]. If the spectral quality is sufficient, the CHO peak profile can be used to differentiate benign from malignant tumors. The increased choline content in tumors is associated with abnormally increased cell membrane metabolism from the mitosis and active proliferation of tumor cells [14,15]. Our data showed that in 1 H MRS spectra
Fig. 3. ROC curves. For a threshold of the ratio of choline peak height to lipid peak height of 0.2, the area under curve was the largest (0.819), resulting in sensitivity 76%, specificity 88% (criterion 0.1 express CHO:LIP ≥ 0.1).
of normal marrow tissues, only LIP peaks were identified. CHO peaks were identified in a few benign tumors but were more frequent in malignant tumors. Differences in TE values would result in variations in the CHO:LIP height ratio for malignant tumors. Our experience has shown that shorter TE values make MRS spectral baselines more stable, while longer TE values suppress the LIP peaks, making the CHO peaks more evident. The TE set at 110 ms in the present study struck a balance between the identification of CHO peaks and the quality of MRS spectra. The analysis of ROC curves on MRS identified the peak height ratio of CHO:LIP as the primary criterion for malignant tumors, with a sensitivity of 76% and a specificity of 88%. The CHO:LIP ratios ≥ 0.2 in four patients with giant cell bone tumors are probably associated with the invasive biological behavior of the tumors. In the present study, the site of the MRS voxel was neither pathologically examined nor followed up post-operatively to determine the presence or absence of recurrence. Thus, the cause of CHO peaks in benign lesions remains unclear [16]. The reason for the 11 malignant tumors having CHO:LIP ratios < 0.2 is possibly a voxel position and/or voxel size issue. The area and height of resonance peaks are proportional to the content of metabolites, and both were used for the quantitative analysis of MRS. The area under the peak is greatly influenced by the baseline and bandwidth. The brain shimming of MRS generally requires a bandwidth below 10 Hz. However, the maximum bandwidth in the present study was up to 20 Hz, compromising the accuracy of the area under peak measurement. Thus, the resonance peak heights of the metabolites were used as quantitative measures of MRS in the present study. In previous studies, the presence or absence of the CHO peak alone was used to differentiate
2128
J. Zhang et al. / European Journal of Radiology 82 (2013) 2124–2128
benignancy from malignancy. The quantitative diagnosis of bone tumors with MRS has not yet been explored. Even with the acceptable quality of MRS spectra and the exclusion of potential artifacts interfering with the accuracy of MRS, one should be cautious in the diagnosis of bone tumors. The diagnosis should be comprehensively analyzed and evaluated in connection with clinical manifestations and imaging data. Our results showed that evident CHO peaks were identifiable in the 1 H MRS spectra of 36% (4/11) of patients with giant cell bone tumors, although the cause of this phenomenon remains unknown. However, for 17 patients with osteosarcoma, 7 patients exhibited a CHO:LIP peak height ratio of <0.2, which is likely due to the voxel position in four patients and the voxel size in three patients. In conclusion, metabolite data from the 1 H MRS may supplement conventional MRI, and assisting clinical practice. These results are preliminary, and they mix very different benign and malignant tumors. More focused studies will be needed, but even the first results may help improve diagnostic confidence. References [1] Wang CK, Li CW, Hsieh TJ, et al. Characterization of bone and soft-tissue tumors with in vivo 1 H MR spectroscopy: initial results. Radiology 2004;232: 599–605. [2] Oya N, Aoki J, Shinozaki T, et al. Preliminary study of proton magnetic resonance spectroscopy in bone and soft tissue tumors: an unassigned signal at 2.0–2.1 ppm may be a possible indicator of malignant neuroectodermal tumor. Radiat Med 2000;18(3):193–8.
[3] Zhou CX, Meng JF, Chen YM, et al. Preliminary evaluation of 1 H MR spectroscopy in studying bone and soft tissue diseases of the lower limbs. J Clin Radiol 2003;22:1035–8. [4] Fayad LM, Bluemke DA, McCarthy EF, et al. Musculoskeletal tumors: use of proton MR spectroscopic imaging for characterization. J Magn Reson Imaging 2006;23:23–8. [5] Yeung DKW, Griffith JF, Antonio GE, et al. Osteoporosis is associated with increased marrow fat content and decreased marrow fat unsaturation: a proton MR spectroscopy study. J Magn Reson Imaging 2005;22:279–85. [6] Shang W, Lin Q, Yu W, et al. Potential value of vertebral proton MR spectroscopy in determining osteoporosis. Chin J Radiol 2007;41(9):947–50. [7] Yu WJ, Du XK. Primary investigation of the value of multi-voxel 1 H-MRS in differential diagnosis of benign and malignant musculoskeletal lesions of limbs. Chin J Radiol 2007;41(3):264–8. [8] Fayad LM, Barker PB, Jacobs MA, et al. Characterization of musculoskeletal lesions on 3-T proton MR spectroscopy. AJR 2007;188:1513–20. [9] Collins DJ, Padhani AR. Dynamic magnetic resonance imaging of tumor perfusion. Approaches and biomedical challenges. IEEE Eng Med Biol Mag 2004;23:65–83. [10] Sijens PE, Oudkenk M, van k P, et al. 1 H MR spectroscopy monitoring of changes in choline peak area and line shape after Gd-contrast administration. Magn Reson lmag 1998;16:1273–80. [11] Smith JK, Kwock L, Castillo M, et al. Effects of contrast material on single volume proton MR spectroscopy. AJNR 2000;21:1084–9. [12] Bartella L, Huang W. Proton (1 H) MR spectroscopy of the breast. RadioGraphics 2007;27:241–52. [13] Qi ZH, Li CF, Li ZF, et al. Preliminary study of 3 T 1H MR spectroscopy in bone and soft tissue tumors. Chin Med J 2009;122(1):39–43. [14] Mountford C, Lean C, Malycha P, Russell P. Proton spectroscopy provides accurate pathology on biopsy and in vivo. J Magn Reson Imaging 2006;24:459–77. [15] Ackerstaff E, Glunde K, Bhujwalla ZM. Choline phospholipid metabolism: a target in cancer cells? J Cell Biochem 2003;90:525–33. [16] Sah PL, Sharma R, Kandpal H, et al. In vivo proton spectroscopy of giant cell tumor of the bone. AJR 2008;190:133–9.
Contents lists available at ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Study of single voxel 1 H MR spectroscopy of bone tumors: Differentiation of benign from malignant tumors Jing Zhang a , Kebin Cheng a , Yi Ding b , Wei Liang a , Yi Ding c , Daniel Vanel d , Xiaoguang Cheng a,∗ a
Department of Radiology, Beijing Jishuitan Hospital, Beijing, China Department of Orthopaedic Oncology, Beijing Jishuitan Hospital, Beijing, China Department of Pathology, Beijing Jishuitan Hospital, Beijing, China d Rizzoli Institute, Bologna, Italy b c
a r t i c l e
i n f o
Keywords: Bone tumor Proton magnetic resonance spectroscopy Single voxel spectroscopy
a b s t r a c t Objective: To evaluate the clinical application of single voxel 1 H MRS in the discrimination of benign and malignant bone tumors. Materials and methods: Eighty-three patients (64 male, 19 female), presenting with a bone tumor, were examined on a 1.5 T MRI scanner. Using pathological results as a gold standard, there were 34 benign and 49 malignant tumors. After plain MRI scans, a 3D fast SPGR sequence was used for dynamic contrastenhanced scanning. Dynamic images were transferred to the workstation, where the region of maximal enhancement was identified for prescription of the 1 H MRS sequence. Single-voxel 1 H MRS was then performed with the probe-p sequence, TR/TE = 1500/110 ms, VOI ranging from 14.4 mm × 7.3 mm × 20.2 mm to 27.9 mm × 25.5 mm × 20.1 mm, automatic shimming and water suppression, 15 min post-contrast. For control purposes, the 3rd lumbar spine vertebral body of six patients having lumbar disc herniation (LDH) without systemic disease was examined with 1 H MRS of normal bone marrow. The static contrast enhancement scan was used for these LDH patients. Conversion of raw MR signal to an MR spectrum was performed using SAGE 7. Cho/Lip (choline/lipids) peak height ratios were calculated. ROC curve analysis was used to determine the cut-off of Cho/Lip ratio for discrimination. Results: For malignant tumors, one resonance at 3.30–3.19 ppm attributed to choline and another at 1.14–1.55 ppm attributed to lipid were detected. With normal bone marrow and most benign tumors, no choline signal was detected. Choline was only found in six benign lesions. With a threshold for Cho/Lip peak height ratio of 0.2, the area under ROC curve was 0.819. The corresponding sensitivity and specificity of 1 H MRS were 76% and 88%. Conclusions: Single voxel 1 H MRS can help in discriminating benign and malignant bone tumors. © 2011 Published by Elsevier Ireland Ltd.
1. Introduction Proton (1 H) magnetic resonance spectroscopy (MRS) is an assay that uses magnetic resonance and chemical shift to quantitatively measure specific compounds in tissues. Its emergence has advanced imaging from gross morphology to metabolism. Normal bone tissues consist of a cortex and marrow, which contains lipids and water. In contrast, skeletal lesions display ossification, calcification, necrosis and hemorrhage. Difficulties in shimming of MRS occur due to the intensive nonlinear changes in the magnetic fields of these sites. The spectral peaks become widened, overlapped and undifferentiated, which interferes with the detection of metabolites. The characteristics of the skeletal system and its disorders
∗ Corresponding author. E-mail address: [email protected] (X. Cheng). 0720-048X/$ – see front matter © 2011 Published by Elsevier Ireland Ltd. doi:10.1016/j.ejrad.2011.11.033
thus restrict to some extent the use of MRS on the skeletal system. Recently, the feasibility of using 1 H MRS for the skeletal system has been demonstrated. Current skeletal 1 H MRS studies focus primarily on the diagnosis of tumors and osteoporosis [1–6]. In this study, we evaluated single-voxel 1 H MRS for the qualitative diagnosis of bone tumors, and further investigated the cut-off value for Cho/Lip peak height ratio for quantitative diagnosis.
2. Materials and methods Eighty-three subjects were enrolled in this study, which was performed at our Orthopedic Oncology Unit from May 2007 to July 2009. There were 64 males and 19 females, aged 7–76 years. None of the subjects had received biopsies or treatment before the MRI examination. The 83 lesions were verified with needle biopsy or surgical pathology (34 benign, 49 malignant). The pathological results of the patients are summarized in Table 1. To obtain control
J. Zhang et al. / European Journal of Radiology 82 (2013) 2124–2128 Table 1 Pathological results.
2125
Table 2 Qualitative diagnoses results of tumors using MRS.
Pathological results
Case number
Pathological results
CHO:LIP < 0.2
CHO:LIP ≥ 0.2
Benign Giant cell tumor of bone Chondroblastoma Chondromyxoma fibroma Eosinophilic granuloma Enchondroma Hemangioma Neurilemmoma Hyperosteogeny Malignant Osteosarcoma Chondrosarcoma Ewing’s sarcoma Chordoma Metastatic tumors Lymphoma Spindle cell sarcoma Malignant fibrous histiocytoma Myeloma Malignant non-small cell tumor Unclassified malignant tumor
34 20 3 1 2 1 1 1 1 49 17 5 4 3 9 2 2 2 1 1 3
Benign Malignant
30 11
4 38
data in normal osseous tissues, the L3 vertebral bodies of six subjects were examined by 1 H MRS. These patients were generally healthy and required contrast-enhanced scanning due to intervertebral disc herniations. Informed consent was obtained from all subjects. Scanning was performed with a 1.5 T MR scanner (Excite, General Electric, Milwaukee, USA). Routine MRI plain scans consisted of axial SE T1WI and FSE T2WI, sagittal or coronal T2WI with fat suppression. Dynamic contrast-enhanced scanning was performed with the fast acquisition with multiphase enhanced fast GRE (FAME) and the following parameters: TR = 6 ms, TE = 1.4 ms, FA = 15◦ , slice thickness = 3–10 mm and matrix = 288 × 160. Fifteen minutes after injection of contrast media, the singlevoxel 1 H MRS examination was initiated. The location and size of the voxel depended on the location and size of the maximum slope of increase (MSI) area in the dynamic enhancement. The voxel sizes for the patients ranged from 14.4 mm × 7.3 mm × 20.2 mm to 27.9 mm × 25.5 mm × 20.1 mm. PRESS sequences were used to allow for spatial localization of the 1 H MRS voxel, with the following parameters: TR = 1500 ms, TE = 110 ms, and NEX = 8. The sequences had a total of 128 acquisition times, with 4096 spectral data points at a frequency of 2500 Hz and a scanning time of 228 s. The automatic shimming and water suppression were performed before the MRS scanning. The bandwidths of all the patients were below 20 Hz, while the effectiveness of water suppression was all over 80%. The phases and baselines of the data were adjusted by using the GE SAGE 7 software on the workstation, and the peak heights and areas were automatically generated for the metabolites. A Chi square (Pearson) was used to assess the difference of CHO between benign and malignant tumors with SPSS v15.0. A p value less than 0.05 was considered statistically significant. ROC curve analysis was performed using the height ratio of the choline peak (CHO) to the lipid peak (LIP) to evaluate sensitivity and specificity for distinguishing benign from malignant bone tumors. A Chi square (McNemar test) was used to compare the diagnostic measure of MRS with pathology. A p value less than 0.05 was considered statistically significant. A kappa value was used to evaluate the agreement between MRS and pathology. For the subjects with enhanced MRI examinations due to intervertebral disc herniations (n = 6), the routine MRI plain scan was followed by a contrast-enhanced scan. The MRS examination and post-processing procedures were identical to those with bone tumors.
3. Results In the 1 H MRS spectra of osseous tissues in six healthy subjects, only LIP peaks were identified at 1.25–1.38 ppm. Tumor spectra showed CHO and LIP peaks or LIP peaks only at 3.30–3.19 ppm (CHO) and 1.14–1.55 ppm (lipid), with the ratio of signal to noise over 3. Among 34 benign tumors, CHO peaks were identified in 6 subjects (Fig. 1). Among 49 malignant tumors, CHO peaks were not identified in 11 subjects (Fig. 2). Pearson Chi square showed a significantly different presence of CHO between benign and malignant tumors (2 = 28.917, p < 0.05). A more quantitative analysis was critical for the differentiation between benign and malignant tumors because CHO peaks were also identified in the 1 H MRS spectra of benign tumors. The analysis of ROC curves showed that, for a threshold value of the height ratio of CHO peak to LIP peak of 0.2, giving an area under the ROC curve of 0.819, the sensitivity for diagnosis of bone tumors was 76% and specificity was 88% (Fig. 3). A McNemar test showed no significant difference between MRS (height ratio of CHO peak to LIP peak, threshold = 0.2) and pathology for discrimination of benign and malignant bone tumors (p = 0.118). The kappa value was 0.628, which demonstrated a good agreement between MRS and pathology. The diagnostic results with the height ratio of the CHO peak to LIP peak (threshold 0.2) are summarized in Table 2. In four patients histologically diagnosed with benign tumors, where CHO:LIP ≥ 0.2 on MRS, the pathological examinations identified giant cell tumors. All of giant cell tumor with CHO:LIP ≥ 0.2 had no aggressive appearance under microscope. The reason of which is not clear at present. However, in 11 patients diagnosed with malignant tumors from pathology, where CHO:LIP < 0.2 on MRS, the pathological examinations identified osteosarcoma (n = 7), lymphoma (n = 1), spindle cell sarcoma (n = 2), and Ewing’s sarcoma (n = 1). These cases were retrospectively reviewed. The ROI of MRS did not match precisely with the ROI defined by dynamic contrastenhanced scan in two patients with osteosarcoma. This result is likely due to a minimal shift in position during the MRS scanning. The positions of voxels were incorrectly selected in three lesions; specifically, the MRS voxel did not match the maximum slope of increased area in the dynamic enhancement. The misdiagnoses of six patients may be due to the ROI of MRS containing normal tissues surrounding the lesion. 4. Discussion Although previous studies have demonstrated the use of 1 H MRS on the skeletal system, it was still extremely difficult to generate MRS spectra of diagnostic quality. One reason is that the diagnoses of various disorders required variable MRS scanning parameters. The spatial localization techniques commonly used in 1 H MRS consist of the stimulated echo acquisition method (STEAM), point resolved selective spectroscopy (PRESS) and chemical shift imaging (CSI). The MRS methods include single voxel spectroscopy (SVS) and multiple voxel spectroscopy (MVS). Due to the intrinsic T1 and T2 relaxation times of metabolites, the spectral data are dependent on the scan parameters. Longer TRs are associated with higher spectral SNRs, although requiring longer acquisition times. Optimal TRs, therefore, range from 1000 to 1500 ms. The selection of TE depends on the 1 H MRS examination target. For tumors, the
2126
J. Zhang et al. / European Journal of Radiology 82 (2013) 2124–2128
Fig. 1. A 32-year-old male with a right femoral trochanteric chondroblastoma. (A) The lesion manifested as moderate signals on axial T1WI; and moderate signals and focal high signals on coronal T2WI with fat suppression (B); (C) in the dynamic contrast-enhanced slope map, the layer with the maximum slope of increase was selected to set the 1 H MRS voxel; and (D) only a LIP peak was identified in the 1 H MRS images of the lesion.
identification of CHO peaks is used to differentiate benign from malignant tumors. Moderate TEs, which favor the identification of CHO peaks, should therefore be selected for 1 H MRS of tumors. Based on our experience, and considering the skeletal characteristics, we used TR = 1500 ms and TE = 110 ms in the single voxel PRESS sequence for the 1 H MRS examination of bone tumors. Multiple voxel spectroscopy can locate multiple voxels simultaneously, requiring larger shimming volumes. The interference of magnetic susceptibility artifacts results in difficulties in shimming in addition to those caused by the skeletal characteristics. For the 1.5 T MR scanner, it has been technically difficult to perform multiple voxel 1 H MRS in skeletal tissue [7,8]. The single voxel MRS was used in skeletal system, but with limited coverage. Therefore, the accurate localization of the voxel is deemed critical for the accuracy of single voxel MRS. Since tumors, especially malignant ones, are heterogeneous (i.e., biological activity varies across different locations), neither MRI plain scans nor routine contrastenhanced scans can identify the areas where tumors grow most actively. It has been reported that the growth of tumors is dependent on the blood supply and that the most vascularized site should be the area where tumors grow most actively [9]. Therefore, with the dynamic contrast-enhanced scans, the post-processing of data identified the variable enhancement characteristics across different locations in the tumor and objectively identified the voxel for single voxel MRS. The ROI of MRS did not match precisely with the ROI defined by dynamic contrast-enhanced scanning in two patients with osteosarcoma, and by reviewing the DCE-MRI, we found that the ROI selected did not coincide with the maximal enhanced area
for two patients with osteosarcoma and one patient with spindle cell sarcoma, compromising the accuracy of MRS. The impact of the contrast agent Gd-DTPA on MRS remains controversial [10,11]. We started the injection of the contrast agent 15 min prior to the MRS scanning to minimize the impact of the contrast agent on MRS. Moreover, even if the effect of the contrast agent is significant, it should be consistent, ensuring the comparability among subjects. In addition to voxel position, voxel size was also critical for the single voxel 1 H MRS. For the 1.5 T scanner, a voxel of 2 cm × 2 cm × 2 cm usually provides MRS spectra of high SNR. Smaller voxels decrease SNR, often making the spectra unreadable. The bone tumors contained complex components, and the homogenous area (excluding hemorrhage, necrosis, ossification and calcification) was quite limited. In the present study, although the spectra exhibited stable baselines with diagnostic quality, the MRS voxel of 2 cm × 2 cm × 2 cm resulted in false negatives for six malignant lesions in comparison with the pathological examination. Previous studies have attributed this phenomenon to a partial volume effect from normal tissues surrounding the lesion. Referring to the dynamic contrast-enhanced images, the maximum enhanced area was smaller than the ROI of the MRS voxel in those six subjects. In our experience, if the volume of maximum slope of increase is smaller than the MRS voxel volume, the spectra have poor SNR and are unreadable. For these subjects, the voxels used were beyond the maximum slope of the increase area, resulting in false negative results even though the spectra exhibited stable baselines with diagnostic quality. With an MR scanner of higher magnetic field
J. Zhang et al. / European Journal of Radiology 82 (2013) 2124–2128
2127
Fig. 2. A 56-year-old male with a left scapular metastatic tumor. (A) The lesion manifested as moderate signals on axial T1WI; (B) and as high signals on axial T2WI with fat suppression; (C) in the dynamic contrast-enhanced slope map, the layer with the maximum slope of increase was selected as the location of the 1 H MRS voxel; and (D) LIP and CHO peaks were identified in the 1 H MRS images of the lesion. The patient also exhibited an enhanced mass in his left lung.
intensity, the MRS at smaller voxels would likely result in fewer false negatives [12,13]. If the spectral quality is sufficient, the CHO peak profile can be used to differentiate benign from malignant tumors. The increased choline content in tumors is associated with abnormally increased cell membrane metabolism from the mitosis and active proliferation of tumor cells [14,15]. Our data showed that in 1 H MRS spectra
Fig. 3. ROC curves. For a threshold of the ratio of choline peak height to lipid peak height of 0.2, the area under curve was the largest (0.819), resulting in sensitivity 76%, specificity 88% (criterion 0.1 express CHO:LIP ≥ 0.1).
of normal marrow tissues, only LIP peaks were identified. CHO peaks were identified in a few benign tumors but were more frequent in malignant tumors. Differences in TE values would result in variations in the CHO:LIP height ratio for malignant tumors. Our experience has shown that shorter TE values make MRS spectral baselines more stable, while longer TE values suppress the LIP peaks, making the CHO peaks more evident. The TE set at 110 ms in the present study struck a balance between the identification of CHO peaks and the quality of MRS spectra. The analysis of ROC curves on MRS identified the peak height ratio of CHO:LIP as the primary criterion for malignant tumors, with a sensitivity of 76% and a specificity of 88%. The CHO:LIP ratios ≥ 0.2 in four patients with giant cell bone tumors are probably associated with the invasive biological behavior of the tumors. In the present study, the site of the MRS voxel was neither pathologically examined nor followed up post-operatively to determine the presence or absence of recurrence. Thus, the cause of CHO peaks in benign lesions remains unclear [16]. The reason for the 11 malignant tumors having CHO:LIP ratios < 0.2 is possibly a voxel position and/or voxel size issue. The area and height of resonance peaks are proportional to the content of metabolites, and both were used for the quantitative analysis of MRS. The area under the peak is greatly influenced by the baseline and bandwidth. The brain shimming of MRS generally requires a bandwidth below 10 Hz. However, the maximum bandwidth in the present study was up to 20 Hz, compromising the accuracy of the area under peak measurement. Thus, the resonance peak heights of the metabolites were used as quantitative measures of MRS in the present study. In previous studies, the presence or absence of the CHO peak alone was used to differentiate
2128
J. Zhang et al. / European Journal of Radiology 82 (2013) 2124–2128
benignancy from malignancy. The quantitative diagnosis of bone tumors with MRS has not yet been explored. Even with the acceptable quality of MRS spectra and the exclusion of potential artifacts interfering with the accuracy of MRS, one should be cautious in the diagnosis of bone tumors. The diagnosis should be comprehensively analyzed and evaluated in connection with clinical manifestations and imaging data. Our results showed that evident CHO peaks were identifiable in the 1 H MRS spectra of 36% (4/11) of patients with giant cell bone tumors, although the cause of this phenomenon remains unknown. However, for 17 patients with osteosarcoma, 7 patients exhibited a CHO:LIP peak height ratio of <0.2, which is likely due to the voxel position in four patients and the voxel size in three patients. In conclusion, metabolite data from the 1 H MRS may supplement conventional MRI, and assisting clinical practice. These results are preliminary, and they mix very different benign and malignant tumors. More focused studies will be needed, but even the first results may help improve diagnostic confidence. References [1] Wang CK, Li CW, Hsieh TJ, et al. Characterization of bone and soft-tissue tumors with in vivo 1 H MR spectroscopy: initial results. Radiology 2004;232: 599–605. [2] Oya N, Aoki J, Shinozaki T, et al. Preliminary study of proton magnetic resonance spectroscopy in bone and soft tissue tumors: an unassigned signal at 2.0–2.1 ppm may be a possible indicator of malignant neuroectodermal tumor. Radiat Med 2000;18(3):193–8.
[3] Zhou CX, Meng JF, Chen YM, et al. Preliminary evaluation of 1 H MR spectroscopy in studying bone and soft tissue diseases of the lower limbs. J Clin Radiol 2003;22:1035–8. [4] Fayad LM, Bluemke DA, McCarthy EF, et al. Musculoskeletal tumors: use of proton MR spectroscopic imaging for characterization. J Magn Reson Imaging 2006;23:23–8. [5] Yeung DKW, Griffith JF, Antonio GE, et al. Osteoporosis is associated with increased marrow fat content and decreased marrow fat unsaturation: a proton MR spectroscopy study. J Magn Reson Imaging 2005;22:279–85. [6] Shang W, Lin Q, Yu W, et al. Potential value of vertebral proton MR spectroscopy in determining osteoporosis. Chin J Radiol 2007;41(9):947–50. [7] Yu WJ, Du XK. Primary investigation of the value of multi-voxel 1 H-MRS in differential diagnosis of benign and malignant musculoskeletal lesions of limbs. Chin J Radiol 2007;41(3):264–8. [8] Fayad LM, Barker PB, Jacobs MA, et al. Characterization of musculoskeletal lesions on 3-T proton MR spectroscopy. AJR 2007;188:1513–20. [9] Collins DJ, Padhani AR. Dynamic magnetic resonance imaging of tumor perfusion. Approaches and biomedical challenges. IEEE Eng Med Biol Mag 2004;23:65–83. [10] Sijens PE, Oudkenk M, van k P, et al. 1 H MR spectroscopy monitoring of changes in choline peak area and line shape after Gd-contrast administration. Magn Reson lmag 1998;16:1273–80. [11] Smith JK, Kwock L, Castillo M, et al. Effects of contrast material on single volume proton MR spectroscopy. AJNR 2000;21:1084–9. [12] Bartella L, Huang W. Proton (1 H) MR spectroscopy of the breast. RadioGraphics 2007;27:241–52. [13] Qi ZH, Li CF, Li ZF, et al. Preliminary study of 3 T 1H MR spectroscopy in bone and soft tissue tumors. Chin Med J 2009;122(1):39–43. [14] Mountford C, Lean C, Malycha P, Russell P. Proton spectroscopy provides accurate pathology on biopsy and in vivo. J Magn Reson Imaging 2006;24:459–77. [15] Ackerstaff E, Glunde K, Bhujwalla ZM. Choline phospholipid metabolism: a target in cancer cells? J Cell Biochem 2003;90:525–33. [16] Sah PL, Sharma R, Kandpal H, et al. In vivo proton spectroscopy of giant cell tumor of the bone. AJR 2008;190:133–9.