- Email: [email protected]
Improved diagnostic accuracy in differentiating malignant and benign lesions using single-voxel proton MRS of the breast at 3 T MRI
Clinical Radiology 68 (2013) e502ee510
Contents lists available at SciVerse ScienceDirect
Clinical Radiology journal homepage: www.clinicalradiologyonline.net
Improved diagnostic accuracy in differentiating malignant and benign lesions using single-voxel proton MRS of the breast at 3 T MRI S. Suppiah a, b, K. Rahmat a, *, M.N. Mohd-Shah a, C.A. Azlan a, L.K. Tan a, Y.F.A. Aziz a, A. Vijayananthan a, A.L. Wui a, C.H. Yip c a
Department of Biomedical Imaging, University Malaya Research Imaging Centre (UMRIC), Kuala Lumpur, Malaysia Department of Imaging, Faculty of Medicine and Health Sciences, Universiti Putra, Malaysia c Department of Surgery, Faculty of Medicine, University of Malaya, Kuala Lumpur, Malaysia b
article in formation Article history: Received 14 November 2012 Received in revised form 19 March 2013 Accepted 5 April 2013
AIM: To investigate the diagnostic accuracy of single-voxel proton magnetic resonance spectroscopy (SV 1H MRS) by quantifying total choline-containing compounds (tCho) in differentiating malignant from benign lesions, and subsequently, to analyse the relationship of tCho levels in malignant breast lesions with their histopathological subtypes. MATERIALS AND METHODS: A prospective study of SV 1H MRS was performed following dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) in 61 women using a 3 T MR system. All lesions (n ¼ 57) were analysed for characteristics of morphology, contrastenhancement kinetics, and tCho peak heights at SV 1H MRS that were two-times above baseline. Subsequently, the tCho in selected lesions (n ¼ 32) was quantified by calculating the area under the curve, and a tCho concentration equal to or greater than the cut-off value was considered to represent malignancy. The relationship between tCho in invasive ductal carcinomas (IDCs) and their Bloom & Richardson grading of malignancy was assessed. RESULTS: Fifty-two patients (57 lesions; 42 malignant and 15 benign) were analysed. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), of predicting malignancy were 100, 73.3, 91.3, and 100%, respectively, using DCE-MRI and 95.2, 93.3, 97.6, and 87.5%, respectively, using SV 1H MRS. The tCho cut-off for receiver operating characteristic (ROC) curve was 0.33 mmol/l. The relationship between tCho levels in malignant breast lesions with their histopathological subtypes was not statistically significant (p ¼ 0.3). CONCLUSION: Good correlation between tCho peaks and malignancy, enables SV 1H MRS to be used as a clinically applicable, simple, yet non-invasive tool for improved specificity and diagnostic accuracy in detecting breast cancer. Ó 2013 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.
Introduction Single-voxel proton magnetic resonance spectroscopy (SV H MRS) enables non-invasive assessment of the biochemical
1
* Guarantor and correspondent: K. Rahmat, Department of Biomedical Imaging, University Malaya Research Imaging Centre (UMRIC), Faculty of Medicine, University of Malaya, 50603 Kuala Lumpur, Malaysia. Tel.: þ603 79492091; fax: þ603 79494603. E-mail address: [email protected] (K. Rahmat).
composition of breast lesions. Previous studies conducted using 1.5 T MR systems1e6 reported the presence of the resonance of total choline (tCho)-containing compounds at 3.2 parts per million (ppm), which includes contributions from choline, phosphocholine, glycerophosphocholine, and taurine as reliable biomarkers of breast cancer.7 This is because choline-containing metabolites detected in breast lesions are an indicator of the increased cellular metabolism noted in malignant breast tumours.
0009-9260/$ e see front matter Ó 2013 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.crad.2013.04.002
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
Current practices include dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) as part of the standard breast imaging protocol.8 Previously, MRS breast studies at 1.5 T have shown that it is a useful adjunct to standard breast MRI protocols. More recent studies have shown MRS breast at 3 T has the added advantage of providing better spatial resolution for lesion localization.9 It also provides improved spectral resolution of metabolites of the MR spectrum for better detection of tCho peaks.10, 11 Standard breast imaging protocols enable the analysis of the morphological and kinetic patterns of benign and malignant breast lesions detected at MRI. The high sensitivity of breast MRI but its limited specificity poses a diagnostic challenge for radiologists, producing a high false-positive rate of cancer detection leading to unnecessary biopsies being performed.1,8 The ratio of mass to non-mass lesions differed significantly between true-positive (7.0 : 1.0) and false-positive findings (1.2: 1.0), giving a low positive predictive value (PPV) of 51.7% for non-mass lesions detected on standard MRI.12 DCE-MRI sensitivity has been reported to be as high as 94e100%, but with variable specificity in the range of 37e97%.13 By adding MRS to standard breast imaging, the specificity for the detection of breast cancer can be consistently improved from 70% up to 92%.4,6 High field scanning has been proven to be of use in improving diagnostic specificity.10 Although MRS breast at 3 T is gaining momentum as a useful diagnostic tool, there are numerous methods recommended for performing and interpreting it.5e7,14 This leads to uncertainty as to which protocol to be utilized for maximum diagnostic accuracy. Therefore, it poses a problem to radiologists as there is a necessity for selecting a fast, accurate, and clinically applicable protocol. Therefore, the aim of the present study was to investigate the diagnostic accuracy of SV 1H MRS at 3 T by quantifying tCho-containing compounds in differentiating malignant from benign lesions. Subsequently, the relationship between tCho levels in malignant breast lesions were compared with their histopathological subtypes.
Materials and methods Hospital ethical committee approval and written informed consent were obtained. Sixty-one women (mean age 49.7 years old, range 20e83 years old) were prospectively recruited based on findings of breast lesion(s) on clinical examination, mammography, and/or ultrasound. Sixty-five lesions were detected over 22 months and, after exclusion criteria were applied, 57 lesions in 52 patients were obtained for the study cohort (42 malignant: mean size 3.6 cm, size range 1.2e15.8 cm; 15 benign: mean size 1.9 cm, size range 0.7e3.8 cm). Nine patients were excluded due to five having erratic MRS spectra, two did not have conclusive histology results, one patient defaulted biopsy, and one patient did not have a lesion detected at MRI. Single-voxel tCho spectra of all breast lesions and normal breast tissue were obtained and correlated with the final histology results with the reference standard being surgery or core-biopsy results. The lesions were analysed for
e503
morphological and dynamic contrast enhancement kinetics characteristics. Subsequently, selected spectra of malignant and benign lesions, as well as of normal breast tissue, (n ¼ 32) were assessed for their tCho content and whether this corresponded to their histopathological findings. Inclusion criteria were women who had a breast lump detected either by clinical palpation or by radiological investigation, i.e., mammography/ultrasound, and that lesion was going to be either biopsied or excised. Exclusion criteria were women who had a contraindication to MRI, women who had undergone previous surgery/radiotherapy on the lesion-containing breast, and those who only had microcalcifications on mammogram and no solid mass lesions seen at breast ultrasound. Thirty-two spectra were analysed for tCho content after the mathematical software for tCho area under the curve quantification was acquired in the later phase of the study period. Studies were performed on a clinical 3 T MR system (SignaÒ HDx MR Systems; GE Healthcare, Milwaukee, WI, USA) with a dedicated bilateral eight-channel double breast coil using the manufacturer’s proton MRS acquisition software, breast spectroscopic examination (BREASE).
Study protocol/sequences Standard breast MRI protocol The imaging protocol comprised of unenhanced axial T1weighted, T2-weighted, and short tau inversion recovery (STIR) sequences as well as a 3D high-resolution dynamic, contrast-enhanced with T1-weighted fat-suppressed sequences. Axial T2-weighted and STIR images had 4 mm slice thickness, and 40 cm field of view (FOV). Prior to spectroscopy, six-phase contrast-enhanced MR mammography using 0.2 mmol/kg intravenous gadopentate dimeglumine (Magnevist; Schering, Berlin, Germany) was performed [10 flip angle,1.4 mm slice thickness, 40 cm FOV, 256 256 matrix size, 4.2 ms time-to-repeat (TR), and 2.5 ms time-to-echo (TE)]. Six consecutive scans were performed for 1 min each with no intersection gap. The regionof-interest (ROI) was placed on the most enhancing region of the target lesions that were detected. Subsequently, dynamic contrast enhancement kinetic curves were plotted using functional tool (FuncTool) on the manufacturerprovided software [Volume Imaging for BReast AssessmeNT (VIBRANT)] to assess for the contrast medium uptake.
MRS The contrast-enhanced images were scrutinized to identify solid, enhancing lesions suitable for spectroscopic examination. Spectroscopic VOI placement was then performed and adjusted to maximize coverage of the enhancing lesion whilst minimizing coverage of adjacent adipose tissue (minimum VOI size was 10 10 10 mm). Saturation bands were then positioned around the selected VOI to reduce signal coming from the surrounding breast tissue (Fig 1a). An automatic pre-scan was undertaken by selecting a homogeneous VOI in the enhancing lesion. Once full- width halfmaximum (FWHM) of <6 Hz was achieved, image acquisition was performed. A point-resolved spectroscopy (PRESS)
e504
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
Figure 1 MRI of a 28-year-old woman (patient 35) who presented with palpable left breast lump. (a) High-resolution contrast-enhanced image in the axial plane showed a heterogeneously enhancing lesion (red arrow) in the left breast. A VOI was placed in the enhancing solid lesion to optimize coverage of the most enhancing part and to minimize the inclusion of adjacent adipose tissue. Saturation bands were placed around the lesion before MRS data were acquired. /MR spectrum for the lesion showed lipid peak at 1.3 ppm (white arrow), choline at 3.2 ppm (yellow arrow) and water peak at 4.74 ppm (blue arrow). This lesion was revealed to be a grade 3 IDC. (b) Voxel selection and placement for MRS of normal tissue. Saturation bands were placed around normal-appearing breast tissue before MRS data were acquired for purpose of control data. /MR spectrum for normal breast tissue showing absent choline metabolite peak at 3.2 ppm (yellow arrow) and a well-suppressed water peak at 4.74 ppm (blue arrow).
sequence [TR/TE: 2000 ms/155 ms and with number of excitations (NEX) 56 times], single-shot unsuppressed spectrum was performed in the VOI. An automatic water- and fatsuppressed (AWS) sequence, which is a chemical shift selective imaging sequence (CHESS), was included to detect tCho resonance at 3.2 ppm (Fig 1a). The total time taken for all spectroscopic preparation steps and acquisitions was approximately 5e6 min. The control spectra were acquired from VOI placed in healthy-appearing glandular tissue with saturation bands placed around it. The spectra of normal breast tissues were not expected to have a choline peak at 3.2 ppm (Fig 1b).
Data interpretation For the MR spectra that were acquired, chemical shifts were corrected using a water signal at 4.74 ppm. The spectra were shifted accordingly using spectral offset manipulations. The MRI images were interpreted by two radiologists (K.R. and Y.F.) with 8e12 years of subspecialty experience in breast imaging. Objective morphological assessment of the lesions using the American College of Radiology (ACR) BIRADSÒ-MRI (Breast ImagingdReporting and Data System
for MRI) lexicon and final assessment categories from 1 to 6 for each lesion were made. The American College of Radiology’s BI-RADS-MRI lexicon,15 was specifically used as a tool to standardize the image interpretation and appropriate MRI BI-RADS category was documented and scores were given based on morphological features and kinetics characteristics scoring system.16,17 The scores were then correlated with the histopathological reference standard based on core-biopsy histology results but upgraded to surgical histopathology results where applicable. DCE-MRI characteristics were interpreted as explained by Kuhl et al.18 and correlated with BI-RADS MRI descriptors of malignancy.19 Initially, the presence of tCho in a lesion was identified by a peak height at 3.2 ppm that was two-times above baseline noise.7 Subsequently, selected spectra were analysed to calculate the area under the curve of tCho peak at 3.2 ppm using Spectroscopic Analysis for General Electric (SAGE) and MATLAB (Mathworks, Natick, MA, USA) softwares. The extracted peaks in the region of 3.2 ppm, SD 0.2 were fitted with a Voigt line-shape curve-fitting algorithm to calculate the area under the curve. Curve fits giving R2 values of 0.695e1.000 were considered to be good curve
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
fits, and R2 <0.695 were considered poor fits. For lesions with poor curve fit, any measured value of the area under the curve was considered background noise, therefore, these were considered to have no detectable tCho. The tCho levels detected on MRS peaks in these lesions were then correlated with histopathological results. The radiologists were blinded to the patient data and histopathological results at the time of analysis of the tCho peaks.
e505
Table 1 Summary of benign and malignant lesions subgroup pathology. Benign lesions Fibroadenomas Fibrocystic disease Sclerosing adenosis Intraductal papilloma
No. of lesions 6 5 2 2
Statistical analysis Total
Data were entered and tabulated into a computer spreadsheet (Excel; Microsoft, Redmond, WA, USA). Statistical analysis was performed using statistical software SPSS version 17.0 (Chicago, IL, USA). Statistical tests involving the comparison of two categorical variables, e.g., significant tCho peak detection versus malignant histology were performed using Pearson’s chi-squared test. In cases where counts in the cells of the contingency tables were low, Fisher’s exact test was used instead. For comparison of continuous variables, e.g., calculated tCho peak values, independent sample t-tests were employed, except when there were more than two groups, in which case analysis of variance (ANOVA) was used. A p-value of <0.05 was considered to be statistically significant with a confidence level set at 95%. Receiver-operator characteristic (ROC) curves were used for continuous data with a binary outcome to determine the sensitivity of data cut-off points. Kappa values for inter- and intra-observer variability were calculated and showed good agreement. For categorical variables, both intra- and interobserver agreement was good to excellent (Kappa value range ¼ 0.78e0.85, p < 0.05). For scale variables, Spearman’s correlation were used to look for intra- and interobserver agreement, and found good agreement for all variables measured (Spearman’s rho range ¼ 0.81e0.88, p < 0.05). It was also noted that in most categorical and scale variables, the intraobserver agreement yielded a higher agreement value than their interobserver counterpart.
Results Sixty-one women were recruited for the study and postprocessing was performed on 65 detected lesions. There were several lesions/spectra with suboptimal examinations (n ¼ 5) or without definite histopathology results (n ¼ 3), which were discarded from the final analysis. The five suboptimal examinations showed erratic spectrum with reduced signal to noise ratios (SNR) due to patient movement or breathing artefacts, and were discarded from the final analysis. As a result, 52 patients and 57 lesions (42 malignant and 15 benign; Table 1) were used in the final analysis. In the histopathologically proven malignant group, 30 patients underwent mastectomies, eight patients opted for breast conservative surgery, and four patients refused surgery. The MRS spectra for all lesions were initially analysed qualitatively. Forty out of the forty-two histopathologically proven malignant lesions demonstrated a tCho peak; and 14
15
Malignant lesions
No. of lesions
DCIS IDC Grade 1 IDC Grade 2 IDC Grade 3 Lobular carcinoma Malignant phyllodes Breast Osteosarcoma
6 4 15 12 2 2 1 42
out of the 15 benign lesions did not have a significant tCho peak. Fig 2 shows an example of an invasive ductal carcinoma (IDC) with suspicious morphology, demonstrated a washout pattern on kinetic curve and significant tCho on MRS. Two malignant lesions that did not demonstrate a significant choline peak were a 2.1 cm low-grade IDC with prominent ductal carcinoma in situ (DCIS) component (patient 37) and a 1.8 cm low-grade DCIS (patient 39). One benign fibroadenoma gave a false-positive finding. A significant choline peak was also detected in a malignant phyllodes tumour.20 Fig 3 shows the typical benign features and absent choline peak on the spectra of a benign fibrocystic lesion. Significant tCho peak above baseline detected by MRS produced a sensitivity, specificity, PPV, and negative predictive value (NPV) of 95.2, 93.3, 97.6, and 87.5%, respectively (Table 2). Consequently, selected spectra (n ¼ 32) were analysed for good curve fit and the area under the curve was calculated for each spectra. All malignant lesions showed good curve fit, with R2 values between 0.695 and 1.0 (Table 3) compared to benign lesions, which mostly had R2 values below 0.695 (Table 4). ROC curves showed that an R2 value of 0.695 gave a sensitivity of 100% and specificity of 94% (Fig 4). Normalized arbitrary units (nAU) for lesions with R2 values <0.695 were considered to be not applicable. Fig 5 shows the ROC curve for arbitrary units (AU) and nAU correlated with malignancy. Independent sample t-test showed that nAU values significantly correlated with malignancy (p ¼ 0.003). The mean nAU for malignant lesions was higher than the mean for non-malignant lesions (Fig 6). Lesions 20 and 29 had the highest nAU values, from patients 37 and 47, who had grade 1 IDC and grade 2 IDC, respectively. Lesion 22 (from patient 38) was a benign fibroadenoma measuring 3.8 cm, which showed a good curve fit (R2 ¼ 0.92) and relatively higher nAU value compared with the rest of the non-malignant lesions. The optimal threshold for normalized tCho level was 0.33 AU/mldequivalent to 0.33 mmol/l, which gave a good trade-off between sensitivity and specificity for the detection of malignancy. It was also noted that the mean normalized tCho value for grade 1 IDCs was the highest compared with grade 2 and grade 3 IDCs. The mean normalized tCho values for grade 1 (n ¼ 1), grade 2 (n ¼ 5), and grade 3 IDCs (n ¼ 6) were 3.9 4.07 AU/ml; 2.96 2.41 AU/ml; and 1.38 0.90 AU/ml, respectively. However, the ANOVA test did not show that the normalized
e506
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
Figure 2 A 63-year-old woman (patient 47) with grade 2 invasive ductal carcinoma. Left mammogram in the caudo-cranial (a) and medio-lateral oblique (b) views showed an irregular, lobulated, high-density mass in the left upper inner quadrant (white arrow). (c) Colour Doppler ultrasound showed a corresponding hypoechoic solid mass with several penetrating vessels at the 10 o’clock position of the left breast (white arrow). (d) Sagittal contrast-enhanced maximum intensity projection (MIP) images, showed a 5.1 cm irregular mass in the left upper quadrant (red square). (e) Kinetic curve analysis demonstrated rapid uptake with washout pattern (type 3 kinetic curve; true-positive DCE-MRI finding). (f) MRS showed elevated choline peak at 3.2 ppm (yellow arrow) (true-positive MRS finding).
tCho for the three different histopathological grades of IDCs to be statistically significant (p ¼ 0.3). The lesions were also assessed for their morphology and DCE-MRI kinetics characteristics. Lesions that had equivocal DCE findings but suspicious morphology or demonstrated washout kinetics pattern after 3 min post-contrast medium administration were interpreted as malignant.18,19 For total scoring, taking into account morphological features combined with DCE-MRI contrast wash-out, lesions with a score of >5 over 8 criteria were interpreted as malignant16,17; the sensitivity, specificity, PPV, and NPV were 100, 73.3, 91.3, and 100%, respectively (Table 2).
Discussion MRS breast at 3 T has been shown to demonstrate better diagnostic efficacy and has the advantage of better spectral resolution to detect tCho peaks; therefore, there is improved specificity and PPV for better detection of breast carcinoma.9,11 The present study achieved 93.3% specificity by qualitative analysis of tCho peak height and 86.7% specificity for tCho values above 0.33 mmol/l by quantification of the area under the curve of the tCho peak; which is compatible with prior breast MRS study results of between
88.5e100% performed by Sardanelli et al.,6 Bolan et al.,5 Tse et al.21 and Kim et al.3 SVS 1H MRS was chosen as it has a fast acquisition time of less than 6 min compared to multivoxel MRS, which generally takes longer, i.e., approximately 10 min.14 SVS 1H MRS is also easier to interpret as there is only one spectrum to analyse. The ROI is chosen carefully by selecting the most enhancing part of the breast lesion for voxel placement; therefore, giving an accurate representation of tCho level within the lesion. Using an optimized threshold of the absolute tCho peak integral expressed as arbitrary units is a simple and effective method to quantify tCho in breast lesions. A 0.941 sensitivity and 0.867 specificity was achieved with a normalized tCho equal to 0.33 AU/ml, which is comparable to a previous study conducted by Sardanelli et al.,6 which achieved 0.842 sensitivity and 0.885 specificity with a normalized tCho equal to 0.85 AU/ml. A recent MRS study at 1.5 T by Mizukoshi et al.22 in 2013 reported an tCho value of 0.61 mmol/l as an optimal cut-off point, which achieved 68% sensitivity and 79.4% specificity 22. In the present study, the optimal threshold for normalized tCho was 0.33 AU/ml, corresponding to approximately 0.33 mmol/l, and provided a good trade-off between sensitivity and specificity for the detection of malignancy. At 3 T, metabolite peaks at MRS are
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
e507
Figure 3 A 50-year-old woman (patient 27) presented with a left breast lesion that was revealed to be fibrocystic disease. (a) T2-weighted axial image showed the high signal lesion (red arrow) in the left upper outer quadrant. (b) Sagittal post-gadolinium multiplanar reformatted (MPR) image demonstrated the 3 cm lesion (red arrow) in the left upper outer quadrant. (c) Axial image of T1 fat-saturation contrast-enhanced MRI demonstrated a well-defined, fairly homogeneously enhancing oval lesion in the left upper outer quadrant (red arrow). (d) MRS showed no significant elevation of choline peak at 3.2 ppm (yellow arrow; true-negative MRS finding).
better discernible, which enables even low values of tCho peaks to be significant for malignancy. Thus, lesions with normalized tCho levels of 0.33 mmol/l should be biopsied to rule out malignancy.
The VOI was placed after administration of gadolinium, which did not adversely affect the ability to detect choline peaks in malignant lesions as tCho peaks were detected in the majority of malignant lesions. By injecting contrast medium prior to MRS, the voxel prescription was focused on the most metabolically active part of the tumour and adipose tissue was avoided. Kousi et al.9 also recommended performing MRS after gadolinium administration because it improved small lesion localization and enabled better voxel prescription. They performed MRS before and after contrast medium injection, and achieved a sensitivity/specificity of 42.8%/84.6% and 78.5%/92.0%, respectively, for detecting
Table 2 Diagnostic accuracy of using morphology and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) scoring versus magnetic resonance spectroscopy (MRS). Grouping variable
Figure 4 ROC curve using test variable of R2 for discriminating malignant from non-malignant tissue.
Scoring system Scores of >5 versus scores <5 MRS Choline peak detection versus no peak
PPV (%)
NPV (%)
Sensitivity (%)
Specificity (%)
91.3
100.0
100.0
73.3
97.6
87.5
95.2
93.3
PPV, positive predictive value; NPV, negative predictive value.
e508
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
Table 3 Quantitative analysis of malignant lesions. Lesion no.
Patient no.
1 26 2 27 4 28 6 29 8 30 9 31 11 32 13 33 17 35 20 37 23 39 25 40 27 44 28 46 29 47 30 48 31 50 Mean SD Median 25th percentile 75th percentile
Age (years)
LLD (cm)
R2
VOI (ml)
Absolute tCho Peak Integral (AU)
Normalized tCho Peak Integral (AU/ml)
Histopathology findings
54 65 53 55 65 43 65 45 45 39 59 68 31 62 63 65 32 52.5 11.7 54.0 44.0 64.0
5.8 1.8 5.4 5.2 1.8 2.0 2.6 2.5 5.8 2.1 1.8 3.1 7.5 1.4 5.1 3.2 15.8 4.6 3.5 3.2 2.1 5.8
0.98 0.91 0.97 0.98 0.91 0.71 0.98 0.92 0.90 0.75 0.76 0.77 0.89 0.99 0.99 0.81 0.94 0.88 0.11 0.91 0.77 0.98
1.75 1.89 2.59 3.42 1.89 6.40 3.42 3.14 6.00 2.98 2.94 3.69 10.02 1.31 4.36 5.38 7.36 4.11 2.26 3.42 2.77 5.69
5.25 1.62 1.39 2.49 1.62 0.95 3.14 2.86 1.28 10.8 0.94 0.85 11.51 1.30 12.2 1.97 0.14 3.61 3.95 1.97 1.12 4.20
3.28 1.02 0.92 1.66 1.02 0.59 2.09 1.91 0.85 6.78 0.63 0.57 3.05 1.30 6.92 1.23 0.09 2.04 2.00 1.30 0.74 2.57
IDC grade 2 IDC grade 1 IDC grade 3 IDC grade 3 IDC grade 1 IDC grade 2 IDC grade 2 IDC grade 2 IDC grade 3 IDC grade 1 DCIS grade 1 IDC grade 3 IDC grade 3 DCIS grade 3 IDC grade 2 IDC grade 3 Malignant phyllodes
LLD, largest lesion diameter; VOI, volume of interest; R2, best spectral curve fit; tCho, total choline containing compounds; AU, arbitrary unit.
tCho presence. In addition, an intermediate TE value of 155 ms was chosen, which provided a good trade-off between satisfactory SNR and clearly resolvable choline peaks at 3.2 ppm while minimizing possible relaxation losses and motion-related artefacts associated with longer TEs. Normalized tCho levels were detected in the range of 0.09e6.92 AU/ml within 17 malignant and one benign lesions. This finding is comparable to the study performed by Sardanelli et al.,6 which detected tCho levels between
0.38e19.8 AU/ml. It also concurs with studies by Bolan et al.5 and Roebuck et al.,1 which detected levels between 0.4e10 and 0.4e5.8 mmol/l, respectively, using phantoms of known choline concentration to study and further quantify the tCho to millimoles per litre units. Although there was some overlap of values among benign and malignant lesions (Fig 6), malignant lesions had a higher mean normalized tCho value of 2.04 2 AU/ml, compared with non-malignant lesions that had a mean
Table 4 Quantitative analysis of non-malignant lesions. Lesion no.
Patient no.
Age (years)
LLD (cm)
R2
VOI (ml)
Absolute tCho Peak Integral (AU)
Normalized tCho Peak Integral (AU/ml)
Histopathology findings
3 5 7 10 12 14 16 18 19 24 26 15 21 22 32 Mean SD Median 25th percentile 75th percentile
27 28 29 31 32 33 34 35 37 39 40 34 37 38 57
65 53 55 48 65 45 20 45 39 59 68 20 39 20 46 31.3 13.3 29.5 20.0 44.3
NA NA NA NA NA NA NA NA NA NA NA 2.3 1.4 3.8 1.3 2.2 1.2 2.1 1.4 2.7
0.04 0.59 0.03 0.51 0.21 0.48 0.69 0.02 0.23 0.07 0.29 0.69 0.17 0.92 0.55 0.58 0.31 0.62 0.27 0.86
3.49 3.69 3.42 2.98 2.66 3.14 3.69 6.00 2.24 2.94 3.69 4.56 3.45 2.98 2.02 3.25 1.06 3.21 2.26 4.28
0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.76 0.00 0.12 0.45 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.10 0.00 0.09 0.32 0.00 0.00 0.00
Normal Normal Normal Normal Normal Normal Normal Normal Normal Normal Normal FA FA FA FCD
tissue tissue tissue tissue tissue tissue tissue tissue tissue tissue tissue
LLD, largest lesion diameter; VOI, volume of interest; R2, best spectral curve fit; tCho, total choline containing compounds; AU, arbitrary unit; FA, fibroadenoma; FCD, fibrocystic disease.
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
Figure 5 ROC curve using test variable absolute tCho (AU) and normalized tCho (nAU) for discriminating malignant from non-malignant tissue.
normalized tCho value of 0.09 0.32 AU/ml. The IDCs in the present study showed high tCho levels. Nevertheless, contrary to the authors’ expectations, it was found that lowgrade IDCs had the highest mean value of tCho followed by grade 2 and grade 3 IDCs. One would expect a higher grade malignant lesion to have a significantly higher level of detectable tCho, which was not the case in the authors’ experience. This observation may be attributed to the small number of low-grade IDCs in the present study. In addition, this study was able to detect tCho in sarcomas. A previous study by Yeung et al.13 did not detect any tCho in phyllodes tumours, which corroborated with another study by Tse et al. 21, indicating that breast stromal tumours generally do not have detectable tCho. Nevertheless, the
e509
presence of choline peaks were identified in two malignant phyllodes tumours in the present study, which has been reported in a recent publication.20 Furthermore, the quantitative MRS study was able to detect the presence of low levels of choline in one malignant phyllodes tumour, i.e., 0.09 AU/ml, which was likely attributed to better spectral resolution at 3 T. MRS at 3 T shows an increase in specificity of lesion detection along with greater spectral resolution of the metabolite peaks. This can improve detectability of tCho and other metabolites, decrease measurement errors, and enable the study of smaller lesions. The synergistic combination of high-field MR and tCho quantifying methods has the potential to greatly improve the clinical utility and availability of breast MRS. The considerations of MRI breast at higher field scanners, includes the need to reduce the number and duration of the radiofrequency pulses per unit time so as to adhere to specific absorption rate (SAR) limitations. This was compensated for by reducing the flip angle to 10 . Due to prolonged T1 relaxation times at 3 Tesla, the study began with a longer TR of approximately 4e5 ms. These adjustments were made based on the recommendations by Kuhl et al. for optimized breast MRI results at 3 Tesla.18 The limitation of this study was the relatively small population of benign lesions. The whole study patient population was not included in the quantitative analysis due to technical difficulties in acquiring the software for tCho quantification in the earlier phase of this study. Quantification of the tCho based on histopathological subtypes was not statistically significant due to the very small sampling of low-grade IDCs. Some lesions that had erratic spectra due to technical problems (i.e., patient breathing/ movement artefacts, susceptibility artefacts due to field inhomogeneity, and inability to perform proper high-order shimming for certain lesions) were also excluded. It would have also been better to perform MRS for normal tissue on the non-lesion containing breast, as the peritumoural environment could influence the tCho measurements. In conclusion, the present study provides an SV 1H MRS protocol for the detection of breast cancer, which is clinically applicable, has a relatively fast scan time, is simple to interpret, and is diagnostically accurate, i.e., improved specificity. Breast MRS at 3 T can improve diagnostic accuracy in differentiating malignant and benign breast lesions when using a tCho cut-off point of 0.33 mmol/l. SV 1H MRS using tCho peak quantification can be used clinically as a non-invasive technique to improve the specificity of breast cancer detection. Nevertheless, MRS is not recommended as a standalone criterion for the diagnosis of breast cancer but should be interpreted in combination with morphological and dynamic contrast-enhancement kinetics features.
Acknowledgements
Figure 6 Boxplot showing the correlation between normalized arbitrary units/normalized tCho values with histopathological findings.
This research study has been supported by the University of Malaya Research Grant (RG032/09HTM). K. Rahmat was supported by University of Malaya Research Grant (RG 390/ 11HTM). The authors gratefully acknowledge Dr Sharon Tan Ling Ling and the UMRIC scientific committee for their essential contri-butions.
e510
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
References 1. Roebuck JR, Cecil KM, Schnall MD, et al. Human breast lesions: characterization with proton MR spectroscopy. Radiology 1998;209: 269e75. 2. Jagannathan NR, Kumar M, Seenu V, et al. Evaluation of total choline from in-vivo volume localised proton MR spectroscopy and its response to neoadjuvant chemotherapy in locally advanced breast cancer. Br J Cancer 2001;84(8):1016e22. 3. Kim JK, Park SH, Lee HM, et al. In vivo 1H MRS evaluation of malignant and benign breast diseases. Breast 2003;12:179e82. 4. Katz-Brull R, Lavin PT, Lenkinski RE. Clinical utility of proton magnetic resonance spectroscopy in characterising breast lesions. J Natl Cancer Inst 2002;94:1197e203. 5. Bolan PJ, Meisamy S, Baker EH. In vivo quantification of choline compounds in the breast with 1H MR spectroscopy. Magn Reson Med 2003;50:1134e43. 6. Sardanelli F, Fausto A, Di Leo G, et al. In vivo proton MR spectroscopy of the breast using the total choline peak integral as a marker of malignancy. AJR Am J Roentgenol 2009;192:1608e17. 7. Jacobs MA, Barker PB, Argani P, et al. Combined dynamic contrast enhanced breast mr and proton spectroscopic imaging: a feasibility study. J Magn Reson Imaging 2005;21:23e8. 8. Sivarajan U, Jayapragasam KJ, Rahmat K, et al. Dynamic contrastenhancement MRI evaluation of breast lesions d a morphological and quantitative analysis. J HK Coll Radiol 2009;12:46e55. 9. Kousi E, Tsougos I, Vasiou E, et al. Magnetic resonance spectroscopy of the breast at 3 T: pre- and post-contrast evaluation for breast lesion characterization. Sci World J 2012;2012:1e9. 10. Kuhl CK. Breast imaging at 3 T. Magn Reson Imaging Clin N Am 2007;15:315e20. vi. 11. Gruber S, Debski BK, Pinker K, et al. Three-dimensional proton MR spectroscopic imaging at 3 T for the differentiation of benign and malignant breast lesions. Radiology 2011;261:752e61.
12. Baltzer PAT, Benndorf M, Dietzel M. False-positive findings at contrastenhanced breast MRI: a BI-RADS descriptor study. AJR Am J Roentgenol 2010;194:1658e63. 13. Yeung DK, Cheung HS, Tse GM. Human breast lesions: characterization with contrast-enhanced in vivo proton MR spectroscopy d initial results. Radiology 2001;220:40. 14. Dorrius MD, Pijnappel RM, Jansen-van der Weide MC, et al. Determination of choline concentrations in breast lesions: quantitative multivoxel proton MR spectroscopy as a promising noninvasive assessment tool to exclude benign lesions. Radiology 2011;259:695e703. 15. American College of Radiology. Breast imaging reporting and data system (BI-RADS). 4th ed. Reston: American College of Radiology; 2003. 16. Fisher U, Kopka L, Grabbe E. Breast carcinoma: effect of pre-operative contrast-enhanced MR imaging on the therapeutic approach. Radiology 1999;213:881e8. 17. Baum F, Fischer U, Vosshenrich R, et al. Classification of hypervascularized lesions in CE MR imaging of the breast. Eur Radiol 2002;12:1087e92. 18. Kuhl CK, Jost P, Morakkabati N, et al. Contrast-enhanced MR imaging of the breast at 3.0 and 1.5 T in the same patients: initial experience. Radiology 2006;239:666e76. 19. Pinker-Domenig K, Bogner W, Gruber S, et al. High resolution MRI of the breast at 3T: which BI-RADSÒ descriptors are most strongly associated with the diagnosis of breast cancer? Eur Radiol 2012;22:322e30. 20. Alhabshi SM, Rahmat K, Abu Hassan H, Westerhout CJ, Chandran PA. Advanced MRI applications and findings of malignant phyllodes tumour: review of two cases. Jpn J Radiology 2013 Feb 6. http:// dx.doi.org/10.1007/s11604-013-0183-y. 21. Tse GMK, Cheung HS, Pang LM, et al. Characterization of lesions of the breast with proton MR spectroscopy: comparison of carcinomas, benign lesions, and phyllodes tumors. AJR Am J Roentgenol 2003;181:1267e72. 22. Mizukoshi W, Kozawa E, Inoue K, et al. 1H MR spectroscopy with external reference solution at 1.5 T for differentiating malignant and benign breast lesions: comparison using qualitative and quantitative approaches. Eur Radiol 2013;23:75e83.
Contents lists available at SciVerse ScienceDirect
Clinical Radiology journal homepage: www.clinicalradiologyonline.net
Improved diagnostic accuracy in differentiating malignant and benign lesions using single-voxel proton MRS of the breast at 3 T MRI S. Suppiah a, b, K. Rahmat a, *, M.N. Mohd-Shah a, C.A. Azlan a, L.K. Tan a, Y.F.A. Aziz a, A. Vijayananthan a, A.L. Wui a, C.H. Yip c a
Department of Biomedical Imaging, University Malaya Research Imaging Centre (UMRIC), Kuala Lumpur, Malaysia Department of Imaging, Faculty of Medicine and Health Sciences, Universiti Putra, Malaysia c Department of Surgery, Faculty of Medicine, University of Malaya, Kuala Lumpur, Malaysia b
article in formation Article history: Received 14 November 2012 Received in revised form 19 March 2013 Accepted 5 April 2013
AIM: To investigate the diagnostic accuracy of single-voxel proton magnetic resonance spectroscopy (SV 1H MRS) by quantifying total choline-containing compounds (tCho) in differentiating malignant from benign lesions, and subsequently, to analyse the relationship of tCho levels in malignant breast lesions with their histopathological subtypes. MATERIALS AND METHODS: A prospective study of SV 1H MRS was performed following dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) in 61 women using a 3 T MR system. All lesions (n ¼ 57) were analysed for characteristics of morphology, contrastenhancement kinetics, and tCho peak heights at SV 1H MRS that were two-times above baseline. Subsequently, the tCho in selected lesions (n ¼ 32) was quantified by calculating the area under the curve, and a tCho concentration equal to or greater than the cut-off value was considered to represent malignancy. The relationship between tCho in invasive ductal carcinomas (IDCs) and their Bloom & Richardson grading of malignancy was assessed. RESULTS: Fifty-two patients (57 lesions; 42 malignant and 15 benign) were analysed. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), of predicting malignancy were 100, 73.3, 91.3, and 100%, respectively, using DCE-MRI and 95.2, 93.3, 97.6, and 87.5%, respectively, using SV 1H MRS. The tCho cut-off for receiver operating characteristic (ROC) curve was 0.33 mmol/l. The relationship between tCho levels in malignant breast lesions with their histopathological subtypes was not statistically significant (p ¼ 0.3). CONCLUSION: Good correlation between tCho peaks and malignancy, enables SV 1H MRS to be used as a clinically applicable, simple, yet non-invasive tool for improved specificity and diagnostic accuracy in detecting breast cancer. Ó 2013 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.
Introduction Single-voxel proton magnetic resonance spectroscopy (SV H MRS) enables non-invasive assessment of the biochemical
1
* Guarantor and correspondent: K. Rahmat, Department of Biomedical Imaging, University Malaya Research Imaging Centre (UMRIC), Faculty of Medicine, University of Malaya, 50603 Kuala Lumpur, Malaysia. Tel.: þ603 79492091; fax: þ603 79494603. E-mail address: [email protected] (K. Rahmat).
composition of breast lesions. Previous studies conducted using 1.5 T MR systems1e6 reported the presence of the resonance of total choline (tCho)-containing compounds at 3.2 parts per million (ppm), which includes contributions from choline, phosphocholine, glycerophosphocholine, and taurine as reliable biomarkers of breast cancer.7 This is because choline-containing metabolites detected in breast lesions are an indicator of the increased cellular metabolism noted in malignant breast tumours.
0009-9260/$ e see front matter Ó 2013 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.crad.2013.04.002
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
Current practices include dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) as part of the standard breast imaging protocol.8 Previously, MRS breast studies at 1.5 T have shown that it is a useful adjunct to standard breast MRI protocols. More recent studies have shown MRS breast at 3 T has the added advantage of providing better spatial resolution for lesion localization.9 It also provides improved spectral resolution of metabolites of the MR spectrum for better detection of tCho peaks.10, 11 Standard breast imaging protocols enable the analysis of the morphological and kinetic patterns of benign and malignant breast lesions detected at MRI. The high sensitivity of breast MRI but its limited specificity poses a diagnostic challenge for radiologists, producing a high false-positive rate of cancer detection leading to unnecessary biopsies being performed.1,8 The ratio of mass to non-mass lesions differed significantly between true-positive (7.0 : 1.0) and false-positive findings (1.2: 1.0), giving a low positive predictive value (PPV) of 51.7% for non-mass lesions detected on standard MRI.12 DCE-MRI sensitivity has been reported to be as high as 94e100%, but with variable specificity in the range of 37e97%.13 By adding MRS to standard breast imaging, the specificity for the detection of breast cancer can be consistently improved from 70% up to 92%.4,6 High field scanning has been proven to be of use in improving diagnostic specificity.10 Although MRS breast at 3 T is gaining momentum as a useful diagnostic tool, there are numerous methods recommended for performing and interpreting it.5e7,14 This leads to uncertainty as to which protocol to be utilized for maximum diagnostic accuracy. Therefore, it poses a problem to radiologists as there is a necessity for selecting a fast, accurate, and clinically applicable protocol. Therefore, the aim of the present study was to investigate the diagnostic accuracy of SV 1H MRS at 3 T by quantifying tCho-containing compounds in differentiating malignant from benign lesions. Subsequently, the relationship between tCho levels in malignant breast lesions were compared with their histopathological subtypes.
Materials and methods Hospital ethical committee approval and written informed consent were obtained. Sixty-one women (mean age 49.7 years old, range 20e83 years old) were prospectively recruited based on findings of breast lesion(s) on clinical examination, mammography, and/or ultrasound. Sixty-five lesions were detected over 22 months and, after exclusion criteria were applied, 57 lesions in 52 patients were obtained for the study cohort (42 malignant: mean size 3.6 cm, size range 1.2e15.8 cm; 15 benign: mean size 1.9 cm, size range 0.7e3.8 cm). Nine patients were excluded due to five having erratic MRS spectra, two did not have conclusive histology results, one patient defaulted biopsy, and one patient did not have a lesion detected at MRI. Single-voxel tCho spectra of all breast lesions and normal breast tissue were obtained and correlated with the final histology results with the reference standard being surgery or core-biopsy results. The lesions were analysed for
e503
morphological and dynamic contrast enhancement kinetics characteristics. Subsequently, selected spectra of malignant and benign lesions, as well as of normal breast tissue, (n ¼ 32) were assessed for their tCho content and whether this corresponded to their histopathological findings. Inclusion criteria were women who had a breast lump detected either by clinical palpation or by radiological investigation, i.e., mammography/ultrasound, and that lesion was going to be either biopsied or excised. Exclusion criteria were women who had a contraindication to MRI, women who had undergone previous surgery/radiotherapy on the lesion-containing breast, and those who only had microcalcifications on mammogram and no solid mass lesions seen at breast ultrasound. Thirty-two spectra were analysed for tCho content after the mathematical software for tCho area under the curve quantification was acquired in the later phase of the study period. Studies were performed on a clinical 3 T MR system (SignaÒ HDx MR Systems; GE Healthcare, Milwaukee, WI, USA) with a dedicated bilateral eight-channel double breast coil using the manufacturer’s proton MRS acquisition software, breast spectroscopic examination (BREASE).
Study protocol/sequences Standard breast MRI protocol The imaging protocol comprised of unenhanced axial T1weighted, T2-weighted, and short tau inversion recovery (STIR) sequences as well as a 3D high-resolution dynamic, contrast-enhanced with T1-weighted fat-suppressed sequences. Axial T2-weighted and STIR images had 4 mm slice thickness, and 40 cm field of view (FOV). Prior to spectroscopy, six-phase contrast-enhanced MR mammography using 0.2 mmol/kg intravenous gadopentate dimeglumine (Magnevist; Schering, Berlin, Germany) was performed [10 flip angle,1.4 mm slice thickness, 40 cm FOV, 256 256 matrix size, 4.2 ms time-to-repeat (TR), and 2.5 ms time-to-echo (TE)]. Six consecutive scans were performed for 1 min each with no intersection gap. The regionof-interest (ROI) was placed on the most enhancing region of the target lesions that were detected. Subsequently, dynamic contrast enhancement kinetic curves were plotted using functional tool (FuncTool) on the manufacturerprovided software [Volume Imaging for BReast AssessmeNT (VIBRANT)] to assess for the contrast medium uptake.
MRS The contrast-enhanced images were scrutinized to identify solid, enhancing lesions suitable for spectroscopic examination. Spectroscopic VOI placement was then performed and adjusted to maximize coverage of the enhancing lesion whilst minimizing coverage of adjacent adipose tissue (minimum VOI size was 10 10 10 mm). Saturation bands were then positioned around the selected VOI to reduce signal coming from the surrounding breast tissue (Fig 1a). An automatic pre-scan was undertaken by selecting a homogeneous VOI in the enhancing lesion. Once full- width halfmaximum (FWHM) of <6 Hz was achieved, image acquisition was performed. A point-resolved spectroscopy (PRESS)
e504
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
Figure 1 MRI of a 28-year-old woman (patient 35) who presented with palpable left breast lump. (a) High-resolution contrast-enhanced image in the axial plane showed a heterogeneously enhancing lesion (red arrow) in the left breast. A VOI was placed in the enhancing solid lesion to optimize coverage of the most enhancing part and to minimize the inclusion of adjacent adipose tissue. Saturation bands were placed around the lesion before MRS data were acquired. /MR spectrum for the lesion showed lipid peak at 1.3 ppm (white arrow), choline at 3.2 ppm (yellow arrow) and water peak at 4.74 ppm (blue arrow). This lesion was revealed to be a grade 3 IDC. (b) Voxel selection and placement for MRS of normal tissue. Saturation bands were placed around normal-appearing breast tissue before MRS data were acquired for purpose of control data. /MR spectrum for normal breast tissue showing absent choline metabolite peak at 3.2 ppm (yellow arrow) and a well-suppressed water peak at 4.74 ppm (blue arrow).
sequence [TR/TE: 2000 ms/155 ms and with number of excitations (NEX) 56 times], single-shot unsuppressed spectrum was performed in the VOI. An automatic water- and fatsuppressed (AWS) sequence, which is a chemical shift selective imaging sequence (CHESS), was included to detect tCho resonance at 3.2 ppm (Fig 1a). The total time taken for all spectroscopic preparation steps and acquisitions was approximately 5e6 min. The control spectra were acquired from VOI placed in healthy-appearing glandular tissue with saturation bands placed around it. The spectra of normal breast tissues were not expected to have a choline peak at 3.2 ppm (Fig 1b).
Data interpretation For the MR spectra that were acquired, chemical shifts were corrected using a water signal at 4.74 ppm. The spectra were shifted accordingly using spectral offset manipulations. The MRI images were interpreted by two radiologists (K.R. and Y.F.) with 8e12 years of subspecialty experience in breast imaging. Objective morphological assessment of the lesions using the American College of Radiology (ACR) BIRADSÒ-MRI (Breast ImagingdReporting and Data System
for MRI) lexicon and final assessment categories from 1 to 6 for each lesion were made. The American College of Radiology’s BI-RADS-MRI lexicon,15 was specifically used as a tool to standardize the image interpretation and appropriate MRI BI-RADS category was documented and scores were given based on morphological features and kinetics characteristics scoring system.16,17 The scores were then correlated with the histopathological reference standard based on core-biopsy histology results but upgraded to surgical histopathology results where applicable. DCE-MRI characteristics were interpreted as explained by Kuhl et al.18 and correlated with BI-RADS MRI descriptors of malignancy.19 Initially, the presence of tCho in a lesion was identified by a peak height at 3.2 ppm that was two-times above baseline noise.7 Subsequently, selected spectra were analysed to calculate the area under the curve of tCho peak at 3.2 ppm using Spectroscopic Analysis for General Electric (SAGE) and MATLAB (Mathworks, Natick, MA, USA) softwares. The extracted peaks in the region of 3.2 ppm, SD 0.2 were fitted with a Voigt line-shape curve-fitting algorithm to calculate the area under the curve. Curve fits giving R2 values of 0.695e1.000 were considered to be good curve
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
fits, and R2 <0.695 were considered poor fits. For lesions with poor curve fit, any measured value of the area under the curve was considered background noise, therefore, these were considered to have no detectable tCho. The tCho levels detected on MRS peaks in these lesions were then correlated with histopathological results. The radiologists were blinded to the patient data and histopathological results at the time of analysis of the tCho peaks.
e505
Table 1 Summary of benign and malignant lesions subgroup pathology. Benign lesions Fibroadenomas Fibrocystic disease Sclerosing adenosis Intraductal papilloma
No. of lesions 6 5 2 2
Statistical analysis Total
Data were entered and tabulated into a computer spreadsheet (Excel; Microsoft, Redmond, WA, USA). Statistical analysis was performed using statistical software SPSS version 17.0 (Chicago, IL, USA). Statistical tests involving the comparison of two categorical variables, e.g., significant tCho peak detection versus malignant histology were performed using Pearson’s chi-squared test. In cases where counts in the cells of the contingency tables were low, Fisher’s exact test was used instead. For comparison of continuous variables, e.g., calculated tCho peak values, independent sample t-tests were employed, except when there were more than two groups, in which case analysis of variance (ANOVA) was used. A p-value of <0.05 was considered to be statistically significant with a confidence level set at 95%. Receiver-operator characteristic (ROC) curves were used for continuous data with a binary outcome to determine the sensitivity of data cut-off points. Kappa values for inter- and intra-observer variability were calculated and showed good agreement. For categorical variables, both intra- and interobserver agreement was good to excellent (Kappa value range ¼ 0.78e0.85, p < 0.05). For scale variables, Spearman’s correlation were used to look for intra- and interobserver agreement, and found good agreement for all variables measured (Spearman’s rho range ¼ 0.81e0.88, p < 0.05). It was also noted that in most categorical and scale variables, the intraobserver agreement yielded a higher agreement value than their interobserver counterpart.
Results Sixty-one women were recruited for the study and postprocessing was performed on 65 detected lesions. There were several lesions/spectra with suboptimal examinations (n ¼ 5) or without definite histopathology results (n ¼ 3), which were discarded from the final analysis. The five suboptimal examinations showed erratic spectrum with reduced signal to noise ratios (SNR) due to patient movement or breathing artefacts, and were discarded from the final analysis. As a result, 52 patients and 57 lesions (42 malignant and 15 benign; Table 1) were used in the final analysis. In the histopathologically proven malignant group, 30 patients underwent mastectomies, eight patients opted for breast conservative surgery, and four patients refused surgery. The MRS spectra for all lesions were initially analysed qualitatively. Forty out of the forty-two histopathologically proven malignant lesions demonstrated a tCho peak; and 14
15
Malignant lesions
No. of lesions
DCIS IDC Grade 1 IDC Grade 2 IDC Grade 3 Lobular carcinoma Malignant phyllodes Breast Osteosarcoma
6 4 15 12 2 2 1 42
out of the 15 benign lesions did not have a significant tCho peak. Fig 2 shows an example of an invasive ductal carcinoma (IDC) with suspicious morphology, demonstrated a washout pattern on kinetic curve and significant tCho on MRS. Two malignant lesions that did not demonstrate a significant choline peak were a 2.1 cm low-grade IDC with prominent ductal carcinoma in situ (DCIS) component (patient 37) and a 1.8 cm low-grade DCIS (patient 39). One benign fibroadenoma gave a false-positive finding. A significant choline peak was also detected in a malignant phyllodes tumour.20 Fig 3 shows the typical benign features and absent choline peak on the spectra of a benign fibrocystic lesion. Significant tCho peak above baseline detected by MRS produced a sensitivity, specificity, PPV, and negative predictive value (NPV) of 95.2, 93.3, 97.6, and 87.5%, respectively (Table 2). Consequently, selected spectra (n ¼ 32) were analysed for good curve fit and the area under the curve was calculated for each spectra. All malignant lesions showed good curve fit, with R2 values between 0.695 and 1.0 (Table 3) compared to benign lesions, which mostly had R2 values below 0.695 (Table 4). ROC curves showed that an R2 value of 0.695 gave a sensitivity of 100% and specificity of 94% (Fig 4). Normalized arbitrary units (nAU) for lesions with R2 values <0.695 were considered to be not applicable. Fig 5 shows the ROC curve for arbitrary units (AU) and nAU correlated with malignancy. Independent sample t-test showed that nAU values significantly correlated with malignancy (p ¼ 0.003). The mean nAU for malignant lesions was higher than the mean for non-malignant lesions (Fig 6). Lesions 20 and 29 had the highest nAU values, from patients 37 and 47, who had grade 1 IDC and grade 2 IDC, respectively. Lesion 22 (from patient 38) was a benign fibroadenoma measuring 3.8 cm, which showed a good curve fit (R2 ¼ 0.92) and relatively higher nAU value compared with the rest of the non-malignant lesions. The optimal threshold for normalized tCho level was 0.33 AU/mldequivalent to 0.33 mmol/l, which gave a good trade-off between sensitivity and specificity for the detection of malignancy. It was also noted that the mean normalized tCho value for grade 1 IDCs was the highest compared with grade 2 and grade 3 IDCs. The mean normalized tCho values for grade 1 (n ¼ 1), grade 2 (n ¼ 5), and grade 3 IDCs (n ¼ 6) were 3.9 4.07 AU/ml; 2.96 2.41 AU/ml; and 1.38 0.90 AU/ml, respectively. However, the ANOVA test did not show that the normalized
e506
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
Figure 2 A 63-year-old woman (patient 47) with grade 2 invasive ductal carcinoma. Left mammogram in the caudo-cranial (a) and medio-lateral oblique (b) views showed an irregular, lobulated, high-density mass in the left upper inner quadrant (white arrow). (c) Colour Doppler ultrasound showed a corresponding hypoechoic solid mass with several penetrating vessels at the 10 o’clock position of the left breast (white arrow). (d) Sagittal contrast-enhanced maximum intensity projection (MIP) images, showed a 5.1 cm irregular mass in the left upper quadrant (red square). (e) Kinetic curve analysis demonstrated rapid uptake with washout pattern (type 3 kinetic curve; true-positive DCE-MRI finding). (f) MRS showed elevated choline peak at 3.2 ppm (yellow arrow) (true-positive MRS finding).
tCho for the three different histopathological grades of IDCs to be statistically significant (p ¼ 0.3). The lesions were also assessed for their morphology and DCE-MRI kinetics characteristics. Lesions that had equivocal DCE findings but suspicious morphology or demonstrated washout kinetics pattern after 3 min post-contrast medium administration were interpreted as malignant.18,19 For total scoring, taking into account morphological features combined with DCE-MRI contrast wash-out, lesions with a score of >5 over 8 criteria were interpreted as malignant16,17; the sensitivity, specificity, PPV, and NPV were 100, 73.3, 91.3, and 100%, respectively (Table 2).
Discussion MRS breast at 3 T has been shown to demonstrate better diagnostic efficacy and has the advantage of better spectral resolution to detect tCho peaks; therefore, there is improved specificity and PPV for better detection of breast carcinoma.9,11 The present study achieved 93.3% specificity by qualitative analysis of tCho peak height and 86.7% specificity for tCho values above 0.33 mmol/l by quantification of the area under the curve of the tCho peak; which is compatible with prior breast MRS study results of between
88.5e100% performed by Sardanelli et al.,6 Bolan et al.,5 Tse et al.21 and Kim et al.3 SVS 1H MRS was chosen as it has a fast acquisition time of less than 6 min compared to multivoxel MRS, which generally takes longer, i.e., approximately 10 min.14 SVS 1H MRS is also easier to interpret as there is only one spectrum to analyse. The ROI is chosen carefully by selecting the most enhancing part of the breast lesion for voxel placement; therefore, giving an accurate representation of tCho level within the lesion. Using an optimized threshold of the absolute tCho peak integral expressed as arbitrary units is a simple and effective method to quantify tCho in breast lesions. A 0.941 sensitivity and 0.867 specificity was achieved with a normalized tCho equal to 0.33 AU/ml, which is comparable to a previous study conducted by Sardanelli et al.,6 which achieved 0.842 sensitivity and 0.885 specificity with a normalized tCho equal to 0.85 AU/ml. A recent MRS study at 1.5 T by Mizukoshi et al.22 in 2013 reported an tCho value of 0.61 mmol/l as an optimal cut-off point, which achieved 68% sensitivity and 79.4% specificity 22. In the present study, the optimal threshold for normalized tCho was 0.33 AU/ml, corresponding to approximately 0.33 mmol/l, and provided a good trade-off between sensitivity and specificity for the detection of malignancy. At 3 T, metabolite peaks at MRS are
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
e507
Figure 3 A 50-year-old woman (patient 27) presented with a left breast lesion that was revealed to be fibrocystic disease. (a) T2-weighted axial image showed the high signal lesion (red arrow) in the left upper outer quadrant. (b) Sagittal post-gadolinium multiplanar reformatted (MPR) image demonstrated the 3 cm lesion (red arrow) in the left upper outer quadrant. (c) Axial image of T1 fat-saturation contrast-enhanced MRI demonstrated a well-defined, fairly homogeneously enhancing oval lesion in the left upper outer quadrant (red arrow). (d) MRS showed no significant elevation of choline peak at 3.2 ppm (yellow arrow; true-negative MRS finding).
better discernible, which enables even low values of tCho peaks to be significant for malignancy. Thus, lesions with normalized tCho levels of 0.33 mmol/l should be biopsied to rule out malignancy.
The VOI was placed after administration of gadolinium, which did not adversely affect the ability to detect choline peaks in malignant lesions as tCho peaks were detected in the majority of malignant lesions. By injecting contrast medium prior to MRS, the voxel prescription was focused on the most metabolically active part of the tumour and adipose tissue was avoided. Kousi et al.9 also recommended performing MRS after gadolinium administration because it improved small lesion localization and enabled better voxel prescription. They performed MRS before and after contrast medium injection, and achieved a sensitivity/specificity of 42.8%/84.6% and 78.5%/92.0%, respectively, for detecting
Table 2 Diagnostic accuracy of using morphology and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) scoring versus magnetic resonance spectroscopy (MRS). Grouping variable
Figure 4 ROC curve using test variable of R2 for discriminating malignant from non-malignant tissue.
Scoring system Scores of >5 versus scores <5 MRS Choline peak detection versus no peak
PPV (%)
NPV (%)
Sensitivity (%)
Specificity (%)
91.3
100.0
100.0
73.3
97.6
87.5
95.2
93.3
PPV, positive predictive value; NPV, negative predictive value.
e508
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
Table 3 Quantitative analysis of malignant lesions. Lesion no.
Patient no.
1 26 2 27 4 28 6 29 8 30 9 31 11 32 13 33 17 35 20 37 23 39 25 40 27 44 28 46 29 47 30 48 31 50 Mean SD Median 25th percentile 75th percentile
Age (years)
LLD (cm)
R2
VOI (ml)
Absolute tCho Peak Integral (AU)
Normalized tCho Peak Integral (AU/ml)
Histopathology findings
54 65 53 55 65 43 65 45 45 39 59 68 31 62 63 65 32 52.5 11.7 54.0 44.0 64.0
5.8 1.8 5.4 5.2 1.8 2.0 2.6 2.5 5.8 2.1 1.8 3.1 7.5 1.4 5.1 3.2 15.8 4.6 3.5 3.2 2.1 5.8
0.98 0.91 0.97 0.98 0.91 0.71 0.98 0.92 0.90 0.75 0.76 0.77 0.89 0.99 0.99 0.81 0.94 0.88 0.11 0.91 0.77 0.98
1.75 1.89 2.59 3.42 1.89 6.40 3.42 3.14 6.00 2.98 2.94 3.69 10.02 1.31 4.36 5.38 7.36 4.11 2.26 3.42 2.77 5.69
5.25 1.62 1.39 2.49 1.62 0.95 3.14 2.86 1.28 10.8 0.94 0.85 11.51 1.30 12.2 1.97 0.14 3.61 3.95 1.97 1.12 4.20
3.28 1.02 0.92 1.66 1.02 0.59 2.09 1.91 0.85 6.78 0.63 0.57 3.05 1.30 6.92 1.23 0.09 2.04 2.00 1.30 0.74 2.57
IDC grade 2 IDC grade 1 IDC grade 3 IDC grade 3 IDC grade 1 IDC grade 2 IDC grade 2 IDC grade 2 IDC grade 3 IDC grade 1 DCIS grade 1 IDC grade 3 IDC grade 3 DCIS grade 3 IDC grade 2 IDC grade 3 Malignant phyllodes
LLD, largest lesion diameter; VOI, volume of interest; R2, best spectral curve fit; tCho, total choline containing compounds; AU, arbitrary unit.
tCho presence. In addition, an intermediate TE value of 155 ms was chosen, which provided a good trade-off between satisfactory SNR and clearly resolvable choline peaks at 3.2 ppm while minimizing possible relaxation losses and motion-related artefacts associated with longer TEs. Normalized tCho levels were detected in the range of 0.09e6.92 AU/ml within 17 malignant and one benign lesions. This finding is comparable to the study performed by Sardanelli et al.,6 which detected tCho levels between
0.38e19.8 AU/ml. It also concurs with studies by Bolan et al.5 and Roebuck et al.,1 which detected levels between 0.4e10 and 0.4e5.8 mmol/l, respectively, using phantoms of known choline concentration to study and further quantify the tCho to millimoles per litre units. Although there was some overlap of values among benign and malignant lesions (Fig 6), malignant lesions had a higher mean normalized tCho value of 2.04 2 AU/ml, compared with non-malignant lesions that had a mean
Table 4 Quantitative analysis of non-malignant lesions. Lesion no.
Patient no.
Age (years)
LLD (cm)
R2
VOI (ml)
Absolute tCho Peak Integral (AU)
Normalized tCho Peak Integral (AU/ml)
Histopathology findings
3 5 7 10 12 14 16 18 19 24 26 15 21 22 32 Mean SD Median 25th percentile 75th percentile
27 28 29 31 32 33 34 35 37 39 40 34 37 38 57
65 53 55 48 65 45 20 45 39 59 68 20 39 20 46 31.3 13.3 29.5 20.0 44.3
NA NA NA NA NA NA NA NA NA NA NA 2.3 1.4 3.8 1.3 2.2 1.2 2.1 1.4 2.7
0.04 0.59 0.03 0.51 0.21 0.48 0.69 0.02 0.23 0.07 0.29 0.69 0.17 0.92 0.55 0.58 0.31 0.62 0.27 0.86
3.49 3.69 3.42 2.98 2.66 3.14 3.69 6.00 2.24 2.94 3.69 4.56 3.45 2.98 2.02 3.25 1.06 3.21 2.26 4.28
0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.76 0.00 0.12 0.45 0.00 0.00 0.00
0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.10 0.00 0.09 0.32 0.00 0.00 0.00
Normal Normal Normal Normal Normal Normal Normal Normal Normal Normal Normal FA FA FA FCD
tissue tissue tissue tissue tissue tissue tissue tissue tissue tissue tissue
LLD, largest lesion diameter; VOI, volume of interest; R2, best spectral curve fit; tCho, total choline containing compounds; AU, arbitrary unit; FA, fibroadenoma; FCD, fibrocystic disease.
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
Figure 5 ROC curve using test variable absolute tCho (AU) and normalized tCho (nAU) for discriminating malignant from non-malignant tissue.
normalized tCho value of 0.09 0.32 AU/ml. The IDCs in the present study showed high tCho levels. Nevertheless, contrary to the authors’ expectations, it was found that lowgrade IDCs had the highest mean value of tCho followed by grade 2 and grade 3 IDCs. One would expect a higher grade malignant lesion to have a significantly higher level of detectable tCho, which was not the case in the authors’ experience. This observation may be attributed to the small number of low-grade IDCs in the present study. In addition, this study was able to detect tCho in sarcomas. A previous study by Yeung et al.13 did not detect any tCho in phyllodes tumours, which corroborated with another study by Tse et al. 21, indicating that breast stromal tumours generally do not have detectable tCho. Nevertheless, the
e509
presence of choline peaks were identified in two malignant phyllodes tumours in the present study, which has been reported in a recent publication.20 Furthermore, the quantitative MRS study was able to detect the presence of low levels of choline in one malignant phyllodes tumour, i.e., 0.09 AU/ml, which was likely attributed to better spectral resolution at 3 T. MRS at 3 T shows an increase in specificity of lesion detection along with greater spectral resolution of the metabolite peaks. This can improve detectability of tCho and other metabolites, decrease measurement errors, and enable the study of smaller lesions. The synergistic combination of high-field MR and tCho quantifying methods has the potential to greatly improve the clinical utility and availability of breast MRS. The considerations of MRI breast at higher field scanners, includes the need to reduce the number and duration of the radiofrequency pulses per unit time so as to adhere to specific absorption rate (SAR) limitations. This was compensated for by reducing the flip angle to 10 . Due to prolonged T1 relaxation times at 3 Tesla, the study began with a longer TR of approximately 4e5 ms. These adjustments were made based on the recommendations by Kuhl et al. for optimized breast MRI results at 3 Tesla.18 The limitation of this study was the relatively small population of benign lesions. The whole study patient population was not included in the quantitative analysis due to technical difficulties in acquiring the software for tCho quantification in the earlier phase of this study. Quantification of the tCho based on histopathological subtypes was not statistically significant due to the very small sampling of low-grade IDCs. Some lesions that had erratic spectra due to technical problems (i.e., patient breathing/ movement artefacts, susceptibility artefacts due to field inhomogeneity, and inability to perform proper high-order shimming for certain lesions) were also excluded. It would have also been better to perform MRS for normal tissue on the non-lesion containing breast, as the peritumoural environment could influence the tCho measurements. In conclusion, the present study provides an SV 1H MRS protocol for the detection of breast cancer, which is clinically applicable, has a relatively fast scan time, is simple to interpret, and is diagnostically accurate, i.e., improved specificity. Breast MRS at 3 T can improve diagnostic accuracy in differentiating malignant and benign breast lesions when using a tCho cut-off point of 0.33 mmol/l. SV 1H MRS using tCho peak quantification can be used clinically as a non-invasive technique to improve the specificity of breast cancer detection. Nevertheless, MRS is not recommended as a standalone criterion for the diagnosis of breast cancer but should be interpreted in combination with morphological and dynamic contrast-enhancement kinetics features.
Acknowledgements
Figure 6 Boxplot showing the correlation between normalized arbitrary units/normalized tCho values with histopathological findings.
This research study has been supported by the University of Malaya Research Grant (RG032/09HTM). K. Rahmat was supported by University of Malaya Research Grant (RG 390/ 11HTM). The authors gratefully acknowledge Dr Sharon Tan Ling Ling and the UMRIC scientific committee for their essential contri-butions.
e510
S. Suppiah et al. / Clinical Radiology 68 (2013) e502ee510
References 1. Roebuck JR, Cecil KM, Schnall MD, et al. Human breast lesions: characterization with proton MR spectroscopy. Radiology 1998;209: 269e75. 2. Jagannathan NR, Kumar M, Seenu V, et al. Evaluation of total choline from in-vivo volume localised proton MR spectroscopy and its response to neoadjuvant chemotherapy in locally advanced breast cancer. Br J Cancer 2001;84(8):1016e22. 3. Kim JK, Park SH, Lee HM, et al. In vivo 1H MRS evaluation of malignant and benign breast diseases. Breast 2003;12:179e82. 4. Katz-Brull R, Lavin PT, Lenkinski RE. Clinical utility of proton magnetic resonance spectroscopy in characterising breast lesions. J Natl Cancer Inst 2002;94:1197e203. 5. Bolan PJ, Meisamy S, Baker EH. In vivo quantification of choline compounds in the breast with 1H MR spectroscopy. Magn Reson Med 2003;50:1134e43. 6. Sardanelli F, Fausto A, Di Leo G, et al. In vivo proton MR spectroscopy of the breast using the total choline peak integral as a marker of malignancy. AJR Am J Roentgenol 2009;192:1608e17. 7. Jacobs MA, Barker PB, Argani P, et al. Combined dynamic contrast enhanced breast mr and proton spectroscopic imaging: a feasibility study. J Magn Reson Imaging 2005;21:23e8. 8. Sivarajan U, Jayapragasam KJ, Rahmat K, et al. Dynamic contrastenhancement MRI evaluation of breast lesions d a morphological and quantitative analysis. J HK Coll Radiol 2009;12:46e55. 9. Kousi E, Tsougos I, Vasiou E, et al. Magnetic resonance spectroscopy of the breast at 3 T: pre- and post-contrast evaluation for breast lesion characterization. Sci World J 2012;2012:1e9. 10. Kuhl CK. Breast imaging at 3 T. Magn Reson Imaging Clin N Am 2007;15:315e20. vi. 11. Gruber S, Debski BK, Pinker K, et al. Three-dimensional proton MR spectroscopic imaging at 3 T for the differentiation of benign and malignant breast lesions. Radiology 2011;261:752e61.
12. Baltzer PAT, Benndorf M, Dietzel M. False-positive findings at contrastenhanced breast MRI: a BI-RADS descriptor study. AJR Am J Roentgenol 2010;194:1658e63. 13. Yeung DK, Cheung HS, Tse GM. Human breast lesions: characterization with contrast-enhanced in vivo proton MR spectroscopy d initial results. Radiology 2001;220:40. 14. Dorrius MD, Pijnappel RM, Jansen-van der Weide MC, et al. Determination of choline concentrations in breast lesions: quantitative multivoxel proton MR spectroscopy as a promising noninvasive assessment tool to exclude benign lesions. Radiology 2011;259:695e703. 15. American College of Radiology. Breast imaging reporting and data system (BI-RADS). 4th ed. Reston: American College of Radiology; 2003. 16. Fisher U, Kopka L, Grabbe E. Breast carcinoma: effect of pre-operative contrast-enhanced MR imaging on the therapeutic approach. Radiology 1999;213:881e8. 17. Baum F, Fischer U, Vosshenrich R, et al. Classification of hypervascularized lesions in CE MR imaging of the breast. Eur Radiol 2002;12:1087e92. 18. Kuhl CK, Jost P, Morakkabati N, et al. Contrast-enhanced MR imaging of the breast at 3.0 and 1.5 T in the same patients: initial experience. Radiology 2006;239:666e76. 19. Pinker-Domenig K, Bogner W, Gruber S, et al. High resolution MRI of the breast at 3T: which BI-RADSÒ descriptors are most strongly associated with the diagnosis of breast cancer? Eur Radiol 2012;22:322e30. 20. Alhabshi SM, Rahmat K, Abu Hassan H, Westerhout CJ, Chandran PA. Advanced MRI applications and findings of malignant phyllodes tumour: review of two cases. Jpn J Radiology 2013 Feb 6. http:// dx.doi.org/10.1007/s11604-013-0183-y. 21. Tse GMK, Cheung HS, Pang LM, et al. Characterization of lesions of the breast with proton MR spectroscopy: comparison of carcinomas, benign lesions, and phyllodes tumors. AJR Am J Roentgenol 2003;181:1267e72. 22. Mizukoshi W, Kozawa E, Inoue K, et al. 1H MR spectroscopy with external reference solution at 1.5 T for differentiating malignant and benign breast lesions: comparison using qualitative and quantitative approaches. Eur Radiol 2013;23:75e83.