An improved 1H magnetic resonance spectroscopic imaging technique for the human breast: preliminary results

Magnetic Resonance Imaging 23 (2005) 571 – 576

An improved 1H magnetic resonance spectroscopic imaging technique for the human breast: preliminary results Jiani HuT, Stephen A. Vartanian, Yang Xuan, Zahid Latif, Renate L. Soulen Department of Radiology, MR Center/Concourse, Harper Hospital, Wayne State University School of Medicine, Detroit, MI 48201, USA Received 21 September 2004; accepted 3 February 2005

Abstract The high sensitivity but poor specificity of magnetic resonance imaging for detecting breast cancer has stimulated interest in magnetic resonance spectroscopic imaging (MRSI) as a tool to improve specificity and reduce the number of benign biopsies. The challenge of applying 1H MRSI to the diagnosis of cancer in the human breast is the need for robust lipid suppression and a clinically acceptable acquisition time. We present an improved 1H MRSI technique that uses an independently optimized chemical-shift-selective for lipid suppression and weighted elliptical k-space sampling combined with a Hamming filter for improved sampling efficiency. D 2005 Elsevier Inc. All rights reserved. Keywords: Breast cancer; Magnetic resonance spectroscopy; Proton

1. Introduction Breast cancer is the one of the most common causes of cancer-related death among women, being second only to lung cancer. It is estimated that one out of every eight women will receive this diagnosis at some point in their lives. An early diagnosis is critical to successful treatment. Dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) of the breast has demonstrated improved sensitivity for detection of breast cancer compared to X-ray mammography, particularly in settings problematic for the latter such as the dense, augmented, irradiated or postsurgical breast [1– 4]. However, specificity remains poor. Biopsies of breast lesions seen only on DCE MRI prove benign in 53–80% of cases [5–7]. Similar difficulties arise in determining the presence of residual tumor in patients treated with neoadjuvant chemotherapy or minimally invasive treatment such as cryotherapy. Contrary to DCE MRI, which characterizes lesions by their morphology and contrast kinetics, magnetic resonance spectroscopy (MRS) reflects lesion metabolism, thus providing completely independent biochemical information. Breast cancers have been distinguished from benign lesions and from normal breast tissue by elevated content of phosphomonoesters and phosphodiesters detected by phosT Corresponding author. Tel.: +1 313 745 1389; fax: +1 313 966 9172. E-mail address: [email protected] (J. Hu). 0730-725X/$ – see front matter D 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.mri.2005.02.011

phorus (31P) MRS and by the presence of a composite choline (Cho) signal on proton (1H) MRS [8–13]. In vivo evaluation of the breast with 31P MRS is precluded by the low sensitivity for detecting 31P at clinical field strengths. 1H MRS offers the highest signal sensitivity among nonradioactive nuclei and can be easily integrated with MRI in the same setting. Analysis of early pooled clinical experience with 1H MRS in 153 lesions yielded sensitivity and specificity for detection of breast cancer of 83% and 85%, respectively, with near 100% for both in a subgroup of young women [8]. A reduction or disappearance of the Cho signal has been associated with response to neoadjuvant chemotherapy of locally advanced breast cancer [11]. The technical challenge presented by a combination of relatively poor achievable B0 homogeneity due to strong lipid signal and the requirement of bextraQ lipid suppression (compared to the brain) has limited progress in the application of 1H magnetic resonance spectroscopic imaging (MRSI) to the breast. With one exception [14,15], most 1H MRS of the human breast up to now has been performed using singlevoxel techniques, due in part to these difficulties. Using an inversion-recovery method to suppress strong lipid signal, and conventional k-space sampling, investigators in the John Hopkins University have been able to acquire 1H MRSI of the human breast [14,15]. Despite this significant progress, an inversion-recovery suppression technique will suppress desired Cho signal in addition to the undesired large lipid

572

J. Hu et al. / Magnetic Resonance Imaging 23 (2005) 571–576

signals. An independently optimized chemical-shift-selective (CHESS) suppression technique can overcome this problem while still suppressing undesired large lipid signals. Another relevant issue in 1H MRSI of the human breast is the acquisition time. MRSI offers better sampling efficiency because multivoxel methods excite and collect a signal from spins in the entire object, thus the sensitivity/unit time of an individual voxel is similar to that of single-voxel technique [16]. However, it takes much longer to acquire an MRSI data set than a single-voxel data set, particularly when using a conventional k-space sampling technique. For example, it will take 25.6 min to acquire a 3232 data set with a TR of 1.5 s and 1 average using a conventional k-space sampling scheme, a duration unacceptable in the clinical setting of a combined MRI/MRS breast examination. In this report, we present an improved 1H MRSI technique for the human breast using an elliptical-weighted k-space sampling scheme combined with a Hamming filter to improve sampling efficiency and two independent CHESS to suppress undesired water and lipid signals.

center frequencies were at 4.75 and 1.5 ppm, respectively. The OVP is determined by applying slice-excitation pulses to select the area outside the volume of interest (VOI) and can be turned on or off as needed [17,18]. Each OVP excitation consists of a 2.56-ms sinc pulse (with a bandwidth of 3400 Hz) and a 2 mT/m gradient pulse in the presaturation slice direction, corresponding to a thickness of 40 mm. To improve sampling efficiency and reduce the relatively long acquisition associated with conventional MRSI, a weighted k-space sampling scheme is used to acquire the MRSI data set [19,20]. The weighted k-space samples only the points located on or within the k-space ellipse. When the number of averages (NA) is greater than 1, the central points of k-space are measured NA times and points on the boundary of the ellipse at least once. For intermediate points, the sampling frequency is determined by their radial distance from the center of k-space. This incorporates the Hamming filter during the measurement, resulting in an improved signal-tonoise ratio (SNR) per measurement time of approximately 20% on a phantom test [19,20]. 2.2. Patients

2. Materials and methods 2.1. Pulse sequence Fig. 1 shows the pulse sequence utilized in this study. It consists of four parts: water suppression, lipid suppression, the outer volume presaturation (OVP) and a PRESS MRSI with a weighted k-space sampling scheme. Water suppression and lipid suppression are achieved by two independent optimized CHESS pulses. CHESS suppresses undesired signals by means of a selective excitation pulse applied at the undesired frequency followed by a spoiling gradient to destroy the signals. The bandwidths for water and lipid suppression CHESS were 60 and 100 Hz, respectively. The

Seven patients with known or suspected breast cancer and DCE MRI-enhancing lesions were studied. Two of these had serial exams before and after cryotherapy for a total of 10 MRSI studies. 2.3. Experimental procedure All patient studies were performed in a Siemens 1.5-T whole-body clinical imager (Sonata, Erlangen, Germany). A standard phased array breast coil (Siemens or MRI Devices, Waukesha, WI) was used both for MRI and 1H MRS. Patients lay prone with their breasts in the coil wells; when possible, gentle breast compression was applied to minimize motion. After global shimming, standard DCE MRI was

Fig. 1. The pulse sequence for 1H MRSI of the human breast. It consists of four components: water suppression (CHESS-W), lipid suppression (CHESS-lip), OVP, and a weighted k-space sampling MRSI sequence.

J. Hu et al. / Magnetic Resonance Imaging 23 (2005) 571–576

573

Fig. 2. Patient with proven breast cancer being evaluated following neoadjuvant chemotherapy for locally advanced disease. (A) The location of the VOI on multiplanar DCE MRI images; (B) the 1H spectrum acquired from voxel 1 shows a distinct choline peak (Cho); (C) a color-coded Cho image, generated from the MRSI, overlaid on the MRI. Note that portions of the enhanced lesion show no Cho, suggesting that this region may no longer contain viable tumor. TR = 1500, TE = 270 ms, FOV = 1616 cm2, thickness = 1.5 cm, spatial resolution = 1.5 cm3.

performed which defined the VOI for MRSI. The MRSI spectra were then acquired. Typical measurement parameters were: field of view (FOV) = 160160 mm2; phasing– encoding steps =1616, TR =1500 ms, TE = 270 ms, slice thickness = 10–15 mm, signal averages =8, and acquisition time = 11 min 40 s. Data were processed with the Siemens spectral process package. Data processing consisted of line broadening (Gaussian function with 256 ms), standard fast Fourier transformations and phasing. No baseline correction was applied. The metabolite images were formed from spectral fitting and zero-filled to 256256 points using the Siemens spectroscopy software package. 3. Results A Cho peak was seen in four of the seven patients. All four had invasive ductal carcinoma varying from 1.6 to 5.0 cm in greatest dimension on DCE MRI. Two of these

patients thereafter had cryotherapy; serial posttreatment studies showed no Cho. No Cho was seen in two patients. Both presented with metastatic adenocarcinoma to the right axilla; DCE MRI, performed in search of an occult primary, showed enhancing foci measuring 0.8 and 1.1 cm in greatest dimension, respectively. One patient’s study was a technical failure due to poor shimming. Illustrative examples are described below. Fig. 2B shows a 1H spectrum from a gadoliniumenhanced area in the right breast of a patient undergoing neoadjuvant chemotherapy for locally advanced cancer. The spectrum is extracted from both water- and lipidsuppressed 1H MRSI of the small box 1 in Fig. 2A. The spectrum, as well as spectra in nearby voxels (boxes 2–7 in Fig. 2A, not shown), demonstrates a distinct Cho peak (SNR N 10), suggesting malignancy. Fig. 2C shows a colorcoded Cho image generated from the MRSI overlaid on the corresponding MRI. Note that the Gd-enhanced areas in

Fig. 3. Patient with palpable, mammographically occult, biopsy-proven breast cancer. (A) An axial image from the DCE MRI showing the enhancing tumor in the right upper outer quadrant; (B) the MRSI VOI overlaid on a coronal MRI image.

574

J. Hu et al. / Magnetic Resonance Imaging 23 (2005) 571–576

Fig. 4. Spectra from the patient shown in Fig. 3. Spectra from all numbered voxels in Fig. 3B that had Cho signals are shown as well as a spectrum from a normal area of the breast (voxel 8) for comparison; the lower right image shows the color-coded Cho image overlaid on the coronal MRI. Note that in this patient, Cho signals extend beyond the enhanced region. Imaging parameters were the same as in Fig. 2.

the superior portion of the region of interest (top row in Fig. 2A) yielded no Cho signals, suggesting absence of viable tumor in these areas. The 1616 2D spectra were acquired in 11 min 40 s with a TE of 270 ms and 8 averages using the weighted k-space sampling scheme. Biopsy prior to cryotherapy confirmed persistent active tumor at the area with Cho signals. Repeat 2D 1H MRSI study acquired 5 days after cryotherapy showed no Cho signal, suggesting successful tumor ablation. This patient demonstrates the potential of detecting tumor distribution by multivoxel techniques.

Fig. 3A is the DCE MRI of a 49-year-old female with a palpable, mammographically occult, right upper outer quadrant 1.61.51.4 cm3 enhancing mass with malignant morphologic and kinetic features which proved to be poorly differentiated high-grade invasive ductal carcinoma. Fig. 4 shows spectra from all numbered voxels in Fig. 3B, which contained Cho signals plus a typical spectrum from a normal area of the breast (voxel 8). Unlike the previous case, in this patient, Cho signals are seen beyond the Gd-enhanced boundary by MRI. This could due to the motion artifact, the psf artifact or more extensive tumor than indicated by

J. Hu et al. / Magnetic Resonance Imaging 23 (2005) 571–576

imaging alone. To prevent motion, we had the breast gently compressed by pads within the coil wells. To reduce psf influence, we have voxel-shifted the MRSI grid during postprocessing to put the lesion within an MRSI voxel. More patient studies are needed to verify whether MRSI can detect breast tumor beyond the Gd-enhanced area. 4. Discussion Increased difficulty in lipid suppression, compared with a single-voxel technique, is a major problem in 1H MRSI of the human breast. The lipid signals in the breast can be so intense that even the a-methylene of lipid (CH2-CHjCH) is often more than 20 times stronger than the desired Cho signal, as illustrated in Figs. 2 and 4. Our results demonstrate that the proposed independent optimized CHESS pulse for lipid suppression can suppress undesired lipid signal sufficiently for in vivo 1H MRSI of the human breast at a TE of 270 ms. The main advantage of CHESS over a nonselective suppression technique, such as inversion-recovery, is that the former does not disturb the signals outside a defined frequency range. Thus, in contrast to the inversion-recovery suppression technique, the lipid suppression technique used here causes no Cho signal loss. Moreover, lipid information can still be obtained through the a-methylene (a-lip) of lipid (CH2-CHjCH) or the olefinic proton (o-lipid) of lipid (CH2-CHjCH). For a TE less than 135 ms, however, the proposed technique does not work well due to increasing difficulty in lipid suppression. In spectra acquired with a TE of 100 ms from the whole VOI, lipid signal was too strong to perform MRSI in both of two tested cases. A short TE can improve SNR or the spatial resolution for the same SNR. Spectra acquired with poor spatial resolution can mix patterns of tumor, necrosis, edema and normal tissue. Moreover, low spatial resolution precludes the applicability of 1H MRSI to small lesions — the very lesions most likely to be indeterminate on DCE MRI and that have the best chance for favorable outcome if malignant. The failure to detect Cho in the two patients with occult primary breast cancers measuring a maximum of 8 and 11 mm, respectively, is most likely due to their small size. Therefore, this is an issue that warrants further effort. Outer volume presaturation, by reducing contamination from outside the region of interest, can be very helpful when dealing with irregular shape of pathological tissues, particularly those near skin or chest wall. The relatively long acquisition time of 1H MRSI compared to single-voxel techniques presents another challenge if the method is to be incorporated into a clinical MR exam of the breast. Many fast 1H MRSI k-space schemes have been developed to overcome the difficulty of relative long acquisition time. Most of these schemes manipulate the spectral dimensions and the three spatial dimensions to shorten measurement time without improving, or even worsening, sensitivity compared to the conventional MRSI experiments [21,22]. The combination of the elliptical

575

phase encoding and Hamming filtering sampling scheme we used can significantly reduce acquisition time compared with conventional k-space sampling while improving SNR [19,20,22]. For a 1616 2D data set with TR of 1.5 s and 8 averages, the acquisition time is only 11 min 40 s. For a 12128 3D data set with TR of 1.5 s and 4 averages, it is only 12 min 30 s. In contrast, the acquisition time for the same 2D and 3D data set with conventional k-space sampling would be 51 min 10 s and 115 min 10 s, respectively. More importantly, inclusion of weighted elliptical sampling and the Hamming filter can significantly reduce the contamination from undesired signals of neighboring voxels [22] and improve SNR per measurement time by approximate 20% [19,20]. For in vivo 1H MRSI of the human breast, the spatial resolution is ultimately determined by the SNR of the Cho signal, therefore preserving or improving sensitivity is more crucial than pursuing high spatial resolution with a fast acquisition alone. The advantages of MRSI techniques over single-voxel proton MRS include simultaneously evaluating multiple lesions, the potential to identify local heterogeneity of breast cancer and the ability to discover bnewQ lesions. As seen in Fig. 4, MRSI may further improve preoperative assessment of lesion size. A 3D technique with a clinically acceptable acquisition time (V 15 min) should be particularly helpful for screening. An important property of MRSI is the ability to arbitrarily shift the voxels in the localization grid after data acquisition. This powerful bvoxel-shiftingQ capability can be very helpful in extracting a bpureQ spectrum from a desired lesion/area during post-acquisition processing. Moreover, MRSI offers better sampling efficiency because multivoxel methods excite and collect a signal from spins in the entire object; thus, the sensitivity/unit time of an individual voxel is similar to that of single-voxel technique [16]. A recent paper by Huang et al. [23] reported two falsepositive results from single-voxel MRS in 50 patients with abnormal mammograms who also underwent DCE MRI and T2*-weighted perfusion imaging. Both patients also had false-positive DCE MRI and both proved to have fibroadenomas. Interestingly, both patients had negative perfusion studies, though perfusion and standard DCE MRI both depend on blood flow. The primary clinical goal of 1H MRSI is to improve the specificity the MR diagnosis of breast cancer (new, residual or recurrent) by providing independent metabolic information. Reported experience to date, mostly with single-voxel techniques, is promising [8–15]. Clinical added value is likely even though 100% accuracy may not be achievable. 5. Conclusion We have presented a scheme for in vivo 1H MRSI of the human breast and demonstrated its potential advantages over single-voxel techniques. This scheme uses an independently optimized CHESS technique for lipid suppression and a weighted elliptical k-space sampling combined with a

576

J. Hu et al. / Magnetic Resonance Imaging 23 (2005) 571–576

Hamming filter for improved sampling efficiency. It is robust, can be implemented on a standard clinical scanner and can be completed in a clinically acceptable time. It has the potential to improve the specificity of the MR diagnosis of breast cancer.

[10]

[11]

Acknowledgment We thank Dr. Kathryn Amerikia for her assistance in recruitment and obtaining clinical data. This work was supported in part by The Susan G. Komen Breast Cancer Foundation. References [1] Kriege M, Brekelmans CT, Boetes C, Besnard PE, Zonderland HM, Obdeijn IM, et al. Efficacy of MRI and mammography for breastcancer screening in women with a familial or genetic predisposition. N Engl J Med 2004;351:497 – 500. [2] Schelfout K, Van Goetham M, Kersschot E, Colpaert C, Schelfout AM, Leyman P, et al. Contrast-enhanced MR imaging of breast lesions and effect on treatment. Eur J Surg Oncol 2004;30:501 – 7. [3] Hata T, Takahashi H, Watanabe K, Takahashi M, Taguchi K, Itoh T, et al. Magnetic resonance imaging for preoperative evaluation of breast cancer: a comparative study with mammography and ultrasonography. J Am Coll Surg 2004;198:190 – 7. [4] Kneeshaw P, Turnbull L, Drew P. Current applications and future direction of MR mammography. Br J Cancer 2003;88:4 – 10. [5] Lee SG, Orel SG, Woo IJ, Cruz-Jove E, Putt ME, Solin LJ, et al. MR imaging screening of the contralateral breast in patients with newly diagnosed breast cancer: preliminary results. Radiology 2003; 226:773 – 8. [6] Liberman L, Morris EA, Kim CM, Kalan JB, Abramson AF, Menell JH, et al. MR imaging findings in the contralateral breast of women with recently diagnosed breast cancer. AJR Am J Roentgenol 2003; 180:333 – 41. [7] Kuhl CK, Morakkabati N, Leutner CC, Schmiedel A, Wardelmann E, Schild HH. MR imaging-guided large-core (14 gauge) needle biopsy of small lesions visible at breast MR imaging alone. Radiology 2001; 220:31 – 9. [8] Katz-Brull R, Lavin P, Lenkinski RE. Clinical utility of proton magnetic resonance spectroscopy in characterizing breast lesions. J Natl Cancer Inst 2002;94:1197 – 203. [9] Cecil KM, Schnall MD, Siegelman ES, Lenkinski RE. The evaluation of human breast lesions with magnetic resonance imaging and proton

[12]

[13]

[14]

[15]

[16]

[17]

[18]

[19]

[20]

[21]

[22]

[23]

magnetic resonance spectroscopy. Breast Cancer Res Treat 2001; 68:45 – 54. Yeung DK, Cheung HS, Tse GM. Human breast lesions: characterization with contrast-enhanced in vivo proton MR spectroscopy-initial results. Radiology 2001;220:40 – 6. Jagannathan NR, Kumar M, Seenu V, Coshic O, Dwivedi SN, Julka PK, et al. Evaluation of total choline from in vivo volume localized proton MR spectroscopy and its response to neoadjuvant chemotherapy in locally advanced breast cancer. Br J Cancer 2001;84:1016 – 22. Kvistad KA, Bakken IJ, Gribbestad IS, Ehrnholm B, Lundgren S, Fjosne HE, et al. Characterization of neoplastic and normal human breast tissues with in vivo 1H MR spectroscopy. J Magn Reson Imaging 1999;10:159 – 64. Roebuck JR, Cecil KM, Schnall MD, Lenkinski RE. Human breast lesions: characterization with proton MR spectroscopy. Radiology 1998;209:269 – 75. Jacobs MA, Barker PB, Bottomley PA, Bhujwalla Z, Bluemke DA. Proton MR spectroscopic imaging of human breast cancer: a preliminary study. J Magn Reson Imaging 2004;19:68 – 75. Jacobs MA, Barker PB, Argani P, Ouwerkerk R, Bhujwalla ZM, Bluemke DA. Combined dynamic contrast enhanced breast MR and proton spectroscopic imaging: a feasibility study. J Magn Reson Imaging 2005;21:23 – 8. Posse S, Schuknecht B, Smith ME, van Zijl PCM, Herschkowitz N, Moonen CW. Short echo time proton MR spectroscopic imaging. J Comput Assist Tomogr 1993;17:1 – 14. Shungu DC, Glickson JD. Sensitivity and localization enhancement in multinuclear in vivo NMR spectroscopy by outer volume presaturation. Magn Reson Med 1993;30:661 – 71. Singh S, Rutt BK, Napel S. Projection presaturation II. Single-shot localization of multiple regions of interest. J Magn Reson 1990;90:313 – 29. Kuhn B, Dreher W, Norris DG, Leibfritz D. Fast proton spectroscopic imaging employing k-space weighting achieved by variable repetition times. Magn Reson Med 1996;35:457 – 64. Maudsley A, Matson GB, Hugg JW, Weiner MW. Reduced phase encoding in spectroscopic imaging. Magn Reson Med 1994;31: 645 – 51. Scheenen TWJ, Klomp DWJ, Roll SA, Futterer JJ, Berentz JO, Heerschap A. Fast acquisition-weighted three-dimensional proton MR spectroscopic imaging of the human prostate. Magn Reson Med 2004; 52:80 – 8. Pohmann R, von Keinlin M, Hasse H. Theoretical evaluation and comparison of fast chemical shift imaging methods. J Magn Reson 1997;129:145 – 60. Huang W, Fisher PR, Dulaimy K, Tudorica LA, O’Hea B, Button TM. Detection of breast malignancy: diagnostic MR protocol for improved specificity. Radiology 2004;232:585 – 91.