Proton MR Spectroscopy in the Breast: Technical innovations and clinical applications

Abstract

Proton magnetic resonance spectroscopy (MRS) is a promising non-invasive diagnostic technique for investigation of breast cancer metabolism. Spectroscopic imaging data may be obtained following contrast-enhanced MR imaging by applying the Point REsolved spectroscopy sequence (PRESS) or the stimulated echo acquisition mode (STEAM) sequence from the MR voxel encompassing the breast lesion. Total choline signal ‘tCho’ measured in vivo using either a qualitative or quantitative approach has been used as a diagnostic test in the workup of malignant breast lesions. In addition to tCho metabolites, other relevant metabolites including multiple lipids can be detected and monitored. MRS has been heavily investigated as an adjunct to morphologic and dynamic magnetic resonance imaging to improve diagnostic accuracy in breast cancer, obviating unnecessary benign biopsies. Besides its use in the staging of breast cancer, other promising applications have been recently investigated, including the assessment of treatment response and therapy monitoring. This review provides guidance on spectroscopic acquisition and quantification methods and highlights current and evolving clinical applications of proton MRS.

Introduction

Breast cancer is the second leading cause of death in women worldwide. Breast cancer is a complex and heterogeneous disease that poses different clinical challenges, especially in term of prognosis and assessment of treatment response. Hence, a deeper understanding of individual tumor properties is mandatory.

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) plays a pivotal role in any given MRI protocol. Compared with conventional imaging, DCE-MRI is the most sensitive method for the detection of breast cancer, with a negative predictive value of 89 to 99% (1–3).

In clinical practice, radiologists combine both morphologic and enhancement kinetics characteristics to reach a diagnosis. This approach is subjective and prone to interobserver variability and experience-based errors (4). Furthermore, overlaps in enhancement characteristics between benign and malignant lesions have been reported (5). Hence, DCE-MRI suffers from a variable specificity that has been reported to range between 47% and 97% (4,6–10).

Multiparametric MRI (mpMRI) is the groundbreaking solution for improving the specificity of DCE-MRI. mpMRI can be performed at different field strengths (1.5–7 T) and involves the combination of established functional parameters (e.g., diffusion-weighted imaging (DWI) and MR spectroscopy (MRS)) with more recently developed parameters (e.g., sodium imaging and positron emission tomography/MRI) to better understand breast cancer development and disease progression as well as to better assess treatment response by quantifying the physiological and pathological processes at the cellular and molecular levels.

Among the established functional MRI parameters, MRS is a non-invasive diagnostic modality that can measure chemical information from a selected region within the tissue of interest (11).The aim of this review article is to provide the readers with an overview of the basis of MRS with special insights into technical innovations and current applications. First, a summary of technical information on MR sequences and quantification tools will be discussed. Then, the use of MRS in breast cancer detection, characterization, and assessment of treatment response beyond choline metabolites will be reviewed. Furthermore, the main limitations of this technique will be given. Finally, future innovations in MRS will be announced.

Basis of Magnetic Resonance Spectroscopy

Since MRI detects and localizes the signals from hydrogen nuclei in water and lipids to produce images, MRS can be used to acquire a chemical spectrum from a specific tissue region and subsequently the spectrum can be converted into chemical information that is useful in the clinical setting. The spectra generated from MRS represent all observable metabolites with their individual chemical profiles in the region of interest; the position and characteristics of each metabolite peak are determined by the underlying chemical formulae, and the area under each metabolite peak represents metabolite concentration. The chemicals measured as well as the techniques typically used to measure them tend to be disease-specific. The presence of a compound resonance around 3.23 ppm has been attributed to different chemical compounds such as phospoethanolamine, choline, phosphocholine, and glycerophosphocholine (the latter three together are simply referred to as total choline (tCho)), and non-choline compounds, which are nearly completely overlapping at 3.23 ppm.

Increased levels of tCho have been detected in malignant tumors and are ascribed to an increased cellular membrane turnover (5,12). As such, tCho measured in vivo using either a qualitative or quantitative approach has been used as a diagnostic test in the workup of malignant breast lesions. In fact, studies have indicated that the metabolic hallmark of cancer is an abnormal choline and phospholipid metabolism associated with oncogenesis and tumor progression (13–17). Upon successful therapy, the levels of tCho may be lowered and hence tCho has also been investigated as a biomarker for monitoring tumor response to treatment (18).

In addition to Cho metabolites, other relevant metabolites including lipids can be detected and monitored with proton MRS (¹H-MRS). Initial ex vivo nuclear magnetic resonance (NMR) and in vivo MRS studies have demonstrated differences in fat and water concentrations between malignant, benign and normal tissue, showing that cancerous tissue displays higher water concentrations and lower methylene lipid peaks at 1.3 ppm (11,19–22). Recently, Thakur et al. used MRS to detect 5–6 lipid peaks in addition to the 1.3 ppm fat peak (23). They investigated the utility of the various lipid concentrations for distinguishing malignant vs. benign lesions, distinguishing molecular subtypes (Luminal A/B vs. others; Luminal A vs. others) and predicting survival outcomes. A higher field strength (3T vs 1.5T) and the STEAM sequence allowed clearer separation of lipid peaks at 2.1 ppm and 2.3 ppm and facilitated the calculation of polyunsaturated and saturated fatty acid fractions.

MRS Acquisition

Currently, MRS acquisition is not standardized; therefore, a variety of spectroscopic techniques at different field strengths have been reported (24–26). Several studies have explored the use of MRS at both 1.5T and 3T. The 3T system yields nearly double the signal-to-noise ratio (SNR) and spacing between metabolite peak locations compared with a 1.5T scanner. However, the 3T system also increases the number of required corrections for T1, T2, B0 field strength, transmit and receiver channel characteristics as part of metabolite quantification analysis (27–29). From the clinical point of view, although MRS at 3T is expected to perform better, no data supporting this expectation has been found. Baltzer et al. (24) examined 19 different studies with spectroscopic data acquired at both 1.5T and 3T in a systematic review and meta-analysis and did not find any significant differences in diagnostic performance between the two systems. Most of the single voxel protocols involving either PRESS or STEAM sequences were used with TR of 1500–2500 ms and TE of 30–450 ms. Therefore, both 1.5 and 3T scanners are suitable for MRS acquisition.

Currently, a variety of breast receiver coils are available that yield differences in SNR, uniformity, comfort, subject orientation, unilateral imaging options, and biopsy devices (30,31). In 2010, Marshall et al. (31) studied three multi-coil breast arrays for conventional and parallel imaging. They found that placing additional small, adjustable coil elements adjacent to the breast tissue led to a larger SNR. Their study supports the use of a multi-coil breast array with high intrinsic SNR to enable parallel imaging. Hancu et al. (32) compared fifteen shimming approaches in 3T breast MRI in human volunteers, finding that a rectangular shim region of interest encompassing the breast region alone resulted in minimal field variability. More recently, Truong et al. and Darnell et al. proposed integrated parallel reception, excitation, and shimming (iPRES) head and body coils for localized B0 shimming in brain and abdominal imaging (33,34); this approach has potential in breast imaging (i.e., phased array coils can be used to achieve homogeneous shimming and fat suppression, thereby improving SNR, fat suppression, and water suppression).

The main pulse sequences used so far for ¹H-MRS of the breast are STimulated Echo Acquisition Mode (STEAM) and Point Resolved Spectroscopy (PRESS). Both sequences can be used to generate localized spectra from a single volume of interest (VOI) or voxel by applying three RF pulses and slice selective gradients. STEAM generates signal from a rectangular or cubic voxel by using three orthogonal slice-selective 90-degree pulses. Similarly, PRESS generates a voxel by acquiring one orthogonal slice-selective 90-degree pulse followed by two 180-degree refocusing pulses. Voxel spectra generated by using PRESS doubles the SNR but the voxel definition is not as precise as STEAM due to the differences between 90- and 180-degree slice-selective pulse profiles. The voxel size and location of the VOI can be easily controlled by adjusting the slice-selective pulses. Both techniques use a water signal suppression and often fat saturation can be applied in the region of interest (24). To suppress the water signal, CHEmically Selective Saturation (CHESS) is usually applied as a pre-scan technique. MRS is highly susceptible to inhomogeneities in the magnetic field and since inadequate water suppression is responsible of the majority of non-diagnostic MRS examinations, the pre-scan phase is of paramount importance (35). Apart from STEAM and PRESS, pulse sequences that have been used for in vivo spectroscopy include semi-LASER which stands for “localization by adiabatic selective refocusing” and LASER which uses adiabatic excitation as well as refocusing pulses for volume localization (36). These adiabatic pulses are insensitive to B1 inhomogeneities and reduces chemical shift artifacts. Finally, fat suppression is an important step when performing MRI or MRS scans. Thus, inversion recovery-based fat suppression methods have included spectral-selective RF pulses (STIR) or spectral-selective adiabatic pulses (SPAIR or SPECIAL) (37).

Two spectroscopy approaches have been used successfully in different types of cancer including breast cancer: single-voxel spectroscopy (SVS) and multi-voxel spectroscopy. SVS generates a cubic or rectangular voxel for a region of interest which samples either the entire lesion or just its center. Many studies have shown that SVS in vivo can distinguish between malignant and benign tissue based on Cho metabolites (25,38–50). In a recent meta-analysis, Cen et al. (51) included 750 malignant and 419 benign breast lesions from eighteen studies and found that SVS had a pooled diagnostic sensitivity and specificity of 71% and 85%, respectively. Furthermore, the addition of SVS has been shown to improve the diagnostic specificity of DCE-MRI (2,52,53). However, SVS is limited in terms of lesion coverage as elevated Cho levels in vital malignant tumors are diluted by contributing necrotic and cystic tumor areas with low Cho levels, yielding false-negative results (39,54). Meanwhile, multi-voxel MRS, also known as chemical shift imaging (CSI), enables the simultaneous acquisition of multiple voxels in single or multiple slices. It allows for the acquisition of a matrix of multiple spectra, enabling in vivo ‘mapping’ of spatial variations of metabolites in both pathological and normal breast tissue (55). Dorrius et al. (56) found that the accuracy of combined multi-voxel MRS and MRI (AUC = 1.00) in assessing breast lesions exceeded that of MRI alone (AUC = 0.96 ± 0.03) when using a Cho concentration cut-off of ≤ 1.5 mM for benign lesions. Gruber et al. (57) evaluated the diagnostic accuracy of quantitative 3D MRS at 3T for differentiating benign from malignant breast lesions, on the basis of Cho signal-to-noise ratio threshold levels, in a clinically feasible measurement time. The authors studied 50 women who underwent MRS for mammographic or sonographic abnormalities. The median Cho SNR was 5.7 in malignant lesions compared with 2.0 in benign lesions. With a Cho SNR threshold level of 2.6, 3D MRS provided a sensitivity of 97% and a specificity of 84% for differentiating benign from malignant breast lesions. They concluded that 3T 3D MRS yields high diagnostic sensitivity and specificity for the discrimination of benign and malignant breast lesions within reasonable measurement times, simplifying acquisition planning. The clinical application of multi-voxel MRS may be limited due to several issues including installation, difficulty in acquiring adequate shimming over such a large volume of tissue, imprecise spatial localization resulting in a great number of partial volume errors, long acquisition time, and the need for post-processing to enable quantitative analysis (58).

Spectra Analysis

Raw spectral data from the MR scanner can be processed off-line on a PC workstation using one of various available software. Data from a GE scanner is commonly processed with GE’s SAGE/IDL software (General Electric Medical Systems, Milwaukee, WI) (11,25). Other vendors will also provide software to analyze the data. Additionally, freely available software such as JMRUI (59) or commercial software such as LCModel (60) can be used to process data from all vendors. The raw data need to be preprocessed before the actual quantitative spectral analysis. Preprocessing consists of three steps: coil element combination, zero filling and apodization, and frequency shift correction and averaging (61,62). The SNR of the Cho peak is calculated as the ratio of the choline peak amplitude to noise amplitude, which is measured in the flat noise baseline region (> 6 or < 0 ppm). Cho or other metabolites can be quantified using the water signal as well as T1 and T2 relaxation constants to calculate absolute concentrations of metabolites (11) or lipid fractions (23).

MRS spectra can be analyzed both qualitatively and quantitatively to differentiate tissue conditions such as normal, benign, malignant, necrotic, or hypoxic. Qualitative assessment relies on the presence or absence of a distinct resonance peak at 3.2 ppm to represent malignancy. This method has demonstrated a sensitivity of 50% to 100% and a reported specificity of 61% to 100% [50].

Several MRS quantitative biomarkers have been reported as promising imaging biomarkers in addition to DCE-MRI. The tCho SNR is the most used quantitative biomarker as of yet and involves determining the SNR of the spectral region at a defined cut-off value. SNR for choline peak detection has also been adopted to determine if MRS results were positive or negative. Results were deemed positive when SNR was greater than or equal to 2 and was deemed negative otherwise. The method has a sensitivity of 44% to 100% and specificity of 67% to 100% (55). Another quantitative biomarker is the tCho peak integral, which has shown a sensitivity of 88% and specificity of 94% (50).

Two main quantitative methods are available for assessing tCho as a quantitative biomarker. The external phantom reference method (EPRM) quantifies tCho molar concentration in an in vivo VOI using an external reference phantom with known concentration of phosphocholine positioned outside the patient’s body. The internal water reference method (IWRM) quantifies tCho concentrations using unsuppressed water reference scans from the same VOI used for water-suppressed choline scans (63). Applying the IWRM technique, Thakur et al. (11) compared tCho and water-to-fat ratios of subtypes of malignant lesions, benign lesions, and normal breast parenchyma to determine their usefulness in breast cancer diagnosis. In this study involving 93 patients with suspicious lesions (> 1 cm), both tCho and water-to-fat (using methylene lipid peaks at 1.3 ppm) were shown to improve diagnostic accuracy in depicting malignancies. Both methods have pros and cons. The IWRM method has been used in many studies and is preferable as it is easier to add an additional spectrum with long TR and short TE for measuring relaxation compensated water signal. Additionally, this method does not need signal corrections due to voxel location, size, and B0 inhomogeneity as the water suppressed choline signal was obtained from the same voxel too. The limitation of this method is that if water concentration is changing due to treatment, this method may introduce errors within quantification estimation. EPRM method may be beneficial for accurate choline quantification in such cases. The limitation is that one needs to have an external phantom with known concentration placed somewhere within breast coil and requires second spectra collection from that phantom to use as a reference. As the cancer lesion location and voxel size are different from the reference location and size, the spectral quantification requires B0 correction.

Minarikova et al. (64) investigated partial volume correction method of Cho signals form multivoxel MRSI by utilizing information from water/fat-Dixon MRI. In this study, glandular/lesion tissue from five breast cancer patients was segmented from water/fat-Dixon and transformed to match the resolution of 3D-MRSI. This approach allowed quantification of the Cho signal in both glandular and tumor tissue independent of water/fat composition in breast 3D-MRSI. The authors conclude that this approach can improve the reproducibility of breast 3D-MRSI, for both diagnosis and response assessment.

Recently, Thakur et al. (23) proposed an LCModel fitting method where individual lipid peak concentrations can be calculated by simulation of experimental model spectra and deconvolution of the experimental spectra into a linear combination of each spectral peak within the MR spectra.

To date, no quantification method has been proven to be superior to another.

Clinical Applications

Differentiation of malignant lesions from benign lesions using tCho levels:

In vivo studies:

Currently, the use of MRS is to differentiate malignant from benign lesions in the diagnostic setting based on increased tCho levels in malignant lesions (Figure 1). Baek (65) and Sah et al. (66) used absolute tCho concentrations and reported a sensitivity of 66% and 92%, respectively, and a specificity of 76% and 75%, respectively. Malignant lesions showed mean tCho concentrations ranging between 2.7 to 5.3 mmol/kg, whereas benign lesions showed mean tCho concentrations ranging between 0.1 to 1.6 mmol/kg. In a large study with 208 breast cancer patients, Mizukoshi et al. (67) calculated a mean choline concentration of 1.1 mmol/kg and 0.4 mmol/kg in malignant and benign lesions, respectively. In a meta-analysis of 19 studies, Baltzer et al. (24) reported that ¹H-MRS used on its own has a pooled diagnostic sensitivity of 73% and specificity of 88%. While there was substantial heterogeneity in sensitivity between the studies (42–100%), there was little variation in specificity. According to this meta-analysis, MRS seems to be limited for diagnosing early breast cancer, small breast tumors, and non-mass enhancing lesions. No significant advantages were reported in diagnostic performance for 3T over 1.5 or for a multi-voxel over single-voxel technique. Of note regarding voxel technique, Tozaki et al. (49) reported an MRS specificity of 85% for breast lesions using a fixed voxel size which is comparable to that of other studies which used a variable voxel size. However, in Tozaki et al.’s study, the sensitivity of 44% when using a fixed voxel is lower than that of other studies which used a variable voxel size.

Figure 1.

(a–b). Biopsy-proved invasive ductal carcinoma in left breast of 34-year-old woman. (a) Sagittal T1-weighted MR image of left breast immediately after intravenous injection of gadolinium diethylenetriaminepentaacetic acid. (b) Spectrum demonstrates a choline (Cho) peak at a frequency of 3.2 ppm, with a signal-to-noise ratio of greater than 2. This is a true-positive finding. (c–d). New mass in right breast of 59-year-old woman. MR imaging–guided biopsy followed by surgical excision yielded benign papillomas, fibrocystic changes and stromal fibrosis. (c) Postcontrast sagittal T1-weighted MR image of the right breast demonstrates an irregular mass. (d) Spectrum did not demonstrate a choline (Cho) resonance peak; only noise level was observed at a frequency of 3.2 ppm. This is a true-negative finding. Lac = lactate, Lip = lipid.

MRS has been investigated as a potential diagnostic tool for improving the diagnostic accuracy of breast MRI in combination with DCE-MRI (24,68). In an mpMRI setting, Mirka et al. (69) reported that increased Cho levels in 100 women with BIRADS 4–6 lesions had a sensitivity of 68.42% and specificity of 93.10%, diffusion restriction had a sensitivity of 90.79% and specificity of 89.66%, and evaluation of DCE-MRI enhancement curve characteristics had a sensitivity of 45.05% and specificity of 72.41%.

Overall, it is important to note that the sensitivity of MRS in the diagnostic setting is based on its ability to detect tCho metabolite concentrations. However, lesion size can be a potential confounder of choline detection sensitivity. Lesions with larger diameters (> 10 mm) are more likely to exhibit detectable tCho peak than smaller lesions with smaller diameters (< 10 mm). Thus, larger lesions are measured with greater sensitivity. Different studies to date have not published concordant results with respect to the ability of MRS for detecting the tCho signal according to lesion size (50,56,57,67,70).

It is also important to note also that, to date, while the results from many single voxel MRS studies are reasonably consistent, the variability in choline concentration levels and optimal cutoff thresholds have resulted in variable diagnostic performance in the literature.

In regards to 2D MRS, Lipnick et al. (22) performed 2D localized correlated spectroscopy (L-COSY) from a prescribed MRS voxel. They concluded that the tCho SNR with the threshold of 2 led to nearly 100% accuracy with a 1-mL voxel size. Jacobs et al. (71) and Baek et al. (70) both used 2D multi-voxel MRS at 1.5 T with 1-mL nominal voxel sizes and found a similar SNR threshold (4 and 3.6, respectively), a similar sensitivity (85% for both) and a similar specificity (81% and 78%, respectively). In regards to 3D MRS, Gruber et al. (57) used 3D MRS at 3T from 32 malignant and 12 benign lesions using 12 × 12 × 12 matrix size scan times of approximately 11 minutes. They used Cho SNR to classify malignancy and found that an SNR threshold of 2.6 led to sensitivity/specificity of 97%/84%.

Lastly, Sardanelli et al. (50) obtained both the absolute tCho peak integral and tCho peak integral normalized for the volume of interest from single-voxel water- and fat-suppressed MRS. No significant change in diagnostic performance was reported when using the normalized tCho peak integral. A surprising result was that the number of false-negative or false-positive findings changed from four using the absolute tCho peak integral (tCho concentration and lesion size) to six using the normalized tCho peak integral, resulting in better diagnostic performance with 90% sensitivity and 92% specificity.

Differentiation of malignant lesions from benign lesions using lipid metabolites:

According to recently published papers, MRS has been also used to differentiate benign and malignant breast lesions (30,72–75) by using the spectral information from multiple spectral regions containing choline, olefinic acids, methylene, and water.

Ex vivo studies:

An ex vivo NMR study showed that malignant carcinosarcoma tissue had higher amounts of water content compared with normal tissue (76). In ex vivo human mammary tissues, the water-to-fat ratio in benign specimens measured with NMR spectroscopy was significantly lower compared with malignant specimens.

In vivo studies:

MRS has been proven to improve the specificity of MRI for the determination of fat necrosis, which may obviate the need for unnecessary biopsies. Hassan et al. (77) reported that the presence of a fat signal in a lesion revealed a sensitivity of 98.04%, specificity of 100%, positive predictive value of 100%, and negative predictive value of 96.88%, whereas non-enhancement assessment of the lesion revealed a sensitivity of 96.08%, specificity of 100%, positive predictive value of 100%, and negative predictive value of 93.94%. However, adding both the non-restriction on diffusion analysis and the lack of tCho at 3.23 ppm increased the sensitivity and specificity to 100%, as well as the positive predictive value of 100% for fat necrosis and hence a negative predictive value for malignancy of 100%.

Several in vivo MRS studies (24,71,78,79) have also reported that the water-to-fat ratio in invasive ductal carcinomas was higher compared with that in benign lesions or normal breast parenchyma. The water-to-fat ratio was also shown to be useful for monitoring the response of breast cancer to neoadjuvant chemotherapy (78,80).

In a feasibility study, Freed et al. (81) analyzed spectroscopic data to investigate the associations between fatty acid fractions in breast adipose tissue and breast cancer status with further stratification with respect to pre- and post-menopausal status. Monounsaturated fatty acid (MUFA) was significantly lower and saturated fatty acid (SFA) was significantly higher in women with invasive ductal carcinoma than in post-menopausal women with benign tissue. They found no correlation between body mass index and fatty acid fractions in breast adipose tissue. Among the women with benign tissue, post-menopausal women had significantly higher polyunsaturated fatty acid and significantly lower SFA than pre-menopausal women (Figure 2).

Figure 2.

Fatty acid composition comparison between diagnostic groups for (a) premenopausal women, (b) for postmenopausal women, and (c) between diagnostic groups with benign tissue. Values displayed are unadjusted values for reader 1. ∗ = Statistically significant difference between adjusted values for both readers. DCIS = ductal carcinoma in situ, IDC = invasive ductal carcinoma. Reprinted from: Freed M, Storey P, Lewin AA, Babb J, Moccaldi M, Moy L, Kim SG. Radiology. 2016 Oct;281(1):43–53. Link to original content: https://pubs.rsna.org/doi/full/10.1148/radiol.2016151959

To evaluate the effectiveness of the water-to-fat ratio and Cho as a priori biomarkers for breast cancer diagnosis, Thakur et al. (11) measured both Cho and the water-to-fat ratio of malignant subtypes, benign lesions, and normal breast parenchyma. For Cho, if every patient with a Cho > 0.1 was classified as malignant, 98% of malignant lesions and 92% of benign lesions were detected correctly. For the water-to-fat ratio, a threshold of 1.0 for the water-to-fat ratio resulted in a sensitivity of 79% and a specificity of 64%. When Cho and the water-to-fat ratio were combined in a model to differentiate malignant from benign lesions, the predictive accuracy of the model was 0.99, which was slightly higher than the area under the curve for Cho alone, although this did not achieve statistical significance (P = 0.16).

Clauser et al. (82) investigated 93 patients undergoing breast MRI for routine clinical indications on a 1.5T scanner to study whether the evaluation of multiple spectral regions can increase the diagnostic performance of ¹H-MRS of the breast and reduce false positive findings. The authors found that a significant AUC for differentiation between benign and malignant lesions was identified for choline (0.733, P = 0.001), olefinic acids (0.769, P = 0.0001) and water-to-methylene ratio (0.704, P = 0.003). The combined evaluation of multiple spectral regions reduced false positive diagnoses in benign lesions of 70.8% (17/24).

Determination of tumor aggressiveness and biological behavior:

MRS not only adds significant value for the detection of breast cancer but might be helpful for assessing the aggressiveness and biological behavior of breast cancer.

In an ex vivo study, Chae et al. investigated whether metabolic profiling of tissue samples using high resolution magic angle spinning (HR-MAS) ¹H NMR spectroscopy could be used to distinguish between ductal carcinoma in situ (DCIS) lesions with or without an invasive component. Metabolite concentrations of choline-containing compounds did not significantly differ between pure DCIS and DCIS without an invasive component. However, the glycerophosphocholine/phosphocholine ratio as well as the concentrations of myo-inositol and succinate was significantly higher in the pure DCIS than in DCIS without an invasive component. Orthogonal projections to latent structure-discriminant analysis models based on the metabolic profiles were able to clearly discriminate between pure DCIS and DCIS without an invasive component (83).

MRS also allows the non-invasive identification and monitoring of triple-negative breast cancer (TNBC) metabolic aberrations. Several human and experimental studies have shown that phosphatidylcholine metabolism is altered in TNBCs and phosphocholine levels are increased in TNBCs. Further understanding of these metabolic mechanisms in triple-negative breast cancers will improve the development and implementation of Cho-based imaging techniques and advance novel therapeutics (84).

The tCho integral value has been reported to be increased in human studies with well-differentiated tumor types with high proliferative activity (57). Tumors with very low proliferative activity, especially lobular cancers, show a scarce choline fraction with the MRS spectra.

Differentiation of malignant lesion subtypes:

Recently, investigators have explored different metabolites for MRS beyond choline to differentiate benign from malignant lesions in ex vivo as well as in vivo studies. During the progression of malignancy, lipid metabolism is activated due to increased activity of lipogenic enzymes. With an increased number of cancer cells, there is an increase in membrane demand, leading to elevated synthesis of membrane phospholipids (85,86). Based on these signal spectra which provide information about the varying levels of associated detectable metabolites, MRS is able to differentiate tissue conditions such as normal, benign, malignant, necrotic, or hypoxic. Moreover, it has been demonstrated that, in breast imaging, the additional application of 1H-MRS aids in the characterization of breast tumors (2).

Ex vivo:

In an ex vivo study, Choi et al. (80) analyzed MR metabolic profiling of ER+ breast cancers using high-resolution magic angle spinning (HR-MAS) MRS and suggested that Cho metabolites may be biomarkers for the aggressive ER+ subtypes. They found that HER2+ cancers had significantly higher levels of both glycine and glutamate compared with HER2− cancers and that luminal B cancers had significantly higher levels of glycine compared with luminal A cancers. Cancers in the high Ki-67 group showed higher levels of glutamate than those in the low Ki-67 group but this did not reach statistical significance. Overall, this study demonstrated that metabolic profiles of core needle biopsy samples assessed by HR-MAS MRS are feasible for detecting potential prognostic biomarkers for ER+ cancers as well as for understanding the differences in metabolic mechanism among the ER+ subtypes. The authors noted that the higher levels of phosphocholine and choline found in the HER2-positive and luminal B subgroups in their ER+ samples were in accordance with results from recent metabolomics studies using core needle biopsy specimens or surgical tissue (87,88). Elevated levels of choline-containing metabolites may be caused by the upregulation of choline kinase activity, which is associated with tumor aggressiveness and drug sensitivity (89–91) (Table 1).

Table 1.

Value of MRS biomarkers in malignant lesion subtypes.

Biomarker	Metabolite	Significance
MRS tCho
Differentiation of molecular subtypes and proliferation rate of breast cancer [Baek HM et al. 2008; Shah T et al., 2010; Choi et al., 2017; Iorio et al., 2016)
HER2+ vs. HER2−	higher glycine and glutamine	P < 0.01, and P < 0.01, respectively
Luminal B vs. Luminal A	higher glycine	P = 0.01
Basal type (TNBC)	phosphocholine through upregulation of choline kinase-alpha
High Ki 67 vs. low Ki 67	higher glutamate	without statistical significance
Differentiation of DCIS from invasive breast carcinoma (Chae et al., 2016)
Pure DCIS vs. DCIS and invasive ductal	higher GPC/PC ratio	P = 0.004
Differentiation of SPC from other invasive breast carcinoma (Chae et al., 2016)	absence of Cho peak
MRS FF (Agarwal et al., 2018)
Malignant vs. benign	lower FF	median 0.12 (range 0.01–0.70) vs. 0.28 (range 0.02–0.71); p < 0.05; 75% sensitivity: 68.6% specificity
ER−/PR− vs. ER+/PR+	higher FF	P < 0.05
HER2neu+ vs. HER2neu−	higher FF	P < 0.05
W/F (Thakur et al., 2011)
Malignant vs. benign	higher W/F	P < 0.05
IDC vs. ILC	higher W/F	P < 0.05

Abbreviations: DCIS, ductal carcinoma in situ; IDC, Invasive ductal carcinoma; ILC, Invasive lobular carcinoma; ER, estrogen receptor; FF, fat fraction (fraction of major fat peak area to the sum of water and fat peaks); W/F, water to major fat peak ratio; GPC, glycerophosphocoline; HER2, human epidermal growth factor receptor 2; MRS, magnetic resonance spectroscopy; PC, phosphocholine; PR, progesterone receptor; SPC, solid papillary carcinoma; tCho, total choline; TNBC, triple-negative breast cancer

In vivo:

In an in vivo study, Agarwal et al. (92) studied lipid metabolism by estimating the fat fraction in different breast tissues and in various breast tumor subtypes using ¹H-MRS. They reported a significantly lower fat fraction in malignant lesions (median 0.12; range 0.01–0.70) compared with benign lesions (median 0.28; range 0.02–0.71) and normal breast tissue (median 0.39; range 0.06–0.76). No significant difference in the fat fraction was seen between benign lesions and normal breast tissue. The fat fraction had sensitivities and specificities of 75% and 68.6% for differentiating malignant from benign lesions and 76% and 74.5% for differentiating malignant from healthy breast tissue. In addition, they found that ER−/PR− tumors had a higher fat fraction compared with tumors, and HER2neu+ tumors had a higher fat fraction compared with HER2neu− breast tumors (Figure 3) (Table 1).

Figure 3.

Representative example of T2-weighted axial MR image of (a) a 45-year-old female suffering from invasive ductal carcinoma; (b) from a woman suffering from fibroadenoma benign lesion, and (c) from a healthy volunteer. The corresponding ¹H MR spectrum obtained from these subjects without water and lipid suppression is shown in (d), (e) & (f), respectively. Reprinted from Magn Reson Imaging vol 49. Agarwal K, Sharma U, Mathur S, Seenu V, Parshad R, Jagannathan NR. Study of lipid metabolism by estimating the fat fraction in different breast tissues and in various breast tumor sub-types by in vivo (1)H MR spectroscopy. Pages 116–122. Copyright (2018), with permission from Elsevier.

Recently, Zhang et al. reported that the absence of the Cho peak on MRS spectra may provide valuable information to distinguish solid papillary carcinomas from other invasive breast carcinomas (93).

Response assessment:

An accepted indication for breast MRI is the evaluation of response to neoadjuvant therapy as recommended by American College of Radiology and the European Society of Breast Imaging (94,95). Neoadjuvant therapy is generally used in locally advanced breast cancer, i.e., stage III disease but can also be used for stage II disease or inflammatory breast cancer (96). Breast MRI has been shown to be effective in monitoring response to chemotherapy compared with mammogram and ultrasound (97,98).

Changes in the tCho concentration and lipid peaks from MRS may be an early indicator of treatment response to precede changes in lesion size (Figure 4). The majority of the studies that have compared MRS with pathologic assessment of treatment response found that early reduction in tCho concentrations are associated with the pathologic response. As part of a multicenter trial (ACRIN 6657), Meissamy et al. (76) showed that changes in the tCho concentration between baseline and 24 hours as measured by single-voxel ¹H-MRS after the first dose of chemotherapy can predict response to treatment. The authors noted that the technical challenges of implementing quantitative spectroscopy in a multi-institutional setting significantly limited the number of usable patient studies. However, this study has not been reproduced and therefore, its clinical value remains controversial.

Figure 4.

Experimental (blue), LCModel fit (red), and the residuals between them from MRS were shown as a function of neoadjuvant tumor response. A) Baseline B) 7-day post after starting the treatment. Reduction in ‘cho’ is visible as shown by black arrow. tCho concentration was decreased. Multiple other lipid peaks were present as shown in addition to choline. Fat fraction (FF) was calculated to be 0.89 at 7-day post treatment which is higher compared to 0.43 at baseline. Enhanced lipid L13+L16 signal is visible on ¹H MRS as a consequence of cell death which occurs through apoptosis or necrosis.

Zhou et al. recently reported that in tumors with a non-concentric shrinkage pattern after neoadjuvant chemotherapy, decreased tCho integral may be useful as a predictive marker. The changes in the tCho integral after chemotherapy were significantly different between response and non-response groups even though the changes in tumor size were not significantly different between the two groups. During follow-up, the maximum area under the receiving operating curve of the change in the tCho integral was 0.747, the sensitivity was 93.75%, and the positive predictive value was 78.9% (79).

There are cases where no tCho can be measured before chemotherapy which limits the clinical utility of the MRS technique. The tCho detection rate in the pre-treatment cohort ranges between 60% and 100% (99–101). Thus, improvements to increase the sensitivity of MRS for measuring tCho would be needed. tCho metabolite detection level could be decreased following several cycles of post chemotherapy in pathological responders (87–90). However, pathological non-responders generally tend to have detectable tCho peak after therapy. Hence tCho may still be a valuable prognostic marker in pre-operative assessment when combined with contrast-enhanced MRI (101–104) (Figure 5).

Figure 5.

An example of breast MR spectroscopy for early monitoring of treatment response. (a) Pretreatment postcontrast gradient echo (repetition time [TR]/echo time [TE] 5 4.4/1.5 milliseconds) 1.5 T MR image of a 51-year-old woman with invasive ductal carcinoma. The red box indicates the voxel placement. (b) Corresponding single-voxel water- and lipid-suppressed spectrum showing residual water and lipids and a tCho peak at 3.2 ppm, acquired with PRESS (Point-Resolved Spectroscopy), 5.1 mL, TR/TE 5 3000/125 milliseconds, 128 averages, CHESS (Chemical Shift Selective) water suppression, and MEGA/BASING (Mescher-Garwood technique/band-selective inversion with gradient dephasing) lipid suppression. Absolute quantification of tCho using T2-corrected water as internal reference gave [tCho] 5 4.1 mmol/kg. The MR imaging/MR spectroscopy study was repeated 4 days later, 1 day after beginning neoadjuvant chemotherapy. (c) Posttreatment scan on day 1 showing voxel placement. (d) Posttreatment spectrum, acquired with same parameters except voxel size 5.4 mL. Although no anatomic changes are evident, MR spectroscopy quantification showed a decrease in [tCho] to 1.5 mmol/kg. The subject demonstrated pathologic complete response at surgery. Reprinted from Magn Reson Imaging Clin N Am vol 21(3). Bolan PJ. Magnetic resonance spectroscopy of the breast: current status. Pages 625–639. Copyright (2013), with permission from Elsevier.

As alterations in cellular metabolism may contribute to the development of a malignant phenotype and cell resistance, MRS applications in tumor response assessment is particularly valuable. In a recent study, Maria et al. investigated ¹H HR-MAS NMR spectroscopy for the characterization of the metabolic profile of metastatic mammary adenocarcinoma breast cancer MDA-MB-231 cells either untreated (control) or treated with tamoxifen, cisplatin, and doxorubicin (105). NMR spectra showed that phosphocholine (a biomarker of breast cancer malignant transformation) signals were stronger in control than in treated cells but significantly decreased upon treatment with tamoxifen/cisplatin.

As in the diagnostic setting, factors such as poor shimming of the magnetic field, patient motion, and even the presence of gadolinium contrast can limit metabolic quantitative measurements for assessing treatment response (78).

MRS is a valuable non-invasive modality in assessing early treatment response, as changes in metabolites (tCho peak) often precedes anatomical changes. MRS performance in predicting response is limited, however, by various factors such as tumor heterogeneity and methods of analysis. More recent semi-quantitative techniques offer a higher level of performance by combining both physiologic and anatomic information; however, they are prone to bias.

Another limiting factor is change in lesion size during treatment. Modification of voxel size to match lesion size has been previously introduced to overcome this limitation.

Given these limitations, MRS has not been yet widely adapted in mpMRI protocols in predicting tumor response. Future multicenter clinical trials are needed to prove its reproducibility and accuracy in response prediction. 2D and 3D multivoxel MR spectroscopic imaging methods need to be optimized to be able to extract information on metabolic tumor heterogeneity. Further, new directions such as radiomics from these metabolite maps can improve tumor characterization, though these need to be validated.

Limitations

In spite of the reported high specificity rates (85–100%) for MRS, its sensitivity remains variable (44%–82%) (49,106,107). Secondly, MRS has a relative long acquisition time, 10–15 minutes, as the detection of metabolites is limited by their concentration in the tissue. Thirdly, technical challenges like poor shimming and chest wall motion in breast MRS contribute to its low-quality spectrum. Fourthly, although MRS can be used to detect multiple metabolites, in breast cancer it is predominately limited to choline compounds (108).

To overcome technical challenges in MRS, training for the technician should be considered. As decisions about the placement of the MRS voxel are usually based on a review of the lesion morphology and the kinetics of contrast agent uptake while the patient is still in the magnet, with SVS, the placement of the voxel is of critical importance. The voxel should be placed so that it contains as much of the lesion as possible while excluding other tissues such as normal fibroglandular or adipose tissue. In studies using MRS to monitor response to treatment, the voxel size and position can be adjusted to cover the same anatomical region of the tumor, decreasing the voxel size as the tumor shrinks.

In addition to the limitations of MRS presented above, lesion size is a particular limiting factor in MRS. The sensitivity of MRS has been reported to decrease with smaller lesions (13) and lesions smaller than 15 mm have been reported with lower sensitivity (49).

Several methods have been reported to improve sensitivity of breast MRS. Stanwell et al. (106,107) showed that setting the spectral resonance at 3.28 ppm resulted in improved accuracy (a sensitivity of 80% and specificity of 100%). Basara et al. (109) reported an improved specificity of 82% and sensitivity of 79% when cases with a peak at 3.28 ppm were excluded and only cases with a peak at 3.23 were considered. However, while the resolution of the composite Cho resonance into its constituent components of 3.23 and 3.28 ppm improved specificity, this approach was associated with increased false negatives of 19%.

Baltzer et al. (110) reported several tCho peaks (3.23 ppm, 3.28 ppm, and 3.3 ppm) for malignant lesions whereas all benign lesions had a single resonance frequency peak at 3.28 ppm. Tozaki et al. (111) reported a tCho peak of 3.22 ppm for malignant lesions and a tCho peak of 3.28 ppm for benign lesions. One patient, however, showed resonance at 3.27 ppm and this was defined as a false negative case. This case was then reported negative for 3.27 ppm after first and fourth cycles of chemotherapy.

Recently, more newer methods have been introduced to decrease false positive results in breast MRS by evaluating multiple spectral regions. Clauser et al. reported that the diagnostic performance of ¹H-MRS could be significantly increased by using a combined analysis of multiple spectral regions. This approach overcame the limited sensitivity of Cho and allowed unnecessary biopsies in benign lesions to be reduced if ¹H-MRS is negative (82). With the discovery of new iPRES coils (36,37) and semi-LASER pulse sequences, it may be possible to perform spectral editing of a metabolite of interest and acquire 2D or 3D MRSI with uniform B0 and B1 homogeneity over the MRSI volume. Additionally, one can apply the Echo Planar Spectral Imaging (EPSI) techniques for speeding up 2D/3D acquisition times and making this tool clinically feasible.

Multiparametric MRI

MpMRI of the breast simultaneously and non-invasively acquires multiple imaging biomarkers and thus has the potential to significantly improve breast cancer diagnosis, staging, and assessment of treatment response. Several recent studies have assessed multiparametric MRI using DCE-MRI and MRS for breast cancer diagnosis. Pinker et al. (52) compared the diagnostic accuracy of DCE-MRI as a single parameter to multiparametric MRI with two (DCE-MRI and DWI) and three (DCE-MRI, DWI and ¹H-MRSI) parameters in breast cancer diagnosis. MpMRI with three parameters yielded significantly higher AUCs (0.936) compared with DCE-MRI alone (0.814) (P < 0.001). MpMRI with just two parameters at 3T did not yield higher AUCs (0.808) than DCE-MRI alone (0.814). MpMRI with three parameters resulted in elimination of false-negative lesions and significantly reduced the false-positives (P = 0.002). The authors concluded that mpMRI with three parameters increases the diagnostic accuracy of breast cancer, compared with DCE-MRI alone and MP MRI with two parameters, and should be considered for future implementation in breast cancer care

Recently, the concept of multiparametric imaging has been extended to ultra-high-field MRI. Schmitz et al. (112) investigated mpMRI with three parameters, i.e., DCE-MRI, DWI, and phosphorus spectroscopy (31P MRSI) at 7T for the characterization of breast cancer. The authors concluded that multiparametric 7T breast MRI with three parameters is feasible in the clinical setting and shows an association between ADC and tumor grade as well as between 31P MRSI and mitotic count.

Conclusion

MRS is a promising non-invasive technique, which provides both tumor metabolic information and insight into physiology of malignant transformation in breast cancer. It has been heavily investigated with its clinical applications in both diagnosis and assessment of treatment response in breast cancer. MRS has been proven to have a role in clinical care to improve the specificity of breast MRI by obviating the need of benign lesion biopsy, although the choice of the optimal technique remains a matter of debate. Further work is needed in improving its technique and sensitivity.

Abbreviations

MRI

Magnetic resonance imaging

DCE-MRI

Dynamic contrast-enhanced magnetic resonance imaging

DWI

Diffusion-weighted imaging

MRS

Magnetic resonance spectroscopy

¹H-MRS

Proton magnetic resonance spectroscopy

NMR

Nuclear magnetic resonance

²³Na MRI

Sodium imaging

³¹P MRS

Phosphorus spectroscopy

mpMRI

Multi-parametric magnetic resonance imaging

DCIS

Ductal carcinoma in situ

IDC

Invasive ductal carcinoma

SNR

Signal-to-noise ratio

tCho

Total choline