Current and emerging quantitative magnetic resonance imaging methods for assessing and predicting the response of breast cancer to neoadjuvant therapy

Abstract

Reliable early assessment of breast cancer response to neoadjuvant therapy (NAT) would provide considerable benefit to patient care and ongoing research efforts, and demand for accurate and noninvasive early-response biomarkers is likely to increase. Response assessment techniques derived from quantitative magnetic resonance imaging (MRI) hold great potential for integration into treatment algorithms and clinical trials. Quantitative MRI techniques already available for assessing breast cancer response to neoadjuvant therapy include lesion size measurement, dynamic contrast-enhanced MRI, diffusion-weighted MRI, and proton magnetic resonance spectroscopy. Emerging yet promising techniques include magnetization transfer MRI, chemical exchange saturation transfer MRI, magnetic resonance elastography, and hyperpolarized MR. Translating and incorporating these techniques into the clinical setting will require close attention to statistical validation methods, standardization and reproducibility of technique, and scanning protocol design.

Introduction

Chemotherapy and hormonal therapy for early stage breast cancer can be administered in either the adjuvant (after surgery) or neoadjuvant (before surgery) settings. Potential advantages to neoadjuvant therapy (NAT) include presurgical reduction of tumor burden, which may allow certain patients to undergo breast conservation therapy rather than mastectomy, and earlier treatment of possible occult micrometastatic disease with the primary breast mass acting as a “marker” for treatment effectiveness (as opposed to adjuvant chemotherapy, where no such marker exists for response of systemic micrometastases).¹ NAT also allows patients time to undergo genetic testing if there is a suspicion of an underlying BRCA 1/2 mutation, which if found may prompt patients to consider mastectomy rather than lumpectomy. At present, NAT is offered primarily to patients with larger tumors, tumors fixed to the chest wall, or clinically matted lymph nodes or skin involvement. However, current clinical trials are evaluating the use of different NAT regimens in multiple patient groups, including those with smaller tumors,² and it is anticipated that these studies will lead to increased use of NAT in early stage breast cancer across a broad spectrum of patients.

With more breast cancer patient receiving NAT, more available NAT regimens, and multiple emerging drug agents and combinations requiring evaluation and comparison with existing options, reliable assessment of treatment response has emerged as an important challenge in both the clinical and research environments. Pathologic response has been established as an independent prognostic marker for overall survival in breast cancer and is currently the gold standard for assessing response to NAT,³ but techniques that can provide reliable response assessment earlier in the course of therapy are in high demand, driven by several considerations. First, from a clinical perspective, accurate early response assessment would provide the opportunity to replace an ineffective treatment with an alternative regimen, and in so doing potentially avoid or curtail debilitating side effects or toxicities, such as cardiotoxicity from anthracycline agents or neuropathy from taxanes. Second, patients with disease that is refractory to multiple NAT regimens could be referred directly to surgery, assuming this determination is made early enough that their disease is still surgically resectable. Third, from a research perspective, accurate early response assessment would allow for determination of treatment efficacy on a much shortened timescale, with important ramifications for clinical trial design.

At present, there is no uniform approach to the early assessment of breast cancer response to NAT. Palpation, probably the most widely used technique in the clinical setting, is inaccurate for predicting pathologic response^4,5 and is poorly suited for assessment of small tumors. Serial biopsy is invasive and is associated with sampling problems in heterogeneous tumors. Noninvasive imaging techniques, including mammography, ultrasound, and conventional MRI, are often deployed in the clinical setting as ad hoc problem-solving tools and in the research setting as secondary endpoints, but no imaging-based, early-response biomarker has been suitably validated as sufficiently predictive of long-term outcomes to become incorporated either as standard of care in the clinical setting or as a routine component of all clinical trials. The stage is now set for development of accurate, noninvasive, early-response biomarkers for integration into both the clinical and research environments.

This contribution discusses quantitative magnetic resonance imaging (MRI) as a promising platform from which to develop and deploy these biomarkers. We begin by discussing the motivations for using quantitative MRI for assessing breast cancer response to NAT. We then describe, with illustrative examples, several currently available quantitative MRI methods including lesion size measurement, dynamic contrast-enhanced MRI (DCE-MRI), diffusion-weighted MRI (DW-MRI), and proton magnetic resonance spectroscopy (MRS). Looking a bit further on the horizon, we then discuss magnetization transfer (MT) MRI, chemical exchange saturation transfer (CEST) MRI, MR elastography, and hyperpolarized MR, methods that have been deployed in other disease sites and that may be particularly well suited to breast cancer response assessment. We conclude by addressing certain practical challenges in the clinical translation of quantitative MRI methods.

Rationale behind quantitative MRI methods for response assessment

Cancer imaging is undergoing a paradigm shift in which quantitative answers are increasingly being sought for questions that have historically motivated a qualitative response. The query “is this cancer responding to therapy” has traditionally been answered by clinical imaging based on the subjective impressions of the observer. However, modern cancer research and treatment now requires objective and reproducible response assessment variables with which to evaluate and compare different treatment strategies. For evaluating treatment response, qualitative imaging interpretation is yielding ground to quantitative imaging response parameters that can be integrated with other quantitative clinical datasets for rigorous statistical evaluations.^6,7

The current mainstay of objective imaging-based response assessment for solid malignancies is the Response Evaluation Criteria in Solid Tumors (RECIST), a set of published guidelines for image acquisition, lesion evaluation and measurement, and response categorization.⁸ RECIST has been successful in providing a standardized approach for imaging-based response assessment and facilitating “apples to apples” comparisons of different cancer treatments, but its emphasis on changes in lesion size has been criticized as failing to capture meaningful changes in tumor biology.^9,10 In particular, RECIST may underestimate the antitumor efficacy of newer drug agents with cytostatic rather than cytotoxic effect, where tumor shrinkage may be minimal or quite delayed. The future of imaging-based quantitative response assessment will likely involve supplementing traditional size-based techniques with more advanced morphological, functional, physiological, cellular, and molecular imaging methods.

For a variety of reasons, MRI is a very promising platform from which to develop advanced quantitative imaging biomarkers for assessing breast cancer response to NAT. First, MRI is already well established as a clinically useful technique in breast cancer detection and characterization,¹¹ and its use for response assessment would be a natural extension of its current role in clinical care. Second, as an intrinsically digital technique, MRI is capable of generating quantitative datasets for direct entry into statistical analyses, thus bypassing the need for analog-to-digital conversion or subjective user interpretation that can add to measurement variability. Third, MRI is an extraordinarily flexible and powerful modality with the ability to report on multiple structural and functional parameters that may be relevant to lesion response assessment, as detailed in this review.

It is important to note that response assessment is only one of two goals for imaging during NAT, with the other being assessment of residual disease as an adjunct to surgical planning.^11–14 While it is tempting to think of these two objectives as one and the same, assessment for residual disease fundamentally seeks to depict anatomy, whereas response assessment seeks to evaluate changes in tumor biology and may do so with a variety of techniques reporting on either anatomical or functional changes. The distinction is especially important for MRI, where certain tradeoffs in how images are acquired (eg, between high temporal and high spatial resolution techniques) may theoretically render one imaging protocol more useful for response assessment and another protocol more useful for demonstrating tiny foci of residual tumor. This review focuses on MRI techniques for response assessment, but we address practical issues of protocol design in a later section on clinical translation.

Currently available quantitative MRI techniques for response assessment

A number of MRI techniques are already available for potential deployment into both clinical and research settings for the assessment of breast cancer response to NAT. These methods include anatomical measurement of lesion size (unidimensional, multidimensional, and volumetric), DCE-MRI, DW-MRI, and MRS. In this section we review the basic theory underlying these techniques (including methods of quantitative analysis), provide examples of relevant clinical and research applications, and discuss opportunities for future development.

Lesion size measurement

Theory

By virtue of its exquisite soft tissue contrast and very high spatial resolution, MRI is an extremely powerful technique for demonstrating the morphology of breast lesions. MRI has therefore been evaluated for assessing breast cancer response to therapy using changes in lesion size as the primary measurement variable. MRI-based evaluation of lesion size relies almost exclusively on “contrast-enhanced” imaging ie, imaging following intravenous injection of a paramagnetic contrast agent (typically a gadolinium chelate) so as to maximize the conspicuity of an enhancing breast lesion against a background of normal breast parenchyma.¹⁵ The literature has incorporated unidimensional, bidimensional, and volumetric (ie, three-dimensional) tumor measurements (Figure 1).

Figure 1.

Lesion size measurement by MRI. (A) Unidimensional measurement of tumor long axis diameter. (B) Bidimensional measurement of tumor long and short axis diameters. (C) Three-dimensional measurement of tumor volume.

Abbreviation: MRI, magnetic resonance imaging.

Applications

Evaluations of MRI-based lesion size measurement in the setting of NAT have focused on two clinical scenarios: (1) evaluation for residual disease at the end of NAT and (2) early assessment of treatment response. With regard to evaluation for residual disease, a significant literature has demonstrated the effectiveness of MRI-based lesion size measurement in predicting results at surgical pathology.^5,16–22 Abraham et al¹⁹ showed that contrast-enhanced MRI at the end of NAT predicted pathologically determined residual disease in 97 percent of cases, outperforming both physical examination and mammography.¹⁹ Balu-Maestro et al²³ compared physical examination, mammography, ultrasound, and MRI and found that MRI was the most reliable for evaluating residual tumor size after NAT. Londero et al²² found that MRI after NAT identified residual disease better than mammography and also showed that MRI-measured tumor diameters correlated better with pathologic diameters than either mammography or sonography. Chou et al¹⁶ found that tumor volume after NAT (calculated using a segmentation algorithm) was highly correlated with histopathologic estimation of tissue volume.¹⁶

With regard to early assessment of treatment response, a number of studies have evaluated the ability of early changes in MRI-based lesion size measurement to predict eventual response.^24–26 Martincich et al²⁵ found that a minimum 65 percent reduction in tumor volume after two cycles of NAT was associated with an eventual major histopathological response. Padhani et al²⁷ demonstrated that a decrease in bidimensional tumor area after two cycles of NAT predicted pathological response. Loo et al²⁸ showed that a change in the longest diameter of enhancing tumor predicted pathologic response after two cycles of NAT.²⁸ The recently published ISPY-1 trial showed that volumetric tumor measurement at mid-treatment outperformed clinical assessment in predicting pathologic response.²⁹

Future directions

Although MRI techniques for lesion size measurement are mature and have been shown to predict changes in true lesion size, MRI-based size measurement has not been widely incorporated into the clinical or research settings for NAT response assessment. In the clinical setting, this is probably due to the prevailing opinion that physical examination provides adequate response assessment in the patient population for whom NAT is currently indicated, ie, patients with large tumors. In the research setting, RECIST-based response biomarkers may be incorporated as secondary endpoints, but current trials invariably rely on pathologic response as the primary endpoint for determining antitumor efficacy. The latter remains true despite some preliminary studies correlating MRI-based size measurement changes with long-term outcomes.³⁰

In the future, with the anticipated increased use of NAT in patients with smaller tumors that are poorly assessed by palpation, MRI-based lesion size measurement may become more integrated into clinical algorithms, but there remain several potential problems with lesion size-based response assessment. First, many breast cancers tend to have infiltrative, irregular, and/or multifocal growth, creating problems for reproducible size measurement. Second, as mentioned previously, some new and emerging treatment agents have cytostatic rather than cytotoxic effect and may produce delayed or attenuated lesion shrinkage such that a size-based response assessment methodology may not fully capture relevant changes in tumor biology. Third, when treated, breast cancers can exhibit a variety of different morphological changes including shrinkage from the outside, melting from the inside, and irregular internal (“Swiss cheese”) liquefaction; these different patterns would be poorly assessed by methodologies focusing exclusively on unidimensional, bidimensional, or even volumetric measurement.

The biggest long-term challenge for lesion size measurement-based techniques will therefore probably involve integration with other techniques focused more on assessing functional changes in tumor biology, as described in subsequent sections. Integration of different techniques into a comprehensive, multi-purpose MRI examination may be challenging, and will also be described later.

Dynamic contrast-enhanced MRI (DCE-MRI)

Theory

DCE-MRI involves the rapid, sequential acquisition of T₁-weighted images before and after the injection of a paramagnetic contrast agent.³¹ As the contrast agent perfuses or diffuses into a voxel or region of interest (ROI), it shortens the native magnetic relaxation times of the tissue as determined by the local concentration of contrast. When the contrast agent leaves the voxel, the relaxation times increase toward their baseline value at a rate determined by local tissue characteristics. Each voxel thus yields a signal intensity time course that can be analyzed to yield estimates of tissue vascularity parameters including perfusion, permeability, and tissue volume fractions.

Both semiquantitative and fully quantitative methods have been developed to perform DCE-MRI analyses. Semiquantitative methods include calculations of the signal enhancement ratio (SER) and the initial area under the enhancement curve (iAUC).^32–36 The SER method, used in the ISPY-1 trial, employs the following calculation:

$$SER=\frac{S_1-S_0}{S_2-S_0}$$ (1)

where S₀ represents the signal intensity within the lesion before contrast administration, S₁ represents the signal intensity early after contrast injection, and S₂ represents the late postcontrast signal intensity³² Given a map of SER values, various derivative parameters are accessible including the SER total tumor volume, SER partial tumor volume, SER washout tumor volume, and the peak SER.

Fully quantitative DCE-MRI analysis typically involves the application of various pharmacokinetic equations to model the movement of contrast agent molecules between tissue and blood vessels over time.³⁷ Two main approaches have been deployed in breast cancer: the standard Tofts-Kety (TK) and the extended Tofts-Kety (ETK) models.³⁸ The standard TK model is summarized by:

$$C_t(T)=K^{trans}\cdot {\displaystyle \underset{0}{\overset{T}{\int }}{C}_p(t)\cdot e^{-\left(K^{trans}/v_e\right)\cdot \left(\text{T}-t\right)}dt,}$$ (2)

where K^trans is the volume transfer constant, v_e is the extravascular extracellular volume fraction, C_t(T) is the concentration of contrast in the tissue of interest, and C_p(t) is the concentration of contrast in blood plasma (also known as the arterial input function). The ETK model incorporates the blood plasma volume fraction, v_p, as follows:

$$C_t\left(T\right)=K^{trans}{{{\displaystyle \int }}^{\text{}}}_0^T{C}_p\left(t\right)\text{exp}\left(-\left(K^{trans}/v_e\right)\left(T-t\right)\right)dt+v_p{C}_p\left(t\right).$$ (3)

Both the TK and ETK models assume that all water compartments within tissue are well mixed (ie, are at the fast exchange limit of the nuclear magnetic resonance time scale) so that MRI signal change is completely described by a single relaxation rate constant.³⁸ After measurement (or estimation) of C_t(t) and C_p(t) on a voxel or ROI level, Equations 2 and/or 3 are used to return estimates of K^trans, v_e, and v_p, along with the derived efflux rate constant k_ep (k_ep≡ K^trans/v_e). The working hypothesis for fully quantitative DCE-MRI in the setting of NAT is that observed changes in these pharmacokinetic parameters can predict treatment response, as illustrated in Figure 2. For a more extensive discussion of DCE-MRI methods in oncology, the interested reader is referred to the review by Yankeelov and Gore.³¹

Figure 2.

Fully quantitative DCE-MRI analysis in a breast cancer patient undergoing NAT. (A) Pretreatment DCE-MRI analysis yields a baseline calculated mean tumor K^trans value of 0.3 min⁻¹. (B) DCE-MRI analysis after one cycle of NAT yields a calculated mean tumor K^trans value of 0.2 min⁻¹. (C) Imaging after completion of NAT shows that the lesion is no longer visible; at surgery, the patient had a pathologic complete response.

Note: Ongoing studies are investigating whether early changes in mean tumor K^trans can reliably differentiate pathologic responders from nonresponders.

Abbreviations: DCE-MRI, dynamic contrast-enhanced magnetic resonance imaging; NAT, neoadjuvant therapy; K^trans, volume transfer constant.

Applications

Both semiquantitative and fully quantitative DCE-MRI techniques have been evaluated for assessing treatment response in breast cancer patients undergoing NAT. The best known example of semiquantitative DCE-MRI in this setting is the I-SPY 1 trial,²⁹ in which patients underwent contrast-enhanced MRI at multiple time points before, during, and after NAT; the authors found that midtreatment change in SER predicted pathologic response with an area under the curve (AUC) of 0.71, higher than clinical assessment (0.63) but slightly lower than change in tumor volume measurement (0.72).²⁹ Other investigations of semiquantitative DCE-MRI include employment of iAUC for response assessment. For example, Tateishi et al³⁶ used the percentage area under the time-intensity curve (%AUC) to predict treatment response and reported that although the sensitivity of %AUC (50%) was not acceptable, the specificity of %AUC of 95.2% was sufficiently high to predict pathologic complete response.³⁶ Our group has shown that semiquantitative analysis of high temporal resolution DCE-MRI data can predict pathologic response after a single cycle of NAT (unpublished data).

Quantitative analysis of DCE-MRI data has been shown to assess³⁹ and predict treatment response^27,40 and has also been shown to correlate with 5-year survival.^41–44 Padhani et al²⁷ found that after two cycles of NAT, both change in tumor size and change in K^trans range on histogram analysis were equally able to predict pathologic response. Ah-See et al⁴⁰ analyzed multiple quantitative DCE-MRI parameters and reported that change in K^trans was the best predictor of pathologic nonresponse.⁴⁰ Our group found that not only the mean but also the standard deviation of K^trans as estimated by the TK and ETK models, as well as v_p, can separate complete pathologic responders from nonresponders after a single cycle of NAT⁴⁵

Diffusion-weighted MRI (DW-MRI)

Theory

Like conventional MRI, DW-MRI records signals from mobile water molecules within tissues, but in DW-MRI the contrast reflects the distance water molecules can migrate or “diffuse” from their original spatial position over a short time interval due to random, thermally-induced motion (ie, Brownian motion). DW-MRI exploits applied gradients of the main magnetic field that allow for localization and calculation of the microscopic diffusion of water molecules. By acquiring two or more images with different degrees of “diffusion weighting” (obtained by applying the diffusion-sensitizing gradients with different amplitudes on successive image acquisitions), an estimate of the amount of molecular water diffusion, termed the apparent diffusion coefficient (ADC), can be calculated at each voxel from

$$S=S_0\text{exp}\left(-b\cdot ADC\right),$$ (4)

where S is the signal intensity recorded with application of a diffusion-sensitizing gradient, S₀ is the signal intensity with no diffusion-sensitizing gradient, and b is a composite variable reflecting various acquisition parameters (including the strength of the gradient pulse, duration of the pulse, and interval between pulses).⁴⁶ ADC values from successive voxels can then be aggregated to produce a map of ADC values over the volume of interest (Figure 3).

Figure 3.

DW-MRI in a breast cancer patient undergoing NAT. (A) On a pretreatment image with no diffusion gradient (ie, b = 0 s/mm²), the tumor is difficult to distinguish from background normal parenchyma. (B) Pretreatment diffusion-weighted image (b = 660 s/mm²) demonstrates subtle patchy increased signal in the deep upper breast, corresponding to an infiltrative tumor. (C) Pretreatment quantitative ADC map, with color-coded voxels corresponding to tissue ADC. The tumor region is outlined in white. (D) ADC map derived from DW-MRI after one cycle of NAT; the tumor volume (again outlined in white) has markedly decreased.

Note: This patient went on to have a complete pathologic response.

Abbreviations: DW-MRI, diffusion-weighted magnetic resonance imaging; NAT, neoadjuvant therapy; ADC, apparent diffusion coefficient.

In general, experimentally-measured ADC values are lower in organized tissues than in free solution because various structures including cell membranes and intracellular organelles tend to restrict or hinder the free movement of water.^47,48 Moreover, cancerous tissues often show significantly reduced ADC values when compared with healthy tissues, a finding typically attributed to the increased cell density of many malignancies.⁴⁹ With treatment, intratumoral ADC values often rise, presumably because of decreases in cell density consequent to apoptosis and cell death, with concomitant disruption of cell membranes allowing water molecules to diffuse more freely. This basic paradigm – low tumor ADC values before treatment, followed by rising tumor ADC values with treatment – provides the basic model for DW-MRI as a response assessment technique. Importantly, it has been shown that rising ADC values can occur quite early during treatment,^50–52 thus providing the motivation for studying DW-MRI as an early response biomarker. For a more extensive discussion, the reader is referred to the review by Arlinghaus and Yankeelov.⁵³

Applications

Several studies have correlated changes in ADC on DW-MRI with treatment response in breast cancer patients undergoing NAT. Early investigations demonstrated increases in mean tumor ADC following chemotherapy³⁹ and correlated mean tumor ADC increases with radiological response,^54,55 while subsequent studies examined the relationship between mean tumor ADC changes and pathological response. In general, mean tumor ADC has been found to increase after NAT in both pathologic responders and nonresponders but to increase more for pathologic responders.^56–59 Sharma et al⁵² showed that at the end of therapy, DW-MRI had a lower sensitivity but a higher specificity than lesion size measurement for differentiating pathologic responders from nonresponders.

More recent studies have looked into changes in ADC early in the treatment course and have begun to investigate whether DW-MRI performed early in treatment can be used to predict response. Pickles et al⁵¹ demonstrated a significant increase in mean tumor ADC after a single cycle of NAT and showed that mean tumor ADC began to change earlier than tumor longest diameter. In a study of 15 patients, mean tumor ADC increased significantly (P = 0.008) by 11% after a single cycle of NAT.⁶⁰ Li et al⁵⁰ reported that after one cycle of NAT, mean tumor ADC significantly increased by 24% in patients who went on to have a complete or partial response and did not change in patients who ultimately exhibited stable or progressive disease.

Proton magnetic resonance spectroscopy (MRS)

Theory

In contrast to conventional MRI, which generates anatomical images using signal primarily arising from water, MRS provides information on the concentrations of different metabolites in tissue. This technique can be exploited to detect the altered metabolic signatures of cancer cells. For example, many malignancies demonstrate elevated levels of choline and lactate, the former due to high rates of membrane turnover and the latter due to utilization of anaerobic glycolysis. Choline is present in less than one millimolar concentrations in normal breast tissue but is significantly elevated in malignant breast tumors^61,62 due to choline kinase overexpression driven by HIF-1α^63,64

MRS is more challenging in the breast than in other organs due to the large amount of signal from lipid and to increased magnetic susceptibility at air–tissue interfaces, although the latter is less relevant in single-voxel MRS of the breast. Despite the lower spectroscopic resolution and signal-to-noise ratio of breast MRS compared with MRS in other organs, measurements of choline levels and water-fat (W-F) ratios are feasible in the breast and have been used for breast cancer diagnosis as well as for monitoring response to treatment during NAT.^65–72

Applications

A recent clinical trial reported by Kumar et al⁶⁷ demonstrated that malignant breast tissues have elevated W-F levels compared with controls, and that breast cancers decreasing in size with NAT also exhibited decreasing W-F ratios.⁶⁷ In a separate clinical trial reported by Tozaki et al⁷¹ using choline MRS in patients undergoing NAT, the reduction rates of choline were statistically significantly different between pathological responders and nonresponders after two treatment cycles, with positive and negative predictive values of choline MRS of 89% and 100%, respectively. Moreover, it was found that the predictive power of choline MRS was greater than that of volumetric tumor measurements. Danishad et al⁷³ found that choline signal-to-noise ratio may be useful in predicting tumor response to NAT.

Future directions

Although relatively high predictive power has been demonstrated in preliminary trials, the widespread clinical implementation of W-F and choline MRS for breast cancer treatment response assessment will confront several challenges. First, customized data postprocessing and internal/external signal referencing is required to convert raw spectroscopic data into quantitative information, especially for clinical analysis of multi-voxel MRS data. Second, because the majority of tissue metabolites in vivo besides lipids are present at millimolar concentrations, the MRS sampling voxel must typically be very large (1–8 cm³) in order to achieve sufficient signal-to-noise ratio for very dilute metabolites, and this low spatial resolution translates into limited ability to interrogate small tumors and to report on intralesional heterogeneity. Improvements on the low spatial resolution of conventional MRS may result from emerging methods including multivoxel chemical shift imaging^61,70,74 and hyperpolarized MR, discussed below.

Emerging quantitative MRI techniques for response assessment

Several additional MRI techniques are on the horizon for possible future use in assessing breast cancer response to NAT. Some of these techniques have already been deployed in other disease sites. This section discusses possible applications of MT, CEST, and MR elastography as treatment response biomarkers. For each technique, we briefly discuss the underlying theory as well as opportunities and preliminary applications in breast cancer.

Magnetization transfer (MT) MRI

Theory

MT is a method for detecting and quantifying the protons associated with tissue macromolecules. These macromolecular protons cannot be routinely detected with conventional MRI; MT takes advantage of the communication between macromolecular protons and water to elicit an indirect measurement of their properties. MT is typically performed by applying a preparatory off-resonance radiofrequency pulse to selectively saturate protons associated with macromolecules and then measuring the attenuation of the water signal that occurs as a result of the exchange of spin information (magnetization transfer) between the saturated macromolecular protons and free water. The magnetization transfer itself takes place via dipolar coupling and/or chemical exchange.⁷⁵

The typical method for characterizing the MT effect is to compare the signal intensity between a nonsaturated image (or so-called reference acquisition, designated as MT_off) and an image acquired after application of the off-resonance radiofrequency (RF) pulse (MT_on). The magnitude of the saturation is proportional to the quantity of saturated macromolecular protons and the efficiency of exchange (or exchange rate) with free water protons (Figure 4). The MT effect is thus characterized by the magnetization transfer ratio (MTR):

Figure 4.

MT results from a healthy volunteer. (A) MT_off. (B) MT_on. (C) MTR map demonstrating an average 40% reduction in signal (ie, MTR = 0.4) in the fibroglandular tissue with good fat suppression.

Abbreviations: MT, magnetization transfer; MT_on, signal intensity with the saturation pulse; MT_off, signal intensity from the reference image; MTR, magnetization transfer ratio.

$$MTR=1-M{T}_{on}/M{T}_{off},$$ (5)

where MT_on represents the signal intensity with the saturation pulse and MT_off represents the signal intensity from the reference image. The MTR has been shown to be related to the amount of macromolecular protons in tissue.⁷⁶ It should be noted that since the MTR measurement is affected by the relaxation and exchange rates, the field strength, and the RF irradiation power, it is only a semiquantitative metric. A quantitative MT approach is necessary to separate the contributions from MT and relaxation effects,^77,78 and the development of clinically feasible quantitative MT techniques is an active area of research.⁷⁹ For a more extensive discussion of MT methods, the interested reader is referred to the review by Gochberg and Lepage.⁸⁰

Opportunities and preliminary applications in breast cancer

While the primary application of MT has been in studies of demyelinating diseases,^81–83 the demonstration of MT’s sensitivity to collagen content in meningiomas⁸⁴ has led to interest in MT for studying changes in the extracellular matrix (ECM) of the breast. The ECM is a major component of the fibroglandular tissue of the breast and is comprised of a network of macromolecules, including collagen, fibronectin, and laminin. The ECM is known to play a role in tumor development and progression,^85–87 and the role of the ECM in breast cancer has gained increasing attention. For example, Ioachim et al⁸⁸ reported that the expression of the ECM macromolecules fibronectin, collagen type IV, and laminin is altered in breast cancer; Levantal et al⁸⁹ recently demonstrated that cross-linking of collagen type I is involved in the stiffening of the ECM and tumor migration in breast cancer; and Kakkad et al⁹⁰ reported that the concentration of collagen type I is reduced in the hypoxic tumor environment. An initial application of MT in breast cancer showed a significant reduction of MTR in malignancies compared with benign tumors, thought to represent a reduction of the macromolecular pool due to increased proteolytic activity.⁹¹ In the NAT setting, it is hypothesized that changes in macromolecular content in response to successful treatment would result in measurable changes in MTR values. Also, as tumors respond to therapy and shrink, the tumor cells are typically replaced with fibrotic tissue,^92–94 which potentially could lead to an (ultimate) increase in MTR in the case of successful therapy.

Chemical exchange saturation transfer (CEST) MRI

Theory

CEST is similar to MT, but rather than focusing on macromolecules, CEST seeks to specifically irradiate tissue metabolites such as amides, amines, and hydroxyl groups that are also in exchange with free water.^95–97 The exchangeable protons on these metabolites have chemical shifts that are significantly smaller than the broad macromolecular pool and exchange at significantly slower rates, which allows for spectrally selective irradiation via application of an off-resonance RF pulse.^98,99 Under experimental conditions, CEST has been shown to discriminate individual tissue metabolites with high specificity. In addition, because proton chemical exchange rates are pH-dependent, CEST can be used to interrogate for changes in tissue pH.¹⁰⁰ Like MT, CEST can be performed without the use of exogenous contrast agents.

CEST is performed through application of a spectrally-selective saturation pulse prior to an imaging sequence, as shown in Figure 5. This series is repeated while the resonance offset of the saturation pulse is swept through a range of frequencies, typically < ±10 ppm. The saturation will affect specific protons, and the observed water signal will be attenuated via direct chemical exchange.¹⁰⁰ CEST results are often examined via a z-spectrum, a plot of the signal intensity of water as a function of saturation offset, normalized by the signal intensity of water in the absence of saturation.¹⁰¹ Example z-spectra are shown in Figure 6, which depicts the results of CEST analysis at 3 T on a malignant breast tumor (black line) compared with healthy fibroglandular tissue (gray line). One way the CEST effect can be characterized is by examining the amount of asymmetry observed in the CEST spectra, with the effect at particular offset frequencies related to the exchanging protons of interest.

Figure 5.

General pulse sequence diagram for a CEST MRI experiment.

Note: RF irradiation for a time t_s with an amplitude of B₁ precedes the excitation (α degrees) and image acquisition.

Abbreviations: CEST MRI, chemical exchange saturation transfer magnetic resonance imaging; RF, radiofrequency.

Figure 6.

Example z-spectra arising from a CEST MRI experiment at 3 T.

Note: The normalized signal (S/S₀) is shown as a function of saturation offset frequency for regions of interest in malignant tumor (black line) and healthy fibroglandular tissue (gray line).

Abbreviations: CEST MRI, chemical exchange saturation transfer magnetic resonance imaging; S, signal intensity with saturation; S₀, signal intensity in the absence of saturation.

Perhaps the most widely reported CEST effect is derived from the exchange of amide protons on the backbone of proteins and peptides. In this case, the CEST effect of the amide proton resonance can be characterized by the so-called proton transfer ratio (PTR), a measure of the asymmetry of the z-spectrum about the water frequency (CEST_asym) at Δω = 3.5 ppm:

$$CES{T}_{asym}=\frac{S\left(-\Delta \omega \right)-S\left(\Delta \omega \right)}{S_0},$$ (6)

where S(−Δω) and S(Δω) are the signal intensities with the saturation at ±Δω, and S₀ is the signal intensity in the absence of saturation. This calculation negates the confounding effects of direct water saturation, which are symmetric about the center of the z-spectrum.¹⁰² The measured PTR can be affected by the concentration of exchanging protons as well as the pH, which influences the proton exchange rate. Extensive research is underway to determine the underlying mechanism driving the measured CEST effect in both healthy and diseased tissue. Ongoing research is also investigating innovative pulse sequences^103,104 and alternative quantification strategies^105–107 that may minimize asymmetric magnetization transfer effects from macromolecules, which can confound the PTR measurement.¹⁰⁸ For a more comprehensive introduction to CEST, the reader is referred to the review by Gochberg and Lepage.⁸⁰

Opportunities and preliminary applications in breast cancer

CEST focused at the amide proton resonance (3.5 ppm) provides information on the amide protons of protein/peptide backbones and has been used to study tissues where either the protein/peptide concentration or the pH may be altered.^95,109–111 This technique, sometimes termed amide proton transfer (APT) imaging, is thought to be especially relevant to cancer imaging because tumor cells may accumulate defective proteins at a higher rate than normal and/or experience alterations in pH due to hypoxia.¹¹² APT-CEST has been applied to brain,¹¹⁰ prostate,¹¹³ and breast tumors¹⁰⁶ and has been used to differentiate cellular protein content between tumor and healthy cells.¹¹⁴

Our group is actively developing APT-CEST imaging as a potential early treatment response biomarker for breast cancer NAT.^106,115 Figure 7 depicts the measured APT overlaid on an anatomical image for three breast cancer patients before (left column) and after (right column) one cycle of NAT. The top row shows imaging data from a complete responder (27% decrease in measured APT from baseline), the middle row from a partial responder (49% increase in measured APT), and the bottom row from a patient with progressive disease (78% increase in measured APT). These preliminary results demonstrate the potential sensitivity of APT-CEST to the molecular changes occurring early during treatment.

Figure 7.

Amide proton transfer (APT) maps derived from CEST MRI in breast cancer patients undergoing NAT. Baseline (pretreatment) images are presented on the left, and images after one cycle of NAT are presented on the right. (A): patient with complete response after one cycle of therapy (27% decrease in measured APT from baseline). (B): patient with partial response (49% increase in measured APT). (C): patient with progressive disease (78% increase in measured APT).

Abbreviations: CEST MRI, chemical exchange saturation transfer magnetic resonance imaging; NAT, neoadjuvant therapy.

In addition to APT-CEST, there is potential to examine breast cancer by deploying CEST to evaluate tissue glycosaminoglycan content (ie, gagCEST). Many different cell surface and matrix proteoglycan core proteins are expressed in the mammary glands. The level of expression of these core proteins, the structure of their glycosaminoglycan chains, and their degradation are regulated by many of the same effectors that control development and function.¹¹⁶ Loss or overexpression of proteoglycans in carcinoma cells has been associated with malignant progression^117,118 and has correlated with poor prognosis,¹¹⁹ leading to the hypothesis that gagCEST may be developed as a treatment response biomarker in the future. The application of CEST targeting signatures of active tumors brings potential for noninvasive molecular imaging that could be predictive of prognosis.

Magnetic resonance elastography (MRE)

Theory

MRE is based on use of the elastic properties of tissue as an imaging contrast mechanism. The general concept of elastography, realized first in ultrasound¹²⁰ and later developed in MRI,¹²¹ involves the use of imaging to measure tissue response to applied physical deformation. Sometimes described as a form of “21 st century palpation,” elastography allows for generation of tissue elasticity maps providing spatial visualization and quantification of the distribution of elastic properties within an object. Generally, elastography methods employ the simplifying assumption of a linear elastic isotropic constitutive model where mechanical equilibrium is governed by:

$$\nabla \cdot G\nabla \overrightarrow{u}+\nabla \frac{G}{1-2v}\left(\nabla \cdot \overrightarrow{u}\right)+\overrightarrow{F}=-\left({\rho }_{tissue}-{\rho }_{fluid}\right)\overleftarrow{g},$$ (7)

where G is the shear modulus, u is the displacement vector, ν is Poisson’s ratio, F are body forces, ρ is density, and g is the gravitational constant vector. Given a measured tissue displacement field along with appropriate assumptions, the above equation can be used to reconstruct the spatial distribution of shear modulus (or Young’s modulus, E = 2G[1+ ν]).

MRE methods can be broadly classified as either dynamic or static. With dynamic excitation MRE, shear waves are applied to an area of interest by piezoelectric or pneumatic sources; oscillating motion-sensitized gradient sequences are synchronized to the externally-applied excitation, and induced three-dimensional tissue motion is recorded with phase-contrast imaging.^121–124 Static MRE methods^125–128 reconstruct elastic properties of tissue by incorporating a biomechanical finite element model into the nonrigid registration of images acquired under different static loading environments; the application of this technique is not unique to MRI modality, and in fact this methodology is also referred to as modality-independent elastography (MIE)¹²⁸ due to its foundation in image processing. All MRE methods require specially designed hardware to couple the deformation source (either static or dynamic) to the area of interest. Challenges to successful application of MRE include synchronization of the dynamic deformation source to the motion-sensitized MR gradient and often complex biomechanical model-driven inversions of the deformation field to reconstruct tissue elasticity maps. For a more extensive review of MR-based elastography techniques, the interested reader is referred to Mariappan et al¹²⁹ and Manduca et al.¹³⁰

Opportunities and preliminary applications in breast cancer

The rationale for exploring MRE as a possible breast cancer treatment response biomarker derives from observations that cancer progression is accompanied by extracellular matrix remodeling and increasing mechanical stiffness.⁸⁹ MRE has been deployed in liver tumor assessment and has had promising preliminary success in distinguishing between benign and malignant lesions, with malignant tumors exhibiting a significant (>3–4 fold) stiffness increase over benign tumors.¹³¹ While there are very few studies of MR-based elastography in breast cancer, preliminary results show initial promise in quantifying in vivo stiffness of breast tumors (as well as differentiation of fibroglandular and adipose tissue)¹³² and have demonstrated the ability to improve diagnostic sensitivity and specificity over more traditional breast MRI.¹³³

Our group is actively developing MRE as a mechanical property biomarker in breast cancer, and our novel MIE method has been under preliminary investigation in breast cancer.^126,127 Recent breakthroughs involving translation and automation¹²⁷ have generated promising preliminary results towards the ability to spatially reconstruct tissue elastic mechanical properties in breast cancer. Figure 8 depicts the results of MIE analysis on a breast cancer patient.

Figure 8.

Static MRE (MIE) results from a breast cancer patient. (A) undeformed image volume. (B) deformed image volume. (C) undeformed central slice. (D) deformed central slice. (E) reconstructed tissue elasticity map.

Abbreviations: MRE, magnetic resonance elastography; MIE, modality-independent elastography.

Monitoring the changes in mechanical properties along the NAT time course may provide a means of assessing response to therapy. Additionally, quantitative mechanical property information can also be used to inform patient-specific reaction-diffusion tumor growth models that are mechanically coupled to constrained tumor cell diffusion.¹³⁴ The growth and tumor cell diffusion parameters within the tumor growth models can be fit between the initial and intermediate time points and then projected forward to the final time point in order to help predict patient response.^135,136

Hyperpolarized MR

Theory

Hyperpolarized MR seeks to overcome the limitations of conventional MRS by exploiting exogenous contrast agents that have been “hyperpolarized,” ie, a large proportion of their nuclear spins have been aligned with the magnetic field or polarized. Whereas conventional MR imaging depends on spins that have been polarized on the order of a few parts per million, hyperpolarization allows for polarizing nuclear spins to nearly unity. The resulting several-orders-of-magnitude increase in signal intensity translates into increased sensitivity for detecting metabolic markers of cancer such as lactate and bicarbonate that may be present at millimolar and potentially submillimolar concentrations^137,138 and that may not be amenable to conventional MRS.^137,139 The technique may also be exploited to provide better spatial resolution and significantly faster examination times compared with conventional proton MRS. Long-lived nuclear spin sites are typically selected for the preparation of hyperpolarized contrast agents that are suitable for injection into living organisms. Carbon-13 (¹³C) sites without directly-attached protons are most commonly used, due to their long T₁ of ~ 20–40 seconds and their abundance within many metabolically-relevant molecules. At present, clinical translation of hyperpolarized MR technology is limited by demanding instrumentation and software requirements, including multinuclear MRI scanner capability, highly specialized RF pulse sequences (tailored to the RF coil, magnetic field strength, and metabolic contrast agent), and multinuclear RF coils.

Opportunities and preliminary applications in breast cancer

While in vivo hyperpolarized MR technology is less than 10 years old, it has rapidly progressed from proof-of-principle studies in mice to the first clinical trial in humans, including analyses of choline, pyruvate, fumarate, and bicarbonate as potential biomarkers.^{137,140–143} There have been as yet no specific clinical applications to breast cancer, but preliminary biomarker studies in mice and in human prostate cancer suggest this technique may be applicable for breast cancer NAT response assessment in the future.¹⁴³

Challenges and opportunities for clinical translation

Several important challenges must be addressed in working toward the translation and adoption of quantitative MRI techniques. First, investigators will have to choose meaningful and clinically relevant statistical methods for validating novel MRI methods. Prior studies of MRI as a response assessment tool for NAT in breast cancer have chosen a variety of clinical outcome variables – including palpation, imaging-based size measurement change, and pathologic response – against which to evaluate imaging as a response biomarker. Pathologic response is the most clinically relevant of these outcome variables, having been established as an independent prognostic marker for overall survival in breast cancer,³ yet researchers attempting to validate their methods against pathologic response will have to contend with the multiple different definitions of pathologic response now circulating in the breast cancer literature as well as the notion raised in certain recent studies that pathologic response may be a suitable surrogate endpoint for some but not all breast cancer subtypes.¹⁴⁴ Widespread incorporation of quantitative MRI response biomarkers in the clinical and research settings will probably require validation not just as predictors of pathologic response, but as predictors of long-term outcomes including recurrence rates and overall survival.

A second challenge involves ensuring standardization and reproducibility of MRI methods. MR imaging is a complex undertaking, with images and quantitative measurements dependent upon multiple precise software settings and hardware configurations. Results of a controlled experiment on a single scanner in a laboratory environment may be quite difficult to reproduce across multiple imaging sites. Researchers will have to work with vendors to increase reproducibility across platforms, and vendors will likely be called upon to provide increased transparency with regard to proprietary hardware and software architectures. A handful of government–industry consortia including the Radiological Society of North America’s Quantitative Imaging Biomarkers Alliance are attempting to facilitate these developments. We may also see continued incorporation of quantitative MRI techniques into commercially available computer-assisted detection software.

A third challenge for clinical translation lies in designing MR imaging protocols to meet the various objectives of a clinical scan. As mentioned earlier, there are two chief goals for breast imaging during NAT – response assessment and demonstration of residual disease – and these goals may call for different MRI approaches that may be mutually exclusive. For example, basic MRI principles dictate a fundamental tradeoff between spatial and temporal resolution–in general, high spatial resolution images require longer acquisition times and therefore a necessary sacrifice of temporal resolution, and high temporal resolution images, by virtue of the rapidity with which they are acquired, do not provide sufficient time to gather high spatial resolution data. A scan tailored for demonstration of residual disease may call for high spatial resolution imaging, while a DCE-MRI scan tailored for early response assessment may call for high temporal resolution imaging in order to provide the most accurate model of vascular flow. These tradeoffs highlight the challenges in clinical MRI protocol design, where different clinical objectives sometimes compete with each other. Potential solutions may emerge from creating hybrid protocols incorporating a variety of imaging techniques; from exploring newer methods, such as compressed sensing, that may make the above tradeoffs less apparent; and/or from employing different scanning protocols for different settings such as, a high temporal resolution MRI protocol early in the course of therapy and a high spatial resolution protocol at the end of therapy as an adjunct to surgical planning.

Finally, investigators across multiple disciplines will have to work toward overcoming the challenge of information saturation that clinicians will inevitably face as the methods discussed in this article are translated into the clinical setting. As more quantitative MRI metrics are validated as useful biomarkers, we may see development of integrated clinical scoring systems that synthesize multiparametric imaging variables with other quantitative clinical parameters.

Conclusion

There is growing demand for objective and standardized early-response biomarkers for breast cancer NAT, with such biomarkers to be used as an adjunct to and possibly eventually a replacement for pathologic assessment of treatment response. The imaging research community is responding to this demand by developing and testing novel approaches in preclinical models, in single site studies, and in large multisite clinical trials. Quantitative MRI techniques, including the current and emerging methods reviewed in this contribution, hold great potential for incorporation into both clinical and research algorithms. As these techniques are validated and correlated with long-term clinical outcomes, we may witness a broad transformation in the use of imaging with breast cancer NAT.