Detecting blood oxygen level-dependent (BOLD) contrast in the breast

Authors


Abstract

Purpose:

To develop a robust technique for detecting blood oxygenation level-dependent (BOLD) contrast in the human breast and to evaluate the signal in healthy and malignant breast.

Materials and Methods:

The design of this study focused on determining the optimal pulse sequence and stimulus for detecting BOLD contrast in the breast. For this study a single-shot fast spin echo (SSFSE) sequence was compared to a gradient echo (GRE) pulse sequence. Also, several hyperoxic stimuli were tested on 15 healthy volunteers to determine the best stimulus for inducing BOLD contrast in the breast: air interleaved with carbogen (95% O2, 5% CO2), air interleaved with oxygen, and oxygen interleaved with carbogen. The stimulus with the most consistent results among the healthy population was tested on three breast cancer patients.

Results:

An SSFSE pulse sequence produced improved BOLD contrast results in the breast compared to a GRE pulse sequence. Oxygen interleaved with carbogen yielded the most consistent results in the healthy population. BOLD contrast in healthy glandular breast tissue positively correlates with carbogen and malignant tissue mostly negatively correlates to carbogen.

Conclusion:

BOLD contrast can consistently be detected in the breast using a robust protocol. This methodology may be used in the future as a noninvasive method for evaluating tumor oxygenation. J. Magn. Reson. Imaging 2010;32:120–129. © 2010 Wiley-Liss, Inc.

EARLY DETECTION of breast cancer notably decreases mortality due to the disease (1), with magnetic resonance imaging (MRI) becoming an increasingly more valuable technology for breast cancer detection (2, 3). Currently, MRI diagnostic protocols use dynamic contrast-enhanced (DCE) MRI, requiring intravenous injection of an exogenous contrast agent such as gadolinium. The contrast helps distinguish malignant areas of the breast by highlighting unusual morphology and depicting areas with increased perfusion of blood with rapid contrast uptake and clearance kinetics (4). These areas of concern are usually biopsied for definitive diagnosis.

Beyond DCE-MRI, MRI technology has the potential to be increasingly useful in breast cancer detection and management. Functional MRI (fMRI) depicts hemodynamic response in a tissue due to changes in oxygenation using blood oxygen level-dependent (BOLD) contrast (5). Traditionally, this technique has been used to study the brain but has the potential to evaluate tumor metabolism and angiogenesis (6–14). Variations in oxygenation may give insight on tumor types (7), predict susceptibility to antiangiogenic therapeutics (8), and monitor chemotherapeutics (15). Developing fMRI for breast imaging thus may lead to improvements in breast cancer detection, treatment planning, and treatment monitoring.

BOLD contrast MRI, first described by Ogawa and Lee (16), is based on the paramagnetic effects of deoxyhemoglobin. In MRI this phenomenon is reflected as a change in the T2* relaxation rate of a tissue. BOLD contrast imaging has mostly been utilized to study brain function, where local changes in metabolic rate of oxygen due to a cognitive task stimulate vasomotor responses that in turn result in fluctuations in oxygen content. Thus, brain BOLD MRI detects measurable changes in deoxyhemoglobin that localize a cognitive task to an area of the brain.

BOLD contrast has also been evaluated for studying function outside of the brain, where an external stimulus is required to induce a variation in oxygenation in the tissue, such as breathing carbogen (95% oxygen, 5% carbon dioxide) or pure oxygen modulated with room air. There are several studies evaluating BOLD contrast outside of the brain such as in human skeletal muscle (17), human renal tissue (18), rodent tumor models (8–11), human prostate tumors (19), brain tumors (6, 12), and one study by Taylor et al (7) that evaluated 37 tumor patients with 17 different types of tumors, one of which was a breast adenocarcinoma. Taylor et al's article motivated our group to further evaluate BOLD contrast in breast tissue and develop a robust methodology for this effort.

We suggest that breast BOLD contrast MRI provides complementary information to that of DCE imaging. An important article that presents an interpretation of BOLD contrast in tumors was published by Gilad et al in 2005 (8). The complementary benefits of DCE and BOLD contrast imaging were evaluated in a mouse ovarian tumor model. The authors note that DCE imaging provides information on the permeability of a tumor, while BOLD contrast imaging conveys vasoreactive maturation information. For the BOLD application the theory is that tumor angiogenesis usually causes hyperproliferation of immature vasculature. These immature vessels do not properly develop smooth muscle vasculature to appropriately respond to a vasoreactive challenge such as vessel dilation due to hypercarbia, while healthy vasculature will appropriately dilate. In correlating BOLD and DCE MRI to histology, they concluded that in mouse ovarian carcinoma xenografts, tumors were heterogeneously composed of hyperpermeable tissue composed of both mature and immature vasculature as opposed to surrounding healthy, less permeable mature vasculature. This insight helped formulate our hypothesis: when provided a hyperoxic stimulus, healthy breast tissue induces a significant BOLD response while malignant tissue's BOLD response is insignificant. The hypothesis is based on healthy mature vessels reacting to a hyperoxic stimulus while less mature, malignant vessels do not respond as well.

All of the above-discussed tumor studies suggested the value of BOLD contrast in evaluating the oxygenation status of cancer. There does not seem to be agreement on a robust methodology for measuring BOLD contrast in human tumors, and the method may need to vary depending on the target structure. The purpose of our research was to develop a robust method for measuring BOLD contrast in the human breast and to introduce preliminary differentiating characteristics between healthy and malignant breast tissue. To date, no major study has been published that evaluates fMRI BOLD contrast in healthy and malignant human breast tissue.

MATERIALS AND METHODS

Measuring BOLD contrast in the breast poses several challenges. Previous experiments (20–22) indicated that the hemodynamic BOLD contrast response in breast tissue differs significantly from that of more established results in the brain. In the healthy breast, BOLD contrast varies between individuals because breast architecture and hemodynamic response are more highly variable in comparison to brain. There are three potential causes that we took under consideration when designing the experimental methodology. First, B0 magnetic susceptibility effects, the same mechanism for providing traditional BOLD contrast for brain imaging, cause large dynamic artifacts in breast imaging because of motion (due to respiration and gross body motion) and the shape of the breast tissue. This is especially of concern at 3T. A second contribution to the signal variability includes the heterogeneity between volunteers' breast fat and glandular tissue, with glandular tissue being more vascular and thereby modulating the BOLD effect. Third, in a premenopausal population, particularly in evaluating breast tissue, menstrual cycle also contributes to hemodynamic variability because estrogen acts as an endogenous vasodilator and fluctuates cyclically (23). Taking these challenges into account, the experimental design focused on determining the optimal pulse sequence and BOLD stimulus for fMRI in the breast in a healthy population. The optimized protocol was then tested on a sample patient cohort.

Pulse Sequence Selection

BOLD contrast fMRI in the brain is usually conducted with a 2D gradient recalled echo (GRE) sequence sensitive to T2* changes. This works well in brain fMRI at 3T because the brain is mostly centered in the field and is nominally elliptically symmetric, and is thus situated in a relatively homogeneous B0 field. Breasts, however, create an adversely heterogeneous B0 field by nature of the geometry (24). Also, the curvature of the breast-body junction causes sharp nonuniform changes in the magnetic field, which exacerbates the heterogeneity effects. The resulting large B0 gradients lead to high sensitivity of signal strength to small motions of the breast.

Single-shot fast spin echo (SSFSE) sequences are not susceptible to B0 inhomogeneity because the periodic refocusing pulses eliminate sensitivity to off-resonance. They have been evaluated for use in detecting BOLD (25–33) contrast, although not in the breast. When using SSFSE, the T2 component solely contributes to the BOLD signal, and T2* is irrelevant.

Initially a 2D gradient echo (GRE) spiral pulse sequence with fat saturation and spatially selective heart saturation was designed for this study with the following scan parameters (21): 3T (GE Healthcare, Waukesha, WI), 8-channel breast coil (GE), TE = 30 msec, TR = 1.1 msec, flip angle = 60°, bandwidth = 125 kHz, matrix size = 80 × 80, field of view (FOV) = 20 cm, slice thickness/spacing = 5 mm/5 mm, 2 interleaves, 327 time frames/slice, 12 slices. Preliminary results yielded inconsistent data with often obscure BOLD contrast maps (Fig. 1). Regions of apparent activation often did not represent expected areas, for example, showing little differentiation between breast glandular tissue and fat. From these initial results we concluded that T2*-weighted GRE would not be adequate for robust functional breast imaging. Preliminary results obtained with SSFSE provided more robust and consistent data (22) compared to the GRE experience (Fig. 1), and thus was selected for use in this study.

Figure 1.

Correlation coefficient maps of BOLD signal collected with (a) GRE and (b) SSFSE with the air vs. carbogen stimulus on a healthy volunteer. Inset in (b) is the anatomic image of this volunteer and should correlate to the anatomy in both (a) and (b). Note that on the GRE image, correlation is random and not necessarily localized to the anatomy.

Stimulus Selection on Healthy Volunteers

Determining the optimal stimulus for inducing BOLD contrast in the breast depends on choosing 1) an appropriate stimulating agent, and 2) the timing of the stimulus as it relates to the hemodynamic response function in breast tissue. Stimulating a vasomotor response and thus BOLD contrast in the breast can be performed noninvasively by breathing an exogenous gas agent that modulates breast tissue oxygenation. Hyperoxia and hypercarbia are obvious choices, but each has its limitations. Pure oxygen increases oxygen delivery by increasing the partial pressure of oxygen (PaO2) in the vascular system. This could lead to increased tissue oxygenation and consequential positive BOLD contrast (oxyhemoglobin correlates to a positive BOLD contrast signal). However, a competing mechanism is that the hyperoxia causes a consequential decrease in carbon dioxide in the vasculature and may lead to vasoconstriction. Despite delivering a hyperoxic gas, the resulting vasoconstriction can decrease the local tissue oxygenation, producing a negative BOLD contrast (7, 34) (deoxyhemoglobin correlates to a decrease in, or negative BOLD contrast signal). Thus, the tradeoffs between these two mechanisms can lead to variability between subjects due to biodiversity. A major benefit of pure oxygen is that it is well tolerated by subjects and is readily available in medical environments. Carbogen (95% O2, 5% CO2) also increases oxygen delivery by providing a hyperoxic stimulus while maintaining vasodilation. It has been shown to be a robust stimulus, although other reports note that subjects found carbogen difficult to breathe (7, 35, 36). Overall, carbogen may provide a more predictable result, but oxygen is easier to breathe (36).

In selecting the stimulating agent, 15 studies were conducted on healthy premenopausal women (ages 24–29). The study limited enrollment to the premenopausal population for future evaluation of the effects of menstrual cycle on BOLD contrast. Menstrual cycle effects are not evaluated in this article. The local Institutional Review Board (IRB) and cancer center approved the protocol for healthy volunteers and patients. Three variations of hyperoxic stimuli were evaluated on the healthy volunteers (not all volunteers received each stimulus due to scan time restraints; Table 1) using a design that switched between blocks of two gases: 1) room air vs. oxygen, 2) room air vs. carbogen, 3) oxygen vs. carbogen. The two gases cycled in 4-minute periods with four cycles per task (Fig. 2), totaling 16 minutes per paradigm. The 4-minute period length was established by accounting for the uptake time for oxygen saturation, which is ≈1 minute (37). Oxygen and carbogen were delivered at 9 L/min through a nonrebreathing mouthpiece with gas delivery controlled by valves and an E-Prime script (Psychology Software Tools, Pittsburgh, PA). Gas delivery and exhalation were monitored and recorded by a capnograph (Datex Capnomac Ultima; Datex, Helsinki, Finland). MR-compatible nose plugs were used to prevent uncontrolled inhalation. An SSFSE pulse sequence was used with the following imaging parameters: 3T (GE Healthcare), 8-channel breast coil (GE), TE = 60 msec, TR = 4 sec, bandwidth = 83 MHz, matrix size = 128 × 128, 80 phase encode lines, FOV = 20 cm, 240 time frames/slice, 1 sagittal (slice thickness/spacing = 5/5 mm) or oblique/coronal slice (slice thickness/spacing = 3/1 mm), fat suppression. A respiratory belt and pulse oximeter were placed on each subject to record respiratory function and cardiac rate.

Table 1. Root Mean Square BOLD Signal Change, Percent, per Subject per Stimulus
Volunteer/patientOxygen vs. Carbogen yrms/# voxels* threshold P-valueAir vs. Oxygen yrms /# voxels* threshold P-valueAir vs. Carbogen yrms /# voxels* threshold P-valueAir only yrms /# voxels* threshold P-value
  • *

    The number of voxels were those that remained after masking the image meeting the above listed P-value thresholds determined by the correlation coefficient maps. NC, not conducted.

Volunteer 10.31/124NCNCNC
0.05   
Volunteer 21.7/4060.92 /2341.05/2260.69/6
0.050.050.050.05
Volunteer 32.24/2291.08/1311.59/196NC
0.220.220.22 
Volunteer 40.73/2271.26/221.17/62NC
0.270.060.27 
Volunteer 51.2/921.22/251.32 /112NC
0.220.220.22 
Volunteer 6NC3.08/1921.23/125NC
 0.050.05 
Volunteer 71.35/199NC1.34/82NC
0.05 0.05 
Volunteer 80.75/1470.61/90.61/41NC
0.060.060.06 
Volunteer 96.91/30NC1.10/157NC
0.05 0.05 
Volunteer 10NC1.11/1550.97/190NC
 0.050.05 
Volunteer 110.65/180.52/601.22/23NC
0.050.050.05 
Volunteer 120.56/851.52/1870.56/125NC
0.050.050.05 
Volunteer 130.59/1320.57/232NCNC
0.050.05  
Volunteer 140.96/901.2/1241.02/111NC
0.050.050.05 
Volunteer 151.46/7700.69/4910.82/224NC
0.050.050.27 
Patient 10.74/182NCNCNC
Healthy glandular0.03   
Patient 10.88/52NCNCNC
tumor0.03   
Patient 21.07/109NCNCNC
Healthy glandular0.24   
Patient21.27/41NCNCNC
tumor0.05   
Patient 31.53/7NC1.32/11NC
Healthy glandular0.05 0.05 
Patient 30.71/5NC1.56/13NC
tumor0.05 0.05 
Figure 2.

Block design for 16-minute BOLD stimulus. The three stimuli variants included air vs. carbogen, air vs. oxygen, and oxygen vs. carbogen. Gas 1 corresponds to the first gas listed in each stimulus. For example, Gas 1 is air and Gas 2 is carbogen in the air vs. carbogen task.

In this study we conducted a simultaneous study with near infrared (NIR) optical imaging on seven of the 15 healthy subjects and on the three patients. Results of that study are not presented here, but did affect how data were collected. In particular, when simultaneous NIR imaging and BOLD imaging were conducted, light breast compression was necessary to obtain contact of the optical probes with the breast tissue. In addition, although most breast MR is conducted in the sagittal or axial planes, coronal imaging was employed to match image planes for the simultaneous NIR experiments.

Analysis

During preprocessing, RETROICOR was used to reduce image noise from respiratory motion and cardiac pulsation in time (38). Next, the BOLD signal time series for each voxel was cross-correlated with a sinusoidal model of the periodic stimulus. A sinusoid model approximated the hemodynamic response function (HRF) of the stimulus in breast tissue. In general, the BOLD response is modeled as the stimulus design convolved with the HRF. In the case of a hyperoxic stimulus, the hemodynamic response takes effect on the order of 30 seconds and plateaus on the order of several minutes later (17). We were able to model the HRF as a sinusoid because the block periods were 4 minutes long (2-minute intervals of each gas), and the hemodynamic processes heavily filtered the time series. By inspection of the received signal and preliminary modeling using a 2-minute “on” state, higher-order harmonics were filtered out, allowing the HRF to be approximated with a single sinusoid. In order to account for unknown delays, we also added a cosine wave, allowing calculation of both the magnitude and phase of the response (39). As the next step in preprocessing, a sigma filter was applied which averages nearby voxels with less than one standard deviation from the center voxel, thus eliminating noisy single voxels. The first cycle of data was not used in the analysis to allow equilibration of the gas inhalation regimen and reduce effects of a different regimen from the previous scan. This analysis provided correlation coefficients and phase lags between the delivered stimuli and the measured BOLD contrast for each voxel. Activation maps were produced based on the correlation coefficients at the phase with maximum correlation.

To evaluate BOLD signal changes, a region of interest (ROI) mask was generated that eliminated signal in the fat on the outer border of the breast tissue, and included voxels whose activation exceeded a threshold, as shown in Table 1. Time series for the last three cycles of the detected signal were then extracted from this ROI and correlated with the sinusoidal model to obtain Pearson correlation coefficients for each volunteer and stimulus. The root mean square BOLD signal yrms in percent was calculated from the variance explained by the modeled fit determined from the correlation and measured variance in the time series:

equation image(1)

where σy is the standard deviation of the measured signal in the ROI, ȳ is the average of the signal, and r is the correlation coefficient.

In determining the optimal stimulus among the three stimuli variants, we evaluated the maximum correlation coefficients of the stimulus to the detected BOLD signal and maximum phase lag of the BOLD signal relative to stimulus onset. Circular mean and standard deviation analysis was conducted in evaluating the phase lag data to account for phase lag wrapping every 4 minutes (4 min = 2π) (40, 41). For this analysis, consider a set of phases, φi, where i = 1 … n, and n is the number of studies evaluated per given stimulus. The mean is defined by:

equation image(2)

where:

equation image(3)

and the standard deviation is defined by:

equation image(4)

An F-test was then conducted to determine the significance of the circular standard deviations between the following stimuli comparisons: 1) oxygen vs. carbogen and air vs. oxygen, 2) oxygen vs. carbogen and air vs. carbogen, and 3) air vs. oxygen and air vs. carbogen.

Patient Study

Three patients with high likelihood of breast malignancy based on a previous recent gadolinium dynamic contrast-enhanced 1.5T MRI scan participated in this study. Each patient was scanned with a multiphase Dixon sequence (42) for anatomic information with the following scan parameters: 3T (GE Healthcare), 8-channel breast coil (GE, Healthcare), TE = minimum, TR = 9 msec, flip angle = 12°, bandwidth = 62.5 kHz, matrix size = 512 × 192, FOV = 20 cm, slice thickness = 2 mm, number of slabs = 1, locations/slab = 32. The multiphase Dixon scan provided water-only images that allowed a radiologist to help correlate anatomic features to the previous gadolinium scan, so that redelivery of contrast was unnecessary. Next, the same BOLD contrast protocol applied to the healthy volunteers was used in the patients, but instead of using all three stimuli, only the optimal stimulus (oxygen vs. carbogen) was delivered. For one of the three patients a second task of air vs. carbogen was also delivered (this was because the first task failed to significantly differentiate the tumor from the glandular tissue and there was additional time for a second task). As with the healthy volunteers, correlation coefficient maps were calculated as were phase lags between the stimulus and the detected BOLD contrast signal.

RESULTS

Stimulus Selection

The initial intention of this study was to compare air vs. oxygen to air vs. carbogen to determine the optimal stimulus. In the first two studies where only these two gas stimuli were compared, air vs. oxygen and air vs. carbogen produced out of phase contrast (refer to volunteers 6 and 10 in Fig. 3). Thus, for the remainder of the studies oxygen vs. carbogen was added as a third stimulus option expecting that this combination would optimize contrast. Not all subjects were evaluated for all three conditions because of time constraints imposed by the length of each 16-minute scan and technical difficulties. In determining which stimulus provided the most consistent results across volunteers, we evaluated the maximum (in temporal phase) correlation coefficients between the task and the measured BOLD signal as well as the circular mean of the phase lag between the stimulus delivery and corresponding measured signal. Our evaluation focused on correlation analysis as a robust method for evaluating time series (43). BOLD signal intensity, extracted using Eq. [1], is shown in Table 1. For most volunteers a large number of voxels composed the ROIs. The voxels used for analysis were limited by the thresholding P-value determined by the correlation coefficient maps. The goal was to include voxels meeting a P-value of less than or equal to 0.05. We found, however, that in a few subjects we were unable to achieve any voxel signal with P less than 0.05 and thus increased the threshold to examine trends, using the P-values listed.

Figure 3.

Phase lag between each stimulus and detected BOLD signal per healthy volunteer. Note the clustering of the oxygen vs. carbogen phase lags between 0.2π and 1.2π.

The maximum correlation coefficient results shown in Fig. 4 indicate the trend that oxygen vs. carbogen provided higher correlation coefficients between the task and the measured signal, although insignificantly relative to the other tasks. The phase lag analysis employing the circular mean (to accommodate the cyclical nature of this comparison) indicated that oxygen vs. carbogen provided the smallest standard deviation in comparison to the two other stimuli variants (Table 2). In particular, an F-test evaluating the significance of the variance of oxygen vs. carbogen compared to air vs. carbogen yielded a P-value of 0.0009. For oxygen vs. carbogen compared to air vs. carbogen, P = 0.02. For air vs. carbogen and air vs. oxygen, P = 0.09. The phase lag results thus indicate that oxygen vs. carbogen is significantly less variable than air vs. oxygen or air vs. carbogen.

Figure 4.

Mean maximum correlation coefficient across volunteers per stimulus. Oxygen vs. carbogen produced the greatest correlation coefficient. However, the standard deviation (indicated by the error bars) did not significantly differentiate this task from the others.

Table 2. Mean Phase Lags Between BOLD Response and Block Gas Stimulus in Healthy Volunteers
Gas stimulusMean phase lagaStandard deviationNumber of studies
  • a

    The lag is comparing the received signal compared to the second gas listed in the “Gas Stimulus” column.

  • b

    F-test results indicated that the standard deviation for the oxygen vs. carbogen stimulus was significantly different than the air vs. oxygen results (P = 0.0009) and the air vs. carbogen results (P = 0.02).

  • c

    The difference between the standard deviations for the air vs. carbogen and air vs. oxygen results was insignificant (P = 0.09).

Oxygen vs. carbogen1 min 26 s (0.72 π)29 sb (0.24 π)13
Air vs. oxygen20 s (0.17 π)1 min 22 sc (0.68 π)12
Air vs. carbogen1 min 23 s (0.69 π)54 s (0.45 π)13

BOLD Results

In evaluating the BOLD signal we were particularly interested in the positive or negative correlation of the gases to the stimulus. We found in healthy volunteers that carbogen results in positive BOLD contrast, ie, signal greater during carbogen than air (Fig. 5, Table 2) and that oxygen generally yields a negative BOLD contrast, although more variably. The phase lag values in Table 2 do not account for the 1-minute time delay (π/2 phase lag) for blood oxygen saturation effects given an inhaled hyperoxic gas stimulus. Due to the time delay for oxygen saturation, we consider a phase lag between 0 and 2 minutes (0 and π) to be positive correlation and between 2 and 4 minutes (π and 2π) to be negative correlation. In Fig. 5 the activation maps demonstrate the maximized BOLD signal with the oxygen vs. carbogen stimulus. In this volunteer, an air-only study for 16 minutes was conducted to demonstrate potential false positives from random correlation.

Figure 5.

Top row: Correlation coefficient maps at the phase of maximum correlation, overlaid on an average coronal image on one volunteer over each stimulus. The color bar indicates the correlation coefficient range multiplied by 1000. Bottom row: corresponding time series to the drawn ROI in the top row. The blue signal is the detected BOLD signal and the green bars represent the second listed gas. This volunteer provided maximum contrast with the oxygen vs. carbogen stimulus.

Patient Study

For two out of the three patients, results indicated inverse correlation in the tumor tissue relative to the healthy glandular tissue (Table 3) with the oxygen vs. carbogen stimulus. Thus, when healthy breast tissue produced a positive BOLD response to carbogen, tumor tissue produced a negative BOLD response. The third patient's tumor did not differentiate from the glandular tissue with the oxygen vs. carbogen stimulus. We noticed this result immediately after collecting the data (and verified that the patient appropriately breathed the stimulus via the capnograph). We had scan time to spare with this patient and thus added another stimulus paradigm of air vs. carbogen. With this second paradigm, the third patient's tumor responded similarly to the previous two, with a negative BOLD response correlating to carbogen, while the healthy tissue produced a positive BOLD response. Overall, the healthy glandular tissue in the patient population matched positive correlation results to carbogen in the healthy volunteers. The tumor tissue, however, negatively correlated with carbogen delivery (Fig. 6). Later pathological analysis of the tumors' biopsy specimens indicated the following diagnoses: patient 1, invasive lobular carcinoma (2.1 × 1.2 cm); patient 2, invasive ductal carcinoma (1.8 cm); and patient 3, invasive ductal carcinoma (1.2 cm).

Table 3. Phase Lags Between BOLD Response and Block Carbogen Delivery in Patients
PatientStimulusTumor phase lagGlandular tissue phase lag
  • *

    Note the positive correlation of the healthy glandular tissue in the patients (0 to 2 min or 0 to π) and the negative correlation between carbogen and the response for the tumors (ranging from 2–4 min or π τo 2 π).

1Oxygen vs. carbogen3 min 28 s (1.73 π)*1 min 19 s (0.66 π)*
2Oxygen vs. carbogen3 min 44 s (1.87 π)*1 min 16 s (0.63 π)*
3Air vs. carbogen4 min (2 π)*1 min 40 s (0.83 π)*
Figure 6.

BOLD contrast results for the three tumor patients. First column: Reformatted coronal DCE images from previous scan at 1.5T. The yellow arrows identify the tumors. Second column: Water images allowing anatomic identification of tumor (based on DCE image architecture). Third column: Correlation coefficient maps overlaid on average images. The color bars represent the correlation coefficients multiplied by 1000. Fourth column: Corresponding time series to an ROI encompassing the tumor. The blue signal is the detected BOLD signal and the green bars represent the second listed gas. (a) Invasive lobular carcinoma with oxygen vs. carbogen as the stimulus; (b) invasive ductal carcinoma with oxygen vs. carbogen as the stimulus; (c) invasive ductal carcinoma with air vs. carbogen as the stimulus.

DISCUSSION

Pulse Sequence

Changes is deoxyhemoglobin content lead to magnetic susceptibility effects that contribute to BOLD contrast signal in the brain. In the breast, however, magnetic susceptibility adversely affects detectable T2*-weighted signal in the breast because of field distortions caused by the breast geometry. Our proposed solution is to derive the BOLD signal from T2 relaxation using an SSFSE pulse sequence, rather than from T2* with a GRE method. In addition to reducing image distortion, by using T2 contrast the detected signal is more specific because it emanates from capillaries rather than larger vessels which contribute to T2* relaxation (26, 27, 29, 33). With T2 relaxation, diffusion contrast dominates the BOLD signal, which is therefore proportional to the square of the field strength (a significant benefit at 3T). The benefit of SSFSE therefore provided more reliable BOLD contrast in the breast compared to that obtained with GRE. The downside of the SSFSE sequence used in this study was the SAR limitation due to the long train (80 pulses) of high flip angle (155°) refocusing pulses. Consequently, only one or two slices could be collected over the 16-minute sequence without exceeding SAR limits. Potential solutions to this limitation include applying techniques such as variable flip angle pulse trains, lower-SAR pulse designs, GRASE, and parallel imaging (44–50).

Stimulus

Breathing 100% oxygen interleaved with carbogen in a block design produced the largest and most consistent BOLD contrast in the healthy breast in comparison to air interleaved with carbogen or air interleaved with oxygen. Previous studies have indicated that some patients find carbogen difficult to breathe (36), and this very well may be the case for breathing carbogen over an extended period of time. In this study, healthy volunteers and patients breathed carbogen for only 2 minutes at a time alternating with either air or oxygen for 2 minutes before breathing carbogen again. Participants did not find the carbogen difficult to breathe. The only reported discomfort for the participants in this study was laying prone on the breast coil for the extensive length of the study. Thus, based on healthy volunteers, we conclude that carbogen contrasted with pure oxygen may be the optimal stimulus for demonstrating BOLD effects in the breast. There is a caveat to this conclusion, however, since one of the three patient's tumors responded to the air vs. carbogen stimulus and not the oxygen vs. carbogen stimulus. The reason for this difference in response is unclear, but one possible explanation to this outlier among the patients and volunteers is that there is biodiversity of tumor oxygenation and patient vascularity. A stimulus may appropriately work for most people but not all people.

For this initial evaluation of BOLD contrast in the breast, the stimulus consisted of a 16-minute scan of 4 block periods. For this study we discarded the first block to allow for gas/tissue equilibrium after previous scans. With more experience and optimization, decreasing both the number of blocks and the length of each block may be possible and this certainly can decrease the scan time.

Understanding BOLD Contrast in the Breast

In healthy volunteers, this study determined that glandular tissue produced a positive BOLD signal in response to breathing carbogen. Furthermore, a hyperoxic stimulus in healthy vasculature led to a decrease in deoxyhemoglobin. The carbogen maintained vasodilation while delivering the hyperoxic stimulus. Consistent results were found in the healthy glandular tissue of the patients. These results agreed with the hypothesis that healthy vasculature will vasoreact appropriately with carbogen (ie, cause vasodilation) and consequently deliver more oxyhemoglobin to the region. The tumor results were, however, surprising. We hypothesized that healthy tissue, given mature vasculature would vasoreact appropriately, and that malignant tissue with immature vascular development would vasoreact relatively less. The assumption was that mature vasculature would have a positive BOLD response, as it did, and that malignant vasculature would produce a null or diminished positive response. In this study, tumor yielded a negative correlation to the carbogen, meaning that in response to a vasodilatory stimulus the tumor increased its deoxyhemoglobin content. This may indicate another mechanism is at play: the redistribution of blood in a tumor with the “steal effect” (51, 52). Others have concluded similar BOLD contrast results in tumors (35, 53), particularly in the one published breast adenocarcinoma result presented in Taylor et al's article (7).

The “steal effect” can explain why there is decreased blood flow to the tumors while the patients breathed carbogen. The decrease in blood flow led to the usual increase in deoxyhemoglobin, as the tissue had more time to metabolize the incoming oxygen. Thus, although tumors generally have more vasculature feeding the tissue, their lack of vasoreactive maturity may have prevented an increase in blood flow while other parallel vessels responded appropriately.

In our pilot study we consistently detected negative correlation to carbogen in tumors. When applying this proposed methodology for detecting BOLD contrast in breast tumors to a larger patient population, we expect to have more variable results representing the biodiversity of breast cancer tumor oxygenation. The tumor's oxygenation information, whether positive, negative, or heterogeneous may help characterize the tumor and guide treatment. Specifically, if a tumor is growing fast with immature vasculature (thus most likely to have negative BOLD signal correlated to a carbogen task), treatment targeting angiogenesis may be appropriate. While on the other hand, if a known tumor demonstrates well-developed vasculature the patient may be spared inappropriate therapy and directed otherwise. Further study of the BOLD contrast response in a larger breast cancer patient population is warranted and potentially useful.

In conclusion, we found that BOLD contrast imaging is a promising technique for evaluating breast tissue oxygenation. Results were greatly improved compared with GRE methods using a T2-weighted acquisition with a pulse sequence such as SSFSE, unaffected by magnetic susceptibility effects. We further found that a breathing stimulus comprised of carbogen interleaved with air provided the most consistent evaluation of BOLD contrast across subjects. In our participant study we found positive correlation of BOLD contrast to carbogen in healthy glandular breast tissue, and negative contrast in a small sample of breast cancers. This technique may provide a noninvasive method for evaluating variations in tumor oxygenation, opening avenues for improving breast cancer patient care.

Acknowledgements

The authors thank the volunteers. Also, we thank Laura Pisani and Stephanie Oberfoell for help with this project through its early phases. We thank Christine Law, Catie Chang, and Jason Hsu for many helpful conversations, and Anne Sawyer and Sandra Rodriguez for helping work through the logistics of scanning with a complicated set-up.

Ancillary