Transmit B₁⁺ Field Inhomogeneity and T₁ Estimation Errors in Breast DCE-MRI at 3T

Abstract

Purpose

To quantify B₁⁺ variation across the breasts and to evaluate the accuracy of pre-contrast T₁ estimation with and without B₁⁺ variation in breast MRI patients at 3T.

Materials and Methods

B₁⁺ and variable flip angle (VFA) T₁ mapping were included in our dynamic contrast-enhanced (DCE) breast imaging protocol to study a total of 25 patients on a 3.0T GE MR 750 system. We computed pre-contrast T₁ relaxation in fat, which we assumed to be consistent across a cohort of breast imaging subjects, with and without compensation for B₁⁺ variation. The mean and standard deviation of B₁⁺ and T₁ values were calculated for statistical data analysis.

Results

Our measurements showed a consistent B₁⁺ field difference between the left and right breasts. The left breast has an average 15.4% higher flip angle than the prescribed flip angle, and the right breast has an average 17.6% lower flip angle than the prescribed flip angle. This average 33% flip angle difference, which can be vendor and model specific, creates a 52% T₁ estimation bias in fat between breasts using the VFA T₁ mapping technique. The T₁ variation is reduced to 7% by including B₁⁺ correction.

Conclusion

We have shown that severe B₁⁺ variation over the breasts can cause a substantial error in T₁ estimation between the breasts, in VFA T₁ maps at 3T, but that compensating for these variations can considerably improve accuracy of T₁ measurements, which can directly benefit quantitative breast DCE-MRI at 3T.

INTRODUCTION

Dynamic contrast-enhanced MRI (DCE-MRI) is a widely used method in the diagnosis of breast cancer (1,2). The technique typically acquires a time series of T₁-weighted images before and after injection of an intravenous low molecular-weight paramagnetic contrast agent, and can be used to characterize lesions of breast tissue. Quantitative microvascular properties can also be extracted by either fitting the gadolinium concentration curve to a pharmacokinetic model (3) or computing initial area under the gadolinium concentration curve (4). Both high spatial and high temporal resolution are important to accurately estimate these quantitative properties, which can potentially provide predictive, prognostic and pharmacodynamic response biomarkers for cancers (5-7).

An image of pre-contrast T₁ values, registered to the dynamic series, is necessary to convert the dynamic MR data into the gadolinium concentration, where the concentration changes over time can be used to extract quantitative or semi-quantitative microvascular properties (8). One common method to measure T₁ is variable flip angle (VFA) imaging, also known as Driven Equilibrium Single-Pulse Observation of T1 (DESPOT1), which uses several short TR RF-spoiled gradient-echo (SPGR) acquisitions with varying flip angles (9-11). The VFA method is highly time efficient and allows rapid 3D volumetric T₁ mapping with high resolution (9,12), using the same pulse sequence that is used for the DCE acquisition itself, thus avoiding registration and calibration differences.

Non-uniformity of the transmit radio frequency (B₁⁺) field can cause the actual flip angle in tissue to be different from the prescribed or nominal flip angle. The B₁⁺ inhomogeneity tends to become more severe for higher field strengths, and at 3 Tesla, noticeable B₁⁺ variation over the chest has been observed by many studies (13-16). The flip angle variation tends to be around 30 - 50% in the chest (13,14) and 40% across the breast (15,16), and can result in significant deviation in T₁ measurements using VFA (17,18). Therefore, careful consideration of the B₁⁺ inhomogeneity will be important to achieve accurate T₁ mapping.

In this work, we measure B₁⁺ inhomogeneity and pre-contrast T₁ values to test the accuracy of VFA T₁ mapping in the breast at 3T using the spatial consistency of T₁ estimates of fat, derived from fat-only 2-point Dixon VFA images, as a metric of improvement. We quantify the B₁⁺ inhomogeneity in a total of 25 breast MRI patients at 3T and evaluate the accuracy of the T₁ measurements with and without compensation for B₁⁺ variation by comparing T₁ relaxation times in fat between breasts as well as with a previously-reported value of the breast at 3T.

MATERIALS AND METHODS

Experiments were performed on a 3.0T GE MR 750 system (GE Healthcare, Waukesha, WI). A body coil was used for RF transmission and a commercially available 8-channel high-density breast array coil (GE Healthcare, Waukesha, WI) was used for signal reception. No parallel imaging was used. The automatic pre-scan, provided by the scanner, was used to calibrate the RF transmission gain, and the difference in nominal transmit gains between T₁ and B₁⁺ measurements was applied to achieve a consistent flip angle variation between two sequences.

The B₁⁺ measurement sequence and the T₁ VFA measurement sequence were performed as part of our standard clinical breast DCE-MRI protocol. Although other faster alternatives for B₁⁺ mapping exist, we used a double angle method (DAM) to measure flip angle variation (19,20), as it is robust. We acquired both measurements in a total of 25 women undergoing clinically indicated breast MRI for a history of known or suspected breast disease, ranging in age between 26 and 73 years (age = 50.1 ± 11.4 years and mass = 62.4 ± 10.8 kg). The axial orientation was chosen for both measurements as it is commonly used in breast MRI, and a large B₁⁺ variation is expected from left to right. This retrospective review and analysis of the B₁⁺ and T₁ VFA data was performed in accordance with a protocol approved by our Institutional Review Board. More details on both measurement sequences are described below.

T₁ Measurements - Variable Flip Angle (VFA)

In VFA T₁ mapping, the measured signal intensity using SPGR (S_SPGR) can be used to compute a T₁ value in a linear form:

$$\frac{S_{\mathit{SPGR}}}{\sin \alpha }=E_1\frac{S_{\mathit{SPGR}}}{\tan \alpha }+M_0(1-E_1)$$ [1]

where α is the flip angle, E₁ = e^−TR/T₁ and M₀ is the longitudinal magnetization. By applying different flip angles α_n, we can generate different points (S_SPGR/sinα_n,S_SPGR/tanα_n) in the linearized form. The slope E₁ can then be estimated by linear regression, and T₁ can be extracted from E₁ using:

$$T_1=\frac{\mathit{TR}}{\ln \left(E_1\right)}$$ [2]

Note that Eq. [[2] is highly sensitive to any possible errors on the linear regression. Any perturbation on flip angles α_n can degrade the slope estimation, and for example, +1% and −1% errors in the slope can range T₁ to be from 200 ms to 2000 ms, when the actual T₁ is 370 ms (TR = 4.5 ms).

T₁ maps were measured by using a 3D SPGR sequence with a dual-echo bipolar readout, where TEs were chosen to be in- and out-of-phase images (TE = 1.2/2.4 ms). A two-point Dixon fat-water separation algorithm was used to generate fat-only and water-only images (21), which can eliminate partial volume effects of the admixture of fat and glandular tissue in breasts (22). We placed the T₁ mapping before the pre-contrast T₁-weighted imaging and selected two flip angles to be 5° and 10°, which were optimized to symmetrically sample the signal curve of fat (T₁ was assumed to be 400 ms and TR = 4 ms in numerical simulation). We used T₁ measurements on the lipid component in fat and compared T₁ relaxation in fat with and without compensation for B₁⁺ variation. Other imaging parameters were as follows: acquisition matrix size = 256×128×88, slice thickness = 4.2 mm, FOV = 32 cm, and total scan time = 20 sec.

B₁⁺ Measurements - Double Angle Method (DAM)

B₁⁺ maps were measured by using a 2D multi-slice SPGR sequence with prescribed flip angles of α and 2α (α = 60°). The actual flip angle map $\tilde{a}$ was calculated as

$$\tilde{\alpha }=\text{arccos}\left(\frac{I_{2\alpha }}{2I_{\alpha }}\right)$$ [3]

where I_α and I_2α are the magnitude images nominal flip angles of α and 2α. Note that any image non-uniformities except for the flip angle variation are cancelled out here as they are identical for both magnitude images. B₁⁺ mapping followed the DCE-MRI acquisition, and repetition time (TR) of 5 seconds was used to ensure complete T₁ relaxation recovery of all tissue. Other imaging parameters were as follows: echo time (TE) = 2.5 ms, acquisition matrix = 64 × 64, number of slices = 49, slice thickness = 4 mm, field-of-view (FOV) = 44 cm, and total scan time = 9 min. Flip angle maps were computed using Eq. [3], and an error due to imperfect 2D slice profile was corrected by simulating the actual slice profile (23). We pre-computed a look-up table that corrects the slice profile errors based on the RF pulses (Hamming windowed sinc shape with different time-bandwidth products). An actual flip angle map can then be normalized by the prescribed flip angle (60°) to compute the relative flip angle variation in %.

Image Analysis

All image analysis both B₁⁺ and T₁ maps was performed on OsiriX, an open source image visualization software application (24). OsiriX supports a complete plug-in architecture and shares all the advanced features with any plug-in developments. We have developed OsiriX plug-ins to easily compute both B₁⁺ and T₁ maps, shown in Fig. 1, and freely available at http://bmr.stanford.edu/. We aligned the orientation and resampled the relative flip angle variation to match the number of slices with the VFA images (using the “Resample” function in OsiriX). T₁ maps were computed using Eq. [2] with and without correcting B₁⁺ variation. We only used central slices for comparison, to avoid effects from an imperfect 3D slab profile. We manually defined circular regions of interest (ROIs) covering medial and lateral portions of both left and right breasts based on magnitude fat-only images. We computed the mean and standard deviation (SD) across all 25 breast MRI patients and also drew box plots for data analysis (25).

Figure 1.

Screenshots of OsiriX plug-ins for (a) B₁⁺ measurements and (b) T₁ measurements. These plug-ins can allow easily computing all quantitative analysis and are freely available.

RESULTS

Figure 2 shows an example of relative flip angle distribution in three orthogonal planes (axial, coronal and sagittal). One of the DAM images (I₆₀) is shown for the anatomical reference, and the coronal and sagittal planes are reformatted from 2D multi-slice images. In this subject, the left breast has an average 13% (± 4.3 %) higher flip angle than the prescribed flip angle, while the right breast has an average 20% (± 5 %) lower flip angle than the prescribed flip angle.

Figure 2.

An example of relative flip angle variation in percentage on a subject at 3T. One of the double angle method images is shown as an anatomical reference, and relative flip angle maps are shown in three orthogonal planes (axial, coronal, and sagittal).

Figure 3a shows the mean and standard deviation of the relative flip angle variation across different slice locations (over a 11.5 cm range of axial slice locations) in a single subject. There exists a somewhat consistent flip angle difference between left and right breasts at all through-plane locations, and in this subject, the overall flip angle difference between the left and right breasts is approximately 35% (solid lines indicate the average relative flip angles of the left and right breasts). Figure 3b shows box plots (median, 25^th and 75^th percentiles, and lower and upper extremes) of the relative flip angle variation in all 25 breast MRI patients. The mean and standard deviation of the relative flip angle variation are 115.4 ± 9.3 (mean ± SD) % on the left breast and 82.4 ± 6.9% on the right breast, which conform to the literature (16).

Figure 3.

(a) Relative flip angle variation across through-plane slices (−65 mm – 50 mm) for one patient. (b) Comparison of flip angle variation in the left and right breasts in 25 breast MRI patients using a box plot. The central mark on each box is the median, the edges of the box are the 25th and 75th percentiles, and the “whiskers” extend to the most extreme data points that were not considered outliers.

Figure 4 shows pre-contrast T₁ values of fat with and without compensation for B₁⁺ inhomogeneity in one subject. One of the fat only images (VFA, FA = 10°) and the relative flip angle map are displayed for the anatomical reference. The T₁ map generated using the prescribed flip angles of 5° and 10° has a substantial T₁ difference between the left and right breasts, while the B₁⁺ compensated map shows more uniform fat T₁ across the whole breast. Average fat T₁ values over ROIs (see red dots in Fig 4) are 458 ms on the left ROI and 227 ms on the right ROI before B₁⁺ correction, and become 387 ms on the left ROI and 323 ms on the right ROI after B₁⁺ correction. The relative flip angle variation is 109% on the left ROI and 81% on the right ROI.

Figure 4.

(a) One of the VFA images and (b) relative flip angle map are shown as an anatomical reference. Comparison of T₁ estimation of fat (c) without and (d) with correcting B₁⁺ inhomogeneity.

Figure 5 shows box plots (median, 25^th and 75^th percentiles, and lower and upper extremes) of T₁ estimation of fat with and without compensation for B₁⁺ variation in all 25 breast MRI patients. The estimated T₁ values of fat without B₁⁺ correction are 497.9 ± 112.1 ms on the left ROI and 239.0 ± 44.4 ms on the right ROI, while the estimated T₁ values of fat with B₁⁺ correction are 374.4 ± 44.8 ms on the left ROI and 346.5 ± 35.1 ms on the right ROI. The T₁ difference between the left and right ROIs is 52% and is reduced to 7% after correcting for the B₁⁺ variation. More importantly, the estimated fat T₁ values with B₁⁺ correction are close to the literature-reported value (T₁ = 366 ms and solid gray lines in Fig 5) (22).

Figure 5.

Comparison of T₁ estimation in fat (a) without and (b) with correction of B₁⁺ inhomogeneity in 25 breast MRI patients. The literature-reported fat T₁ value is shown as a solid gray line.

DISCUSSION

We have measured B₁⁺ inhomogeneity over the breast in 25 breast MRI patients and have shown improved T₁ measurements by accounting for B₁⁺ variation in 3T breast imaging. Average relative flip angle variations were 115% (on the left breast) and 82% (on the right breast), where the overall difference between two breasts was approximately 35% and confirms to the literature (16). This flip angle difference mainly caused a 52% T₁ estimation bias, as measured in fat, between the left and right breasts, and we were able to reduce this estimation error to 7% by including B₁⁺ correction.

We assumed the T₁ relaxation times in fat to be globally uniform and used the estimated fat T₁ fat as a reference to demonstrate a T₁ measurement bias. Corrected T₁ values in fat are 374 ms (on the left breast) and 347 ms (on the right breast), which are close to the literature-reported value, 366 ms (22). There still exists a small residual difference between two breasts, and we believe the residual bias can be further reduced by increasing the number of flip angles in the T₁ mapping (i.e., more than two flip angles).

Two flip angles in VFA T₁ mapping may not be enough to accurately measure T₁ due to its distinctive flip angle variation between two breasts at 3T. A set of flip angles can be optimized by accounting for the typical patterns of B₁⁺ variation in the breast. The overall flip angle difference between the left and right breasts is relatively consistent among the patients, as shown in Fig 3, and we can design two different sets of flip angles (one for the left and the other for the right) by coarsely assuming the expected flip angle variation or can design a set of flip angles to make up for strong B₁⁺ variations within one breast. We expect these approaches can be more useful for longer T₁ values such as fibroglandular tissue T₁, where the T₁ estimation is known to be more sensitive to the flip angle variation.

Multi-channel parallel excitation techniques can be used to reduce the bias in T₁ estimation due to B₁⁺ inhomogeneity. A recent study has shown that dual-source excitation can improve B₁⁺ inhomogeneity over conventional single-source excitation in breast MRI at 3T (26). This parallel excitation technique, however, requires special hardware equipment and can be vendor specific. In addition, the dual-source excitation can correct large first-order B₁⁺ variations but can not solve all spatial variations, which may still cause a residual bias in T₁ estimation.

The effects of the T₁ estimation bias can also be minimized by pursuing other alternative quantification options. New enhancement indices that are insensitive to the variation of pre-contrast T₁ (T₁₀) values and B₁⁺ inhomogeneity have been proposed to quantify the contrast agent uptake in DCE-MRI (27). The study has shown that these new enhancement indices, derived from saturation-recovery snapshot-FLASH (SRSF) images, are considerably less affected by errors caused by variations in the T₁₀ and B₁⁺ inhomogeneity at the cost of longer total scan time than the enhancement ratio in breast DCE-MRI.

Our study has several limiations. Although the degree of the flip angle variation agrees with the previous literature (15,16), the observed patterns of B₁⁺ variation can be vendor and model specific as our study was performed on a 3T GE MR scanner. In addition, we have shown that our B₁⁺ correction scheme can improve the accuracy of T₁ estimation, but this correction scheme can propagate noise to the T₁ estimation, which might make it less useful in low-SNR protocols. Other B₁⁺ mapping methods (28-30) that are fast and can provide an improved angle-to-noise ratio can be considered to reduce the total scan time and error propagation.

In conclusion, we have shown that there exists a noticeable and somewhat systematic B₁⁺ variation between the left and right breasts (average 33%) and have evaluated the accuracy of VFA T₁ mapping with and without compensation for B₁⁺ variation at 3T. The average difference in fat T₁ between breasts was 52% in a total of 25 breast MRI patients, and we reduced this T₁ estimation error to 7% by accounting for B₁⁺ variation, where the fat T₁ values are close to the literature-reported value. This improved T₁ measurement scheme can benefit quantitative breast DCE-MRI at 3T.