High-Quality Breast MRI

R. Edward Hendrick, PhD

KEYWORDS Breast cancer detection Contrast agent Spatial resolution Temporal resolution Technical quality

KEY POINTS High-quality breast magnetic resonance imaging (MRI) requires the competing factors of high spatial resolution, good temporal resolution, high signal-to-noise ratios (SNRs), and complete bilateral breast coverage. High-quality breast MRI requires a modern MR scanner with a magnetic field strength of 1.0 T or higher, good magnetic field homogeneity, a bilateral breast coil with prone positioning, strong magnetic gradients with short rise times, and good fat suppression over both breasts. The key pulse sequence for high-quality breast MRI is a multiphase 3D gradient-echo sequence performed bilaterally with submillimeter in-plane spatial resolution, thin slices, and temporal resolution of 1 to 3 minutes to correctly capture the morphology and time-enhancement pattern of enhancing breast lesions.

INTRODUCTION One of the limitations of contrast-enhanced breast MRI has been the lack of standardized im- Early breast magnetic resonance imaging (MRI) aging protocols, contrast agent administration, im- studies conducted in the early to mid-1980s at- age postprocessing, and image review. A major tempted to distinguish malignant breast lesions step forward has been standardization of breast from benign lesions and normal breast tissues 1,2 MRI reporting terminology through publication of based on inherent tissue T1 and T2 values. Ma- the fourth edition of the American College of Radi- lignant breast lesions were found to have higher T1 ology’s (ACR’s) Breast Imaging Reporting and and T2 values than normal breast tissues, but Data System (BI-RADS), which included reporting shorter T1 and T2 values than most benign breast 3–5 terminology for breast MRI and breast ultrasound, lesions such as fibroadenomas. Significant as well as mammography.17 Another recent step overlap in both T1 and T2 values between benign forward has been the initiation of the ACR’s Breast and malignant breast lesions, however, discour- MRI Accreditation Program, which is described aged the use of noncontrast breast MRI for cancer briefly at the end of this article. detection and diagnosis. It is now an accepted This article provides specific recommendations standard that high sensitivity to breast cancer re- for achieving high-quality breast MRI. Because of quires contrast-enhanced breast MRI both without different MRI hardware, software, and scanning and with a gadolinium (Gd)-based paramagnetic 6–10 capabilities, it is not possible to achieve complete contrast agent to identify enhancing lesions. uniformity of protocols, but fairly specific guide- Most recent studies of contrast-enhanced breast lines for equipment requirements and scanning MRI have reported sensitivities to breast cancer protocols are given for performing bilateral between 90% and 100%, depending on the sub- contrast-enhanced breast MRI with high spatial ject cohort, imaging techniques, including other resolution, good temporal resolution, and high imaging tests performed along with breast MRI, 11–16 signal-to-noise ratio (SNR). This article describes and criteria for breast cancer. the technical parameters needed to achieve

Department of Radiology, School of Medicine, University of Colorado–Denver, Anschutz Medical Campus, 12700 E. 19th Avenue, MS C-278, Aurora, CO 80045, USA E-mail address: [email protected]

Radiol Clin N Am 52 (2014) 547–562 http://dx.doi.org/10.1016/j.rcl.2013.12.002

consistently high-quality contrast-enhanced Kuhl and colleagues21 compared contrast- breast MRI. enhanced breast MRI at 1.5 T and 3.0 T in the same group of 37 patients. Overall image quality scores were slightly higher and differential diag- EQUIPMENT REQUIREMENTS nosis of enhancing lesions was made with greater MRI systems used for breast cancer detection and confidence at 3.0 T, as shown by larger areas un- diagnosis should include (1) adequate magnetic der the receiver operating characteristic curve.21 field strength with good magnetic field homogene- The investigators pointed out, however, that ity across both breasts, (2) adequate magnetic technical problems exist at 3.0 T beyond those field gradients to permit fast gradient-echo imag- observed at 1.5 T.22 These included increased ing, (3) a bilateral breast coil enabling prone posi- nonuniformity of transmitted B1 RF waves, partic- tioning, and (4) good fat suppression over both ularly between left and right breasts with the larger breasts. field-of-view (FOV) used for transaxial scanning. These requirements are discussed individually. This in turn led to reduced enhancement of lesions 22 located in “low B1 areas.” The investigators used a 2-dimensional (2D) gradient-echo pulse Adequate Magnetic Field Strength and sequence, pointing out that the adverse effects Homogeneity of low B1 areas on lesion enhancement should The magnetic field strength of whole-body MRI be reduced with the 3D (volume) sequences systems ranges from 0.064 T to 8 T (1 T 5 more commonly used in the United States 10,000 G; the naturally occurring magnetic field because of the shorter repetition times (TRs) at the earth’s surface ranges from 0.25 to used in 3D imaging. Others have pointed out that 0.65 G). MRI systems approved by the Food and B1 nonuniformities can be reduced by using 3D Drug Administration (FDA) for clinical use have techniques, careful choice of flip angle to match magnetic field strengths up to 3.0 T. Above a few the TR of the imaging sequence, smaller FOV tenths of a Tesla, image SNRs per voxel go up (eg, sagittal rather than transaxial acquisitions), approximately linearly with magnetic field and optimized shimming of the acquisition strength, if receiver coil design, voxel size, and volume.23 imaging parameters other than field strength Another reason for performing breast MRI at remain constant.18 Thus, higher magnetic field magnetic field strengths of 1.0 T or greater, strength (B0) should provide higher SNR per voxel beyond higher SNR, is to ensure higher static for breast imaging for the same pulse sequence, magnetic field homogeneity over the entire imaged although the linear increase in SNR with field volume. High magnetic field homogeneity for strength is moderated somewhat by the increase breast imaging means that the static magnetic in tissue T1 values at higher field strengths. T1 field strength (B0) should remain nearly constant values increase by about 20%, going from 1.5 T across both breasts, including the chest wall and to 3.0 T.19 axillae. Because hydrogen nuclei in water and fat Most breast MRI is done on 1.5-T scanners, with differ in resonant frequencies by 3.4 parts per a few sites performing breast MRI at 1.0 to 1.2 T million (ppm), the magnetic field homogeneity and a growing number of sites performing breast must be significantly less than 3.4 ppm to achieve MRI at 3.0 T. Although SNR per voxel is nearly good chemically selective fat suppression of doubled for the same pulse sequence at 3.0 T, hydrogen signals from fat, while preserving compared with 1.5 T, 3.0 T systems have some hydrogen signals from water. additional technical challenges.20 It is more diffi- The standard criterion to ensure that chemically cult to get uniform fat suppression on 3.0-T sys- selective fat suppression is effective is that the tems than on 1.5-T systems (Fig. 1). In addition, magnetic field strength should vary by less than artifacts are often more pronounced at 3.0 T than 1 ppm over the entire volume of tissue being at 1.5 T. Most importantly, higher-frequency radio imaged. At 1.5 T, a nonuniformity of 1 ppm would waves used for tissue excitation are more highly amount to a magnetic field difference of 1.5 mT (mi- attenuated. Because 3.0-T systems have double croTesla), or a resonant frequency difference of the resonant frequency of 1.5-T systems, the 63.9 Hz, compared with the water-fat frequency penetration of radiofrequency (RF) waves trans- difference of 224 Hz (3.4 ppm). The static magnetic mitted to excite breast tissues, the B1-field, is field should be homogeneous to this level across a less uniformly distributed within breast tissues at FOV 30 to 35 cm in diameter encompassing both 3.0 T because of greater absorption by external breasts. This is generally not possible for low-field tissues. This causes nonuniformity of signal exci- to midfield scanners (less than 1.0 T), and is a chal- tation and, thus, nonuniform measured signals. lenge even for high-field scanners, as the location High-Quality Breast MRI 549

Fig. 1. Same slice of a fat-suppressed T1-weighted transaxial scan of volunteer imaged on a 1.5-T scanner (A) and 3.0-T scanner (B). All acquisition parameters other than magnetic field strength were matched between the 2 scanners. Note the more uniform fat-suppression in (A) and higher SNR in (B). Aliasing artifacts (signal wrap) of the volunteer’s arms from the opposite sides (arrows) are visible in (A). of the breasts in prone-positioned breast MRI typi- coils, the received signals are recorded simulta- cally is not at the isocenter of the magnet. Instead, neously using multiple amplifier and analog-to- in most magnets, the breasts are below isocenter digital converters. Multiple-channel receiver to allow prone positioning of the patient, with elements require a scanner capable of simulta- breasts in the breast coil, and to allow adequate neously recording multiple channels of data, so it space for the patient’s torso in the magnet bore. is important to make sure that scanner hardware and software can accommodate the number of channels in the breast coil. A Bilateral Breast Coil with Prone Positioning Multiple receiver channels were developed Bilateral imaging is recommended for the following to boost coverage and signal uniformity, but ac- reasons: (1) Clinical comparison of both breasts is quired a single dataset for image reconstruction. as important in breast MRI as it is in mammog- A technique developed over the past decade, par- raphy. Bilateral comparison helps identification of allel imaging, modifies data acquisition so that focal enhancement and helps prevent overcalling different channels or sets of channels simulta- of physiologic enhancement, which tends to occur neously acquire different datasets simulta- bilaterally, especially in premenopausal women neously.32,33 In parallel imaging, each coil and postmenopausal women on hormone replace- element (or set of coil elements) acquires different ment therapy.24,25 (2) Data from recent breast MRI pieces of the image simultaneously, or in parallel; studies indicate that when a breast cancer occurs, then, more complex image reconstruction tech- there is a 3% to 5% chance that breast MRI will niques are used to recombine the different partial detect a mammographically occult cancer in the datasets into planar images or volumes. Parallel contralateral breast.26–31 (3) Unilateral imaging in imaging also requires a short prescan to map out the transaxial or coronal plane can incur image the sensitivity profile of each coil element (or set wrap (or aliasing) artifacts from the contralateral of coil elements) on each patient. This is done in breast, especially if phase encoding is set left-to- a separate acquisition on some scanners and right (as it typically is in transaxial imaging), a bilat- within the parallel imaging pulse sequence itself eral breast coil is used, and the field-of-view is on other scanners. Some parallel imaging tech- narrowed to include only the breast being imaged. niques acquire multiple channels of data in phys- Bilateral breast imaging is typically performed us- ical space, whereas others acquire multiple ing the body coil as the RF-transmit coil and a channels of data in spatial frequency (or k-) prone-positioned bilateral breast coil as the space.32,33 Parallel imaging speeds image acquisi- RF-receiver coil. A few systems, such as the Aurora tion by a prespecified factor (eg, 2, 3, or 4), but re- breast MRI system, have bilateral breast coils quires longer for image reconstruction after all serving as RF-transmit-receive coils. Modern data have been acquired. Parallel imaging typically breast coils, whether receive-only or transmit- is done with an acceleration factor of 2, which cuts receive coils, have multiple-channel elements. the acquisition time nearly in half. Use of higher- Bilateral breast coils currently have between acceleration factors in breast imaging tends to 2 channels (1 channel for each breast) and 18 chan- cause image reconstruction artifacts (Fig. 2) and nels (9 channels for each breast). In multichannel has been avoided in most clinical practices. 550 Hendrick

Fig. 2. The same slice of sets of 3D gradient-echo sequences acquired on the same volunteer using a 3.0-T scanner without (A) and with (B–D) parallel imaging using acceleration factors (AF) of 2 to 4. All were acquired with the same FOV, matrix, and slice thickness. Total acquisition time for each 3D gradient-echo acquisition covering both breasts was 102 seconds without parallel imaging (A), 57 seconds with AF 5 2(B), 42 seconds with AF 5 3(C), and 34 seconds with AF 5 4(D). Note the presence of parallel imaging reconstruction artifacts in (C) and (D)(arrows). As a result of more artifacts with higher AF values, most sites performing parallel imaging in breast MRI use an AF of 2.

Adequate Magnetic Field Gradients of x, y, or z location. The knocking noise heard from MR units as they scan is due to magnetic field Magnetic field gradients are produced by addi- gradients being turned on and off. tional coils that generate magnetic fields (each Two parameters characterize the performance pointing along the static magnetic field, B ) that 0 of magnetic field gradients: (1) Maximum gradient intentionally vary the magnetic field strength line- strength, expressed in milliTesla per meter (mT/m), arly along each of the 3 perpendicular axes: x, y, plays a role in determining how small voxels can and z (Fig. 3). Gradient fields are switched on be made. Modern MR scanners have magnetic and off rapidly during each repetition of the pulse field gradient strengths of up to 50 mT/m. sequence to spatially resolve the source of signal (2) Gradient rise times describe the time interval by briefly altering the precessional frequencies of needed for a magnetic gradient to go from zero hydrogen nuclei at different locations as a function to maximum strength, which in turn determines

Fig. 3. The x, y, and z gradients shown within the bore of a solenoidal magnet. The main static magnetic field, B0, points along the z direction, which is also the direction of the magnetic field of each gradient; x, y, or z gradients, when applied, alter the strength of the magnetic field pointing along the z-axis as a function of the x, y, or z direction, respectively. High-Quality Breast MRI 551 how quickly pulse sequences can be performed. fat suppression makes it more difficult to separate The shorter the gradient rise time, the shorter TR enhancing breast lesions from fat, because fat and echo time (TE) can be made in 2D or 3D and enhancing breast lesions have similar signal gradient-echo imaging. Modern MR scanners intensities. In subtracted images (postcontrast im- have gradient rise times as short as 200 microsec- ages minus precontrast images), even a small onds, yielding TR values as short as 4 ms and TE amount of motion or misregistration between pre- values as short as 1 ms. Generally, to achieve contrast and postcontrast scans causes structured adequate spatial resolution and short enough im- noise artifacts that complicate interpretation and, in aging times in 3D gradient-echo imaging, TR some cases, simulate enhancing lesions. Good fat needs to be shorter than 10 ms and TE needs to suppression in both precontrast and postcontrast be less than 4 ms. In gradient-echo sequences images minimizes the structured noise of misregis- without or with incomplete fat suppression, TE tration artifacts in subtracted images, allowing should be carefully chosen to minimize chemical- detection of smaller enhancing lesions or non- shift artifacts.33 masslike lesions with greater reliability.

Good Fat Suppression Over Both Breasts PULSE SEQUENCE REQUIREMENTS In 2D or 3D MRI, fat suppression is typically Beyond good equipment, high-quality breast MRI achieved by applying a frequency-selective 90 requires the use of appropriate pulse sequences. saturation pulse that acts only on the hydrogen Breast MRI pulse sequences should include nuclei in fat (Fig. 4). In 2D imaging, this saturation several noncontrast series, performed before pulse is applied to each slice at the start of each contrast administration, along with a multiphase pulse sequence repetition; in 3D (volume) imaging, series of pulse sequences applied just before this saturation pulse is applied to the entire volume and several times after contrast agent administra- of tissue within the RF-transmit coil. If applied tion. Recommended pulse sequences include uniformly across both breasts, the fat suppression (Fig. 6): pulse effectively eliminates fat signal from signal measured during the pulse sequence repetition 1. Scout images obtained in transaxial, sagittal, that follows. Fig. 5 shows examples of T1- and coronal planes. weighted bilateral breast MRI without fat sup- 2. A T1-weighted non–fat-saturated series ob- pression, with good fat suppression, and with tained bilaterally, including axillae and chest incomplete fat suppression. Fat suppression is use- wall, to distinguish fat from water-based tissues ful in contrast-enhanced breast MRI because it re- including fibroglandular tissue, cysts, lymph duces the signal from fat in both precontrast and nodes, and other benign lesions, muscle, and postcontrast scans. In postcontrast scans, lack of cancers.

Fig. 4. Schematic of the resonant frequency difference between hydrogen nuclei in fat and hydrogen nuclei in water at 1.5 T. The MR scanner’s center frequency is tuned to the resonant frequency of hydrogen nuclei in water. When fat-saturation is selected, a saturation pulse is applied with a narrow frequency range to cancel the signal from hydrogen nuclei in fat molecules, which resonate at about 220 Hz (1 Hz 5 1 cycle per second) lower frequency than the hydrogen nuclei in water at 1.5 T. At 3.0 T, the frequency shift between hydrogen fat and water peaks doubles to about 440 Hz. 552 Hendrick

Fig. 5. T1-weighted breast MRI: (A) sagittal plane image without fat suppression; (B) sagittal plane image with uniform chemically selective fat suppression; (C) transaxial image with incomplete fat suppression. Fat saturation was adequate distally, but failed near the chest wall and on the lateral side shown on the left.

3. A T2-weighted fat-saturated series obtained as possible, and using a flip angle that is bilaterally to distinguish cysts from solid le- relatively small and based on the TR value to sions. A STIR (short inversion time [TI or tau] optimize SNR.33 inversion recovery) series can be used in place of a T2-weighted series (Fig. 7) if TI is set Any modern MR scanner should be able to correctly to minimize the signal from fat. A TI deliver the first 3 pulse sequences without diffi- of about 180 ms at 1.5 T, or about 215 ms at culty. Scout images acquired in all 3 perpendicular 3.0 T, should do a good job of suppressing fat planes are routine and should take less than 1 min- signal.33 Good fat-suppression is important in ute to acquire and display. Both T1-weighted non– either T2-weighted or STIR images, so that fat-saturated and T2-weighted fat-saturated the brightest tissues in the image are fluid- series can be obtained using accelerated spin- filled cysts or blood vessels. echo sequences, called fast spin-echo (FSE) or 4. A multiphase 3D Fourier transform (3DFT or vol- turbo spin-echo (TSE) sequences, in a time of ume acquisition) gradient-echo T1-weighted less than 3 to 4 minutes for each series. If parallel pulse sequence acquired once before and mul- imaging with an acceleration factor of 2 can be tiple times after contrast agent administration, applied to these noncontrast series, scan times preferably with chemically selective fat- can be decreased by nearly a factor of 2, to about suppression, is used to identify the vascular 2 minutes per series. bed and any enhancing lesions in the breast. The key pulse sequence for breast cancer detec- T1-weighting is achieved by setting the pulse tion and lesion characterization is the multiphase sequence TR short relative to the T1-values of 3D gradient-echo T1-weighted series acquired tissues being imaged, setting the TE as short before and several times after MR contrast-agent High-Quality Breast MRI 553

Fig. 6. (A) Scout images obtained on a woman with double-lumen implants. Scout images were obtained in the transaxial, coronal, and sagittal planes through both breasts. The second sagittal scout shows slice locations for subsequent transaxial T1-weighted and T2-weighted acquisitions. (B) One slice of 42 non–fat-saturated T1-weighted images using a 2D FSE sequence. (C) The same single slice of 42 fat-suppressed T2-weighted STIR images. Note on both T1-weighted and T2-weighted series a cyst anterior to the right implant with signal similar to that in the inner (saline) portion of the double-lumen implant. The cyst is dark on T1-weighted images and bright on T2-weighted or STIR images. (D) A single slice through the left breast in a bilateral multislice precontrast T1-weighted 3D gradient- echo image set. (E) The same slice from the first postcontrast series, acquired using the same bilateral multislice 3D gradient-echo series as in the precontrast set. On core biopsy, the enhancing lesion just anterior to the left implant was found to be a 4-mm Grade 3 invasive ductal carcinoma. (F) Subtraction of the precontrast slice shown in (D) from the postcontrast slice shown in (E), making more apparent the rim effect of contrast uptake in the enhancing lesion. (G) MIP reconstructed from the entire set of subtracted images of the left breast. The dataset is projected in a direction rotated nearly 180 from the primary sagittal acquisitions to better display the enhancing invasive ductal carcinoma. Note the absence of signal in the implant area of both subtracted and MIP images, since no contrast uptake occurred in that region. Note also that rim enhancement of the lesion, an important indicator of malignancy, is more clearly seen in postcontrast and subtracted series than in MIP images. administration. Stronger gradients permitting very The important features of a contrast-enhanced short TR and TE values, along with multichannel multiphase T1-weighted series are as follows: coils and scanner software permitting parallel imaging, have sped multiphase acquisitions, allowing 1. Consistent Gd-chelate contrast agent admin- improved spatial resolution by using a higher matrix istration based on patient mass or weight: (that is, more phase-encoding and frequency- 0.1 mmol/kg, followed by a 20-mL saline flush. encoding steps), while maintaining adequate 2. Bilateral acquisition with prone positioning. SNRs per voxel. This has enabled breast MRI to 3. A multiphase 3D gradient-echo T1-weighted meet all of the spatial resolution and temporal pulse sequence (1 precontrast and multiple resolution goals listed below when proper pulse postcontrast series extending at least 6 minutes sequence techniques are used. after contrast injection). 554 Hendrick

Fig. 6. (continued)

4. Adequately thin slices of 3 mm or less. Gd-chelate Contrast Agent Administration: 5. Pixel sizes of less than 1 mm in each in-plane 0.1 mmol/kg Followed by 20 mL of Saline direction. Although MR contrast agents are not labeled spe- 6. Phase-encoding direction chosen to minimize cifically for breast cancer detection, use of an artifacts across the breasts. appropriate contrast agent is essential for high 7. Total acquisition time for each series (or “phase” sensitivity to breast cancer. There are 6 MR of the multiphase series) of 1 to 3 minutes. contrast agents that are FDA-approved for use in 8. Adequate SNR to visualize small enhancing the brain and spine (Table 1) and suitable for vessels on 3D maximum intensity projection breast MRI. All are labeled for a recommended (MIP) images. dose of 0.1 mmol/kg of patient body mass. Each of these items is described in more detail All but one (Gadavist, Bayer Healthcare Pharma- in the following sections. ceuticals Inc, Wayne, NJ, USA) are packaged in High-Quality Breast MRI 555

Fig. 7. The same slice of 44 transaxially acquired slices using (A) T2-weighted FSE (TR 5 4350 ms, TE 5 111 ms, echo-train length [ETL] 5 15) and (B) fast STIR images (TR 5 6080, TE 5 59 ms, TI 5 180 ms, ETL 5 13). Tissue contrast is similar in both scans. Note the bright cyst near the chest wall of the left breast in both. Fat- suppression typically is more uniform in STIR images than in T2-weighted images, because in STIR images fat- suppression is based on the TI setting. Fat-suppression in T2-weighted images depends on achieving uniform chemically selective fat-suppression over the entire FOV, which in turn depends on the uniformity of the static magnetic field B0 and the uniformity of transmitted RF excitations. a concentration of 0.5 mmol/mL; therefore, in be administered for a given body mass to achieve terms of packaged contrast agent volume, the rec- a dose of 0.1 mmol/kg of body mass. A simple rule ommended dose is 0.2 mL per kg of body mass. with Gadavist is to administer 1 mL (or 1 cc) of For example, a 140-pound woman has a body agent for every 22 lb of body weight. mass of 64 kg (140 lb/2.2 lb/kg 5 64 kg) and her Contrast agent should be administered with a administered dose of MR contrast agent pack- controlled flow rate (most sites use a rate of aged at a concentration of 0.5 mmol/mL should 2 mL per second) followed immediately by a bolus be 13 mL (64 kg * 0.2 mL/kg 5 12.8 mL) rounded of 20 mL of saline administered at a similar rate. to the nearest milliliter. A simple rule to follow to This is best done with an dual-headed MR- administer label-recommended doses of 0.1 mL/ compatible power injector that can administer kg of body mass for agents packaged at both contrast agent and saline flush sequentially 0.5 mmol/mL is to inject 1 mL (or 1 cubic centi- at controlled flow rates. meter, cc) of contrast agent for every 11 pounds of body weight. Using the previous example, a Bilateral Acquisition with Prone Positioning 140-lb woman should receive 140 lb * (1 mL/11 lb) x 13 mL of Gd-chelate contrast agent. Prone positioning in a dedicated bilateral breast coil Gadavist is packaged at a higher concentration positions the breasts pendently and reduces breast of 1.0 mmol/mL, so half a much Gadavist should motion due to respiration and cardiac pulsation.

Table 1 Gd-chelated contrast agents approved for central nervous system indications in the United States (and used for breast cancer detection)

Molecular Molarity, Viscosity, Relaxivity Agent Common Name Weight mol/L cP, 37 C a1, L/(mmol$s) Magnevist (Gd-DTPA) Gadopentetate 938 0.5 2.9 4.1–4.9 dimeglumine Prohance (Gd-HP-DO3 A) Gadoteridol 559 0.5 1.3 4.1–5.4 Omniscan (Gd-DTPA-BMA) Gadodiamide 574 0.5 1.4 4.3–5.4 Optimark (Gd-DTPA-BMEA) Gadoversetamide 662 0.5 2.0 4.7 Multihance (Gd-BOPTA) Gadobenate 1058 0.5 5.3 6.7–9.7 dimeglumine Gadavist Gadobutrol 605 1.0 5.0 5.2

Viscosities are measured in centipoises (cP) at 37 C (viscosity of water is 1.002 cP); Relaxivity a1 is the relaxation rate (the À1 À1 inverse of relaxation time, T1) per unit concentration of agent and is expressed in mmol/L) s or L/(mmol$s). Magnevist (Bayer Healthcare Pharmaceuticals Inc, Wayne, NJ, USA); Prohance (Bracco SpA, Milan, Italy); Omniscan (GE Healthcare, Princeton, NJ, USA); Optimark (Mallinckrodt Inc, St. Louis, MO, USA); Multihance (Bracco SpA, Milan, Italy). Data from manufacturer’s labeling and Rohrer M, Bauer H, Mintorovitch J, et.al. Comparison of magnetic properties of MRI contrast media solutions at different magnetic field strengths. Invest Radiol 2005;40(11):715–24. 556 Hendrick

Because the patient is supported by the coil at the excited volume of tissue, including both breasts, sternum, lateral chest, and above and below the instead of from just a single plane, at each signal breasts, most respiratory and cardiac motion measurement. The 3DFT sequences require more affects chest tissues posterior to the breasts. Any phase-encoding steps (by the factor Nslices)to motion between precontrast and postcontrast resolve not just a plane of tissue, but an entire scans or during scanning causes misregistration volume of tissue, into individual voxels. in subtracted breast images. By positioning the patient comfortably and by properly instructing Adequately Thin Slices of 3 mm or Less the patient before the multiphase T1-weighted se- Slice thickness sets the limit on the smallest lesion ries (rather than during the series, such as just that can be imaged without slice partial volume ef- before administration of contrast agent), the MR fects decreasing lesion contrast. Although slice technologist can help minimize breast and patient thickness may not impair visualization of high- motion. Keeping total scan time reasonably short contrast lesions that enhance dramatically, it can (20 minutes or less) will also help decrease patient play an important role in the detection of low- discomfort and motion during scanning. contrast lesions. To image a low-contrast lesion of a given diameter without partial volume effects, A 3D Fourier Transform Gradient-Echo T1- which would decrease its contrast relative to sur- Weighted Pulse Sequence rounding tissues, a slice thickness of half the lesion’s diameter or less should be used. For T1-weighted pulse sequences are used in example, to be sensitive to a low-contrast 5-mm contrast-enhanced breast MRI because Gd- enhancing lesion, a slice thickness of 2.5 mm or chelates, while shortening both T1 and T2, cause less should be used (Fig. 8). Thin slices are partic- a greater fractional change in T1 than T2 (or ularly important for minimizing partial volume ef- 34 T2*). In gradient-echo imaging, T1 weighting is fects on diffuse, non-masslike enhancing lesions, achieved by using a short TR, very short TE, and such as those sometimes associated with ductal a relatively low flip angle that is matched to the carcinoma in-situ (DCIS) (Fig. 9).33 TR.33 For 3D Fourier transform (3DFT) imaging, extremely short TR values are used to keep the Pixel Sizes of Less than 1 mm in Each In-Plane scan times for each phase of the multiphase series Direction reasonably short, ideally 3 minutes or less. Although 2DFT pulse sequences acquire image Pixel sizes smaller than 1 mm in each in-plane data from a single plane at a time, 3DFT pulse se- direction can be achieved by selecting an acquisi- quences acquire image data from an entire volume tion matrix (number of phase-encoding and at a time. Multislice 2DFT imaging typically has frequency-encoding steps) that exceeds the FOV small gaps between individual slices, with (in mm) in both the phase-encoding and Gaussian slice profiles. The 3DFT imaging ac- frequency-encoding direction. For example, in Â Â quires contiguous slices within the 3D volume, transaxial imaging with a 30 30 cm (300 Â with rectangular slice profiles, so that no signal 300 mm) FOV, an acquisition matrix of 300 300 Â gaps occur between slices. or greater should be used. If a 384 384 matrix For 3DFT imaging, total acquisition time is were used for this FOV, each in-plane pixel would be (300 mm)/384 5 0.78 mm in each direction, Ttotal 5 (TR)(Npe)(Nacq)(Nslices), where TR is the basic which would give excellent spatial resolution. Sub- pulse sequence repetition time, Npe is the number of in-plane phase-encoding steps to resolve signal millimeter in-plane pixels are important for good lesion margin visualization, which helps distinguish in a single in-plane direction, Nacq is the number of times each phase encoding step is repeated benign from malignant enhancing lesions based on their morphology.35 (usually set to 1 in 3DFT imaging), and Nslices is the number of slices, which equals the number of Phase-Encoding Direction Chosen to Minimize phase-encoding steps used to separate the 3D vol- Artifacts Across the Breasts ume in the third (slice-select) direction (in 2DFT imaging, Nslices is automatically set to 1). It is because A primary cause of image artifacts (structured of this additional factor, Nslices, which can be as high noise in MR images), is patient motion, including as 160 with slices comparable in thickness to the cardiac and respiratory motion. These motion arti- in-plane pixel size (isotropic voxels), that gradient- facts propagate across the image in the in-plane echo sequences with very short TR values are phase-encoding direction, regardless of the direc- needed in 3DFT imaging. The 3DFT sequences tion of motion in the patient.33 Therefore, it is have a signal-to-noise advantage over 2DFT se- essential to orient the in-plane phase-encoding di- quences because signal is acquired from the entire rection to minimize artifacts across the breast. For High-Quality Breast MRI 557

Fig. 8. Relationship between slice thickness and minimum lesion size seen without, or with minimal, partial volume effects. (A) Slice thickness equals lesion size. In this case, partial volume effects will significantly decrease lesion conspicuity. (B) Slice thickness equals one-half lesion size. Regardless of alignment of lesion and slices, in this case at least one slice will display the lesion without, or with minimal, partial volume effects. sagittal plane acquisitions, the phase-encoding obscure a minimal amount of breast tissue direction should be head-to-foot (or superior- (Fig. 11). For coronal plane acquisitions, phase en- inferior) (Fig. 10). For transaxial plane acquisitions, coding can be either left-right or head-to-foot, as phase-encoding should be oriented left-to-right to cardiac and respiratory motion will not propagate ensure that cardiac and respiratory motion across the breasts in either in-plane direction.

Fig. 9. Surveillance breast MRI in a 48-year-old woman revealed a subtle non-masslike enhancement in a linear/ ductal, clumped pattern (arrows) in sagittally acquired and axially reconstructed images. Core biopsy demonstrated the enhancing lesion to be intermediate-grade cribriform and solid DCIS. Note that the enhancing lesion appears less sharp in the reconstructed axial image than in the originally acquired sagittal image because the slice thickness of acquired sagittal images exceeded the in-plane pixel size. The thicker slices in sagittally acquired images lowered spatial resolution in the slice-select direction, which is left-to-right in the reformatted axial images. (Courtesy of Robyn Birdwell, MD, Breast Imaging, Brigham and Women’s Hospital, Boston, MA.) 558 Hendrick

Fig. 10. Sagittal plane imaging with the phase-encoding direction: (A), correctly selected in the head-to-foot (or superior-inferior) direction, and (B), incorrectly selected in the anterior-posterior direction. Note in (B) that motion and wrap artifacts are propagated across breast tissue.

Phase-encoding is usually chosen to be head-to- k-space is being acquired), which occurs at one- foot for coronal imaging because it requires less third to one-half of the total pulse sequence acqui- spatial coverage than left-to-right, minimizing the sition time, depending on the manufacturer (eg, number of phase-encoding steps needed. 3DFT sequences on Siemens acquire the center of k-space about one-third of the way into the full acquisition, on GE and Philips at half-way Total 3DFT Acquisition Time for Each Series of through sequence acquisition). The goal is to 1 to 3 Minutes select imaging parameters that place the The temporal resolution required for breast MRI is maximum contrast-weighting of the first postcon- determined by the time course of contrast agent trast series at or near the time of peak contrast uptake. Peak contrast enhancement in malignant agent uptake. Assuming that peak enhancement lesions typically occurs between 60 and 120 sec- of breast lesions occurs 90 seconds after contrast onds after injection.36 It is important to capture injection, you would like the center of k-space of contrast uptake at or near its maximum with one the first postcontrast series to occur 90 seconds of the postcontrast series. To do that, it is also after the end of contrast injection. If you were using important to know that 3DFT acquisitions have a Siemens 3DFT series with a 2-minute series scan maximum contrast-weighting at the low spatial time, peak contrast-weighting would occur at one- frequency acquisitions (ie, when the center of third of 2 minutes, or 40 seconds, into the series,

Fig. 11. Transaxial plane T2-weighted FSE images identically acquired, except with the phase-encoding direction: (A) correctly selected in the left-to-right direction, and (B) incorrectly selected in the anterior-posterior direction. In (B), cardiac and respiratory motion artifacts propagate across breast tissue. High-Quality Breast MRI 559 so to properly time the center of k-space 90 sec- lesions into 3 categories: continuous uptake onds after injection, you would wait 50 seconds af- (Type 1), plateau (Type 2), or washout (Type 3) ter injection to begin the first postcontrast series. If (Fig. 12). Kuhl and colleagues37 demonstrated in you were using a GE or Philips scanner with, for a study of 266 enhancing lesions (101 breast can- example, a 3-minute 3DFT acquisition, peak cers) that only 6% of lesions with Type 1 curves contrast weighting would occur at the midpoint were malignant, 65% of lesions with Type 2 curves of the series, so you would begin scanning imme- were malignant, and 87% of lesions with Type diately after injection to place maximum contrast- 3 curves were malignant. Characterizing lesions weighting at 90 seconds. by both their morphology, which does not require Typically, a precontrast series and several multiple postcontrast time points, and their time- postcontrast series are acquired with identical enhancement curve shape, which does, adds acquisition parameters so that subtractions of pre- specificity to contrast-enhanced breast MRI. contrast from postcontrast images reveal only Thus, it is important to collect data with adequate temporal changes. Thus, all precontrast and post- temporal resolution, and adequate duration, to contrast series should be identical. A single accurately capture the time course of lesion precontrast scan should be acquired, followed enhancement. immediately by contrast agent injection. A pause Initially, European breast MRI protocols empha- of scanning during, and perhaps after, contrast in- sized the need for high temporal resolution, on the jection might be needed, depending on the calcu- order of 1 minute per series, to gain specificity.8 A lation outlined previously to place peak contrast at subsequent article by Kuhl and colleagues35 the center of k-space of the first postcontrast se- demonstrated that temporal resolution could be ries. Then, several postcontrast series should be relaxed to approximately 2 minutes without sacri- acquired without pauses or delays between ficing specificity, especially if that added time them, extending so that the last measurement was used to improve spatial resolution to submilli- samples the center of k-space at least 6 minutes meter in-plane pixels. More recent work by Gutier- after the end of contrast injection. This is done rez and colleagues38 indicated that 3-minute so that the detailed shape of the time- temporal resolution was adequate to correctly enhancement curve can be determined for any characterize time-enhancement curve shapes significantly enhancing lesions. Contrast agent up- when the center of k-space was properly posi- take is best characterized by dividing enhancing tioned at approximately 90 seconds after contrast

Fig. 12. Three time-enhancement curve types typical of enhancing breast lesions. Lesions with Type 1 curves have continuous uptake and have the lowest probability of malignancy. Lesions with Type 2 curves enhance by at least 80% to 100% from their noncontrast signal values and then demonstrate a plateau behavior, not gaining or losing signal appreciably from their peak value. Lesions with Type 2 curves have a moderate (40%–70%) suspicion of malignancy. Lesions with Type 3 curves have rapid uptake of contrast within 3 minutes of administration, then washout, and have a high (60%–80%) suspicion of malignancy. Images producing these curves were acquired every 60 seconds (1-minute temporal resolution), with the first postcontrast series acquired 90 seconds after contrast administration. 560 Hendrick

injection. In addition, the longer acquisition time (MIP) images is a good surrogate for the ability to per series of 3-minute acquisitions captured visualize small enhancing lesions on subtracted greater peak signal than 90-second acquisitions. or MIP images. Failure to see relatively small blood Based on their findings, it appears that 3-minute vessels on subtracted or MIP images gives the temporal resolution is adequate to add specificity radiologist little confidence that small or subtle by correctly characterizing time-enhancement enhancing lesions would be detected, if present. curve shapes. Going faster than 1 minute per se- Fig. 13 provides examples of good, SNR- ries fails to add additional information about curve deficient, and SNR-starved MIP images: good, shape and decreases SNR per series.33,39 Other marginal, and poor-quality breast MR images. studies, such as Schnall and colleagues,39 have shown that, although curve shape and degree of THE ACR BREAST MRI ACCREDITATION lesion enhancement are important, lesion PROGRAM morphology assessment is an even more important factor in the overall accuracy of breast MRI The ACR Breast MRI Accreditation Program for cancer detection. (BMRAP) began accrediting facilities that perform breast MRI on May 10, 2010. As of May 1, 2013, 1264 facilities have been accredited, with 72 facil- Adequate SNR to Visualize Small Enhancing ities under review. The repeat rate for facilities is Vessels on 3D Maximum Intensity Projection 20%. Like other ACR accreditation programs, the Images BMRAP includes requirements for personnel The ability to visualize small enhancing vessels (2– (radiologists, MRI technologists, and medical 3 mm in diameter) on maximum intensity projection physicists), equipment, quality assurance, and

Fig. 13. Sagittal MIP images demonstrating varying degrees of quality in terms of displaying enhancing vessels (and lesions, if they were present): (A) good MIP image, where small vessels are clearly displayed; (B) marginal MIP image, where visibility of small vessels is limited due to low SNR; and (C) poor MIP image, due to extremely low SNR and no display of large or small vessels. The poor image quality in (C) gives low confidence that this scan would demonstrate a small or diffusely enhancing lesion, if present. High-Quality Breast MRI 561 accreditation testing based on submission of clin- 8. Kaiser WA, Zeitler E. MR imaging of the breast: fast ical images to assess breast MRI scanning proto- imaging sequences with and without Gd- DTPA. Pre- cols and clinical image quality. A complete liminary observations. Radiology 1989;170:681–6. discussion of the BMRAP is beyond the scope of 9. Harms SE, Flamig DP, Hensley KL, et al. MR imaging this article, but is available at http://www.acr.org/ of the breast with rotating delivery of excitation off- Quality-Safety/Accreditation/BreastMRI, including resonance: clinical experience with pathologic a complete list of BMRAP requirements, a Breast correlation. Radiology 1993;187:493–501. MRI Clinical Image Quality Guide, and complete 10. Padhani AR. Contrast agent dynamics in breast MRI. procedures for applying for ACR Breast MRI In: Warren R, Coulthard A, editors. Breast MRI in prac- Accreditation. tice. London: Martin Dunitz; 2002. p. 43–52. 11. Kriege M, Brekelmans CT, Boetes C, et al. Efficacy SUMMARY of MRI and mammography for breast-cancer screening in women with a familial or genetic predis- Current MRI systems are capable of meeting the position. N Engl J Med 2004;351:427–37. stringent technical requirements of performing 12. Kuhl CK, Schrading S, Leutner CC, et al. Mammog- multiphase T1-weighted contrast-enhanced scan- raphy, breast ultrasound, and magnetic resonance ning with high in-plane spatial resolution (1mm imaging for surveillance of women at high familial pixel sizes), thin slices (3 mm thick), adequate risk for breast cancer. J Clin Oncol 2005;23: temporal resolution (1–3 minutes), bilateral breast 8469–76. coverage, and adequate SNR to detect small or 13. Tilanus-Linthorst MM, Obdeijn IM, Bartels KC. MAR- diffusely enhancing breast lesions. Careful atten- IBS study. Lancet 2005;366:291–2. tion to breast MRI equipment selection and breast 14. Stoutjesdijk MJ, Boetes C, Jager GJ, et al. Magnetic MRI protocols is required to achieve all of these re- resonance imaging and mammography in women quirements simultaneously. The ACR’s BMRAP with a hereditary risk of breast cancer. J Natl Cancer provides a peer-review system for validating that Inst 2001;93:1095–102. breast MRI personnel, equipment, quality-control 15. Warner E, Plewes DB, Hill KA, et al. Surveillance of procedures, scanning protocols, and image quality BRCA1 and BRCA2 mutation carriers with mag- are adequate to perform high-quality breast MRI. netic resonance imaging, ultrasound, mammography, and clinical breast examination. JAMA REFERENCES 2004;292:1317–25. 16. Morris EA, Liberman L, Ballon DJ, et al. MRI of 1. Ross RJ, Thompson JS, Kim K, et al. Nuclear mag- occult breast carcinoma in a high-risk population. netic resonance imaging and evaluation of human AJR Am J Roentgenol 2003;181:619–26. breast tissue: preliminary clinical trials. Radiology 17. Ikeda DM, Hylton NM, Kuhl CK, et al. BI-RADS: 1982;143:195–205. Magnetic resonance imaging. In: D’Orsi CJ, 2. El Yousef SJ, Alfidi RJ, Duchesnau RH, et al. Initial Mendelson EB, Ikeda DM, et al, editors. Breast im- experience with nuclear magnetic resonance aging reporting and data system: ACR BI-RADS – (NMR) imaging of the human breast. J Comput Breast imaging atlas. 1st edition. Reston (VA): Assist Tomogr 1983;7:215–8. American College of Radiology; 2003. 3. McSweeney MB, Small WC, Cerny V, et al. Magnetic 18. Edelstein WA, Glover GH, Hardy CJ, et al. The resonance imaging in the diagnosis of breast dis- intrinsic signal-to-noise ratio in NMR imaging. ease: use of transverse relaxation times. Radiology Magn Reson Med 1986;3:604–18. 1984;153:741–4. 19. Rakow-Penner R, Daniel B, Yu H, et al. Relaxation 4. Wiener JI, Chako AC, Merten CW, et al. Breast and times of breast tissue at 1.5Tand 3T measured using axillary tissue MR imaging: correlations of signal in- IDEAL. J Magn Reson Imaging 2006;23:87–91. tensities and relaxation times with pathologic find- 20. Barth MM, Smith MP, Pedrosa I, et al. Body MR im- ings. Radiology 1986;160:299–305. aging at 3.0 T: understanding the opportunities 5. Merchant TE, Thelissen GR, de Graaf PW, et al. and challenges. Radiographics 2007;27:1445–62. Application of a mixed imaging sequence for MR im- 21. Kuhl CK, Jost P, Morakkabati N, et al. Contrast- aging characterization of human breast disease. enhanced MR imaging of the breast at 3.0 and 1.5 Acta Radiol 1993;34:356–61. T in the same patients: initial experience. Radiology 6. Heywang SH, Hahn D, Schmid H, et al. MR imaging 2006;239:666–76. of the breast using gadolinium-DTPA. J Comput 22. Kuhl CK, Kooijman H. Effect of B1 inhomogeneity on Assist Tomogr 1986;10:199–204. breast MR imaging at 3.0T [letter to the editor]. Radi- 7. Heywang SH, Wolf A, Pruss E, et al. MR imaging of ology 2007;244:929–30. the breast with Gd-DTPA—preliminary observations. 23. Mountford CE, Stanwell P, Ramadan S. Breast MRI Radiology 1989;71:95–103. at 3.0 T. Radiology 2008;248:319–20. 562 Hendrick

24. Friedman PD, Swaminathan PV, Herman K, et al. The recently diagnosed breast cancer. N Engl J Med importance of bilateral imaging. Am J Roentgenol 2007;356:1295–303. 2006;187:345–9. 32. Glockner JF, Houchun HH, Stanley DW, et al. Parallel 25. Kuhl CK, Kreft BP, Bieling HB, et al. Dynamic breast imaging: a user’s guide. Radiographics 2005;25: MRI in premenopausal healthy volunteers: normal 1279–97. values of contrast enhancement and cycle phase 33. Hendrick RE. Breast MRI: fundamentals and tech- dependency. Radiology 1997;203:137–44. nical aspects. New York: Springer; 2008. p. 156–62. 26. Liberman L, Morris EA, Kim CM, et al. MR imaging 34. Hendrick RE, Haacke EM. Basic physics of MR findings in the contralateral breast of women with contrast agents and maximization of image contrast. recently diagnosed breast cancer. Am J Roentgenol J Magn Reson Imaging 1993;3:137–48. 2003;180:333–41. 35. Kuhl CK, Schild HH, Morakkabati N. Dynamic bilat- 27. Lee SG, Orel SG, Woo IJ, et al. MR imaging eral contrast-enhanced MR imaging of the breast: screening of the contralateral breast in patients trade-off between spatial and temporal resolution. with newly diagnosed breast cancer: preliminary re- Radiology 2005;236:789–800. sults. Radiology 2003;226:773–8. 36. Kuhl C. The current status of breast MR imaging, 28. Lehman CD, Blume JD, Thickman D, et al. Added part I. Choice of technique, image interpretation, cancer yield of MRI in screening the contralateral diagnostic accuracy, and transfer to clinical prac- breast of women recently diagnosed with breast tice. Radiology 2007;244:356–78. cancer: results from the International Breast Mag- 37. Kuhl CK, Mielcarek P, Klaschik S, et al. Dynamic netic Resonance Consortium (IBMC) trial. J Surg breast MR imaging: are signal intensity time course Oncol 2005;92:9–15 [discussion 15–6]. data useful for differential diagnosis of enhancing 29. Slanetz PJ, Edmister WB, Yeh ED, et al. Occult lesions? Radiology 1999;211:101–10. contralateral breast carcinoma incidentally detected 38. Gutierrez RL, Strigel RM, Partridge SC, et al. Dy- by breast magnetic resonance imaging. Breast J namic breast MRI: does lower temporal resolution 2002;8:145–8. negatively affect clinical kinetic analysis? AJR Am 30. Viehweg P, Rotter K, Laniado M, et al. MR imaging of J Roentgenol 2012;199:703–8. the contralateral breast in patients after breast- 39. Schnall MD, Blume J, Bluemke DA, et al. Diag- conserving therapy. Eur Radiol 2004;14:402–8. nostic architectural and dynamic features at breast 31. Lehman CD, Gatsonis C, Kuhl CK, et al. MRI evalu- MR imaging: multicenter study. Radiology 2006; ation of the contralateral breast in women with 238:42–53.