Characterizing the imaging environment for supine breast MRI

Abstract

Purpose

Supine breast MRI has the potential to improve over standard prone breast magnetic resonance imaging (MRI) in terms of efficiency and image quality, image alignment with diagnostic and treatment procedures, and overall accessibility. This study aims to characterize potential technical challenges of imaging in the supine position: (i) ${\mathrm{B}}_0$ field inhomogeneities, (ii) ${\mathrm{B}}_1^{+}$ variations, (iii) respiratory‐induced breast motion, and (iv) supine breast geometry.

Methods

Ten healthy subjects were scanned at 3T in both prone and supine positions to quantify and compare (i) and (ii) between both positions, and to assess (iii) in the supine position. Breast image volumes from a wider population (N = 40, healthy volunteers and patients) were analyzed to obtain breast shape metrics to characterize (iv).

Results

${\mathrm{B}}_0$ field inhomogeneity increased from prone positioning (2SD: $122\text{Hz}\pm 25\text{Hz}$) to supine positioning (2SD: $152\text{Hz}\pm 15\text{Hz}$), and ${\mathrm{B}}_1^{+}$ flip angle variations (from prescribed ${30}^{\circ }$) were greater in the supine position (2SD ranging $7^{\circ }$ to ${13}^{\circ }$) than in the prone position (2SD ranging $6^{\circ }$ to $8^{\circ }$). Breast tissue displacement (median [IQR] across all analyzed locations and subjects) was similar along A‐P (1.4 [0.5] mm) and R‐L (1.9 [1.5] mm) directions. Breast geometry varied greatly, with the outer breast perimeter ranging from 34 to 68 cm, and maximum breast tissue thickness ranging from 2 to 9 cm.

Conclusion

Supine positioning for breast MRI may lead to greater ${\mathrm{B}}_0$ inhomogeneities and greater ${\mathrm{B}}_1^{+}$ variations when compared to prone positioning, and breast motion can be substantial. Breast geometry varies greatly among the female population, and shape metrics can inform supine‐dedicated coil development.

1. INTRODUCTION

Breast cancer accounts for one out of three of all new female cancers each year and affects one out of eight women in their lifetime in the United States. ¹ To detect breast cancer early, annual imaging‐based breast screening is recommended to every woman starting at the age of 40, or even earlier if she is classified as at elevated risk ² . Breast magnetic resonance imaging (MRI) has been shown to be the most sensitive imaging modality for screening. ³ , ⁴ , ⁵ Yet, mammography and ultrasound remain the most frequently used modalities due to simplicity, availability, cost, and short exam times.

Currently, the clinical standard for breast MRI is to image with the patient in the prone position (Figure 1A), which poses several drawbacks and limits widespread adoption of breast MRI as the standard‐of‐care modality. First, long setup times, claustrophobia and patient discomfort ⁶ are barriers to utilization as well as compliance with screening guidelines. Second, the breast RF coil reduces the available cross‐sectional area of the bore, preventing MRI in women with large body habitus. Third, poor adaptation of coil coverage to varying breast size, specifically in case of very small breasts, compromises SNR. ⁷ These limit overall specificity as well as sensitivity to small lesions. Lastly, breast tissue deforms substantially from the position of prone MRI to the position of ultrasound biopsy, radiation therapy and surgery, ⁸ , ⁹ which limits the capability for MRI‐based guidance. Deformable registration frameworks to map breast tissue between the prone and supine position have been proposed, ¹⁰ , ¹¹ but have not been translated to clinical practice.

FIGURE 1.

(A) Prone versus (B) supine positioning for breast MRI. In prone imaging, the patient is elevated, with the RF coil occupying a substantial portion of the magnet bore, which limits patient space within the bore, and with most of the patient's torso mass pressing on their sternum. In supine imaging, the patient simply lays on their back, with a flexible RF coil placed on their thorax.

Breast MRI performed with the patient in the supine position ¹² (Figure 1B) has the potential to address each of these drawbacks. Supine breast MRI was first introduced more than two decades ago, primarily to guide interventional procedures, but clinical diagnostic or screening MRI still uses prone positioning as the standard of care. To establish a paradigm shift from prone to supine breast MRI, we would ideally be able to generate image quality that is comparable or superior to prone breast MRI, which relies on addressing potential technical challenges inherent to supine positioning. Recent works propose supine‐specific RF coil hardware ¹³ , ¹⁴ , ¹⁵ and novel approaches to supine‐dedicated imaging sequences, ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ which show promising results toward robust and high‐quality supine breast MRI. RF coil designs using smaller coil elements for increased SNR ²⁰ are considered for signal gain to enable acquisitions with image resolution superior to state‐of‐the‐art prone breast MRI. Smaller elements must be fitted closely to the imaging for efficient use of the coil's penetration depth, which is defined as the distance at which a coil element's sensitivity is reduced to around 37% of that at its center. The penetration depth of a circular loop coil, by rule of thumb, is approximately its radius, so it decreases with smaller elements. ²¹ Fitting a surface coil array close to the breast may in fact be more feasible with supine patient positioning, in which a universal, flexible blanket‐like coil can be wrapped around the rather convex‐shaped body. Conversely, in the conventional prone position, the breast is falling away from the body center, thus forming a more complex and more patient‐to‐patient variable 3D shape. Here, a close fit would require custom‐fitted coils for example as described by Nnewihe et al. ²² or Hancu et al. ²³

In order to inform these technical developments, with this work we seek to characterize the imaging environment specific to supine positioning. Previous studies assessed ${\mathrm{B}}_0$ inhomogeneities  ²⁴ , ²⁵ and ${\mathrm{B}}_1^{+}$ variation  ²⁶ in prone positioning, but it is unknown how these compare when the patient is positioned supine. Further, one expects increased respiratory motion in the supine position, supported by the nature of the patient setup. While the existence of breast tissue motion seems obvious, we seek to better understand the degree of motion and motion pattern characteristics, in order to develop novel motion‐robust sequences or to deploy reconstruction methods that implement motion‐correction, which have improved substantially in recent years. Lastly, breast geometry in the supine position is expected to vary greatly between subjects, which requires novel breast‐dedicated coil development in order to achieve full breast region coverage with sufficient SNR and depth penetration in a wide population of women.

The objective of this work was to characterize the imaging environment of supine breast MRI by assessing ${\mathrm{B}}_0$ field inhomogeneities, regional ${\mathrm{B}}_1^{+}$ variations, respiratory‐induced breast tissue motion, and overall breast geometry in supine positioning. The output of this work is intended to inform the technical development of supine‐dedicated imaging methods to overcome these technical limitations, to ultimately enable robust supine breast MRI and increase accessibility of breast MRI to a broad population.

2. METHODS

2.1. Scan protocol

2.1.1. Prospective volunteer study

We performed breast MRI using a 3T magnet (Signa Premier, GE HealthCare) in ten female subjects in prone and then supine positions (single session per subject). For all sequences, the build‐in body coil was only used in transmit (Tx) mode. In receive (Rx) mode, we used (in the prone position) a breast‐dedicated 16‐channel Rx coil (Sentinelle), or (in the supine position) a prototype 60‐channel blanket Rx coil. ¹⁴ Arms were positioned along the subject's body in both prone and supine position. Each scan was acquired bilaterally, with full breast coverage, in axial orientation, and included the following sequences:

• (i) ${\mathrm{B}}_0$ mapping using a 3D 6‐point iterative decomposition of water and fat with echo assymetry and least squares (IDEAL) acquisition ²⁷ , ²⁸ ($\text{field of view}=(420\times 420){\text{mm}}^2$, $32\text{to}36$, $\text{slice thickness}=6\text{mm}$, $\text{pixel size}=(1.6\times 1.6){\text{mm}}^2$, $\text{TR}=5.8\text{ms}$, $\text{TE}=(0.86;1.56;2.25;2.94;3.64\text{and},4.33)\text{ms}$, $\text{flip angle}=3^{\circ }$). In both prone and supine position, dual 3D cuboid shim boxes were prescribed to perform a linear fit ${\mathrm{B}}_0$ offset correction with active magnetic shimming. In prone position, shim boxes were placed to cover breast tissue, chest wall, and axilla region. In supine position, shim boxes were placed to cover the same regions (Figure 2D). Thus, the measured field inhomogeneities represent residuals after ${\mathrm{B}}_0$ field inhomogeneity correction, which would, for example, affect fat‐saturation performance. However, higher‐order ${\mathrm{B}}_0$ field inhomogeneities were not accounted for.
• (ii) ${\mathrm{B}}_1^{+}$ mapping using the fast spoiled gradient‐recalled echo Bloch–Siegert shift method ²⁹ with quadrature ${\mathrm{B}}_1^{+}$ transmit correction ($\text{field of view}=(420\times 420){\text{mm}}^2$, $32\text{to}36$ slices, $\text{slice thickness}=6\text{mm}$, $\text{pixel size}=(3.3\times 3.3){\text{mm}}^2$, $\text{TR}=14\text{ms}$, $\text{TE}=7.8\text{ms}$, $\text{flip angle}={30}^{\circ }$).
• (iii)In supine positioning only, five sequential 2D plane acquisitions with a high‐frame‐rate two‐dimensional spoiled gradient echo (2D+t GRE) sequence to assess breast motion over multiple respiratory cycles in anterior–posterior and right‐left direction ($\text{field of view}=(360\times 360){\text{mm}}^2$, $\text{slice}\text{thickness}=4\text{mm}$, $\text{pixel size}=(1.4\times 1.4){\text{mm}}^2$, $\text{TR}=3.8\text{ms}$, $\text{TE}=1.2\text{ms}$, $\text{flip angle}=7^{\circ }$, $\text{acceleration}=3\times$, $\text{temporal resolution}=270\text{ms}$, $\text{number temporal frames}=70$). Owing to the five separate, sequential acquisitions, respiratory‐motion induced displacement in the five axial locations is not directly related. Axial planes were equidistantly spaced with 20 mm to 30 mm spacing (breast size dependent) with the central slice (C) prescribed at the superior‐inferior breast center. The five axial planes will be referred to as I2 (most inferior location), I1, C, S1, and S2 (most superior location) in subsequent sections.

FIGURE 2.

Definition of the six breast geometry metrics. (A) Coronal view to define (B) superior and (C) inferior breast tissue boundaries. (D) Central axial slice with dual shim boxes (dashed orange) and in‐plane bounding box (solid blue). (E) Breast mask segmentation ($M_B$) with definition of remaining breast geometry metrics. ${\text{FOV}}_{S-I}$, ${\text{FOV}}_{A-P}$, ${\text{FOV}}_{R-L}$: S‐I, A‐P, and R‐L extent; ${\text{BR}}_{iso}$: distance of breast tissue to isocenter; ${\text{BR}}_{cov}$: outer breast coverage; ${\text{BR}}_{th}$: maximum radial breast tissue thickness.

2.1.2. Image set for breast geometry

To build a geometry atlas of the female breast in the supine position, we considered all supine breast MRI exams (healthy volunteers and patients) performed at our research center between May 2023 and March 2025 (N = 58). All patients were scanned at 3T (GE Signa Premier, GE HealthCare) with either a 21‐channel blanket receive coil (AIR coil MP Large, GE HealthCare), or the prototype 60‐channel blanket receive coil. ¹⁴ For breast geometry segmentation, we included only the exams in which we acquired data that covered the entire breast and thorax. In addition, repeat subjects were excluded. The final atlas comprised T1‐weighted data from 40 female subjects (N = 29 patients with known or suspected new breast cancer, N = 11 healthy volunteers, age (median [range]): 49 [22 to 75] yrs, weight (median [range]): 64 [48 to 101] kg). All imaging exams, including the prospective volunteer study described in the previous section, were conducted with the approval of the Institutional Review Board, and written informed consent was obtained from every subject prior to imaging.

2.2. Data analysis

2.2.1. ${\mathrm{B}}_0$ field inhomogeneities

3D masks of the region of interest (i.e. combined breast and chest wall) were segmented based on IDEAL water images using semi‐automated methods provided by ITK‐SNAP software (v4, build 2023) and tailored in‐house fully‐automated scripts (Python). ${\mathrm{B}}_0$ maps were “centered” by subtracting the average field mapping value within the defined mask to focus on region‐of‐interest field inhomogeneity rather than center frequency offsets. We then analyzed residual field inhomogeneities for each subject using two standard deviations (2SD) of the normalized field map values. Applied linear shim gradient values, as generated by automated dual volume shimming, were also compared between the prone and supine position for each gradient direction. Statistical significance between the prone and supine position with respect to field inhomogeneity and shim gradient strength was assessed using a two‐sided signed‐rank Wilcoxon test.

2.2.2. ${\mathrm{B}}_1^{+}$ variations

3D masks of the region of interest (i.e. combined breast and chest wall) were segmented using the same approach as for ${\mathrm{B}}_0$ data. ${\mathrm{B}}_1^{+}$ maps were analyzed over the combined breast as well as with respect to the separated left and right breast. ${\mathrm{B}}_1^{+}$ variations were studied as flip angle deviation from the prescribed flip angle (${30}^{\circ }$), and reported regarding average and 2SD over the respective region‐of‐interest. Statistical significance between the prone and supine position with respect to flip angle variation was assessed using a two‐sided signed‐rank Wilcoxon test.

2.2.3. Respiratory‐induced motion

2D+t GRE images at five axial locations (I2, I1, C, S1, S2) were post‐processed with a 2D Sobel filter for edge enhancement along the anterior–posterior (A‐P), using horizontal Sobel, or right‐left (R‐L), using vertical Sobel, directions. To obtain a measurement for maximum displacement of breast features along A‐P and R‐L directions ($d_{A-P}$, $d_{R-L}$), we selected four A‐P lines and two R‐L lines to get temporally resolved 1D+t image signals. We then computed cross‐correlations of the first 1D image signal with the 1D image signal from any subsequent temporal frame, and interpreted the signal shift as the relative displacement at that timepoint. The peak‐to‐peak distance of the relative displacement over time data curve was then used to define a measurement for maximum displacement (per analysis line), referred to as peak‐to‐peak displacement in the following. Each 1D signal was upsampled by a factor of 3 using linear interpolation prior to cross‐correlation, yielding pixel spacing of $<0.5\text{mm}$ for displacement estimation. All post‐processing methods were implemented as in‐house scripts using open‐access Python libraries.

2.2.4. Breast geometry

Breast MRI data from 40 subjects with supine positioning were analyzed in order to obtain the following six metrics per subject (Figure 2): (i) superior‐inferior (S‐I), anterior–posterior (A‐P), and right‐left (R‐L) extent (${\text{FOV}}_{S-I}$, ${\text{FOV}}_{A-P}$, ${\text{FOV}}_{R-L}$), to inform overall imaging volume requirements, particularly to define the necessary acquisition matrix for a given image resolution requirement; (ii) distance of breast tissue to ${\mathrm{B}}_0$ isocenter (${\text{BR}}_{iso}$) (median, 5th percentile, 95th percentile), assessed at the 2D central slice, to assess the need for coverage away from isocenter; (iii) outer breast perimeter, assessed at the 2D central slice (${\text{BR}}_{cov}$), to inform coverage requirements for breast‐dedicated coil engineering. (iv) maximum radial breast tissue thickness (${\text{BR}}_{th}$), assessed at the 2D central slice, to inform investigations on signal penetration depth and signal homogeneity, e.g., to inform coil element size.

Datasets were visually inspected to define the most inferior and the most superior axial slices that contain the clinically relevant region‐of‐interest ${\text{FOV}}_{S-I}$ (Figure 2A). The superior end (Figure 2B) was defined at the axilla and the inferior end (Figure 2C) was defined by the most inferior slice containing breast tissue. A central axial slice (Figure 2D) was defined as the slice showing the widest right‐to‐left extent of the breast. This central slice was further processed to obtain all other metrics based on 2D segmentation masks of the full‐thorax and breast‐only‐tissue using semi‐automated methods provided by ITK‐SNAP software (v4, build 2023) and tailored in‐house image processing scripts (Python).

${\text{FOV}}_{A-P}$ and ${\text{FOV}}_{R-L}$ were defined by the bounding box of the central full thorax mask (arms excluded). ${\text{BR}}_{iso}$ was defined as the distance from each (x,y) breast mask pixel coordinate to the ($x_0$, $y_0$) scanner isocenter (i.e. only in‐plane distance to isocenter was considered). The breast mask was processed using a binary opening operation to inflate the mask by $\approx 1\text{cm}$ to mimic the position of a flexible surface coil. ${\text{BR}}_{cov}$ was defined as the length of the connected boundary points of the (inflated) breast mask convex hull. ${\text{BR}}_{th}$ was defined by the Hausdoff distance (HD) between masks $M_A+M_B$ and $M_A$ (Figure 2E). Visual inspection of data for anatomical features and confirmation of semi‐automated segmentation masks were completed by a research scientist ($>3$ yrs experience in breast MRI), advised by a breast radiologist ($>30$ yrs experience).

3. RESULTS

3.1. ${\mathrm{B}}_0$ field inhomogeneities

Figure 3 shows the ${\mathrm{B}}_0$ field inhomogeneity in three representative subjects with varying breast size (A: small, B: medium, and C: large). In both prone and supine positioning, residual field inhomogeneities (after linear shimming) were non‐linear in all of the three main axes (S‐I, A‐P, R‐L).

FIGURE 3.

${\mathrm{B}}_0$ deviation maps for three subjects (A–C) with different breast size in prone (top row) and supine positioning (bottom row). Two axial slices (separation distance $42\text{mm}$) centered around z‐isocenter are displayed per example. In spite of the appearance in the selected slices, there is no remaining shimmable linear variation.

Figure 4 presents the distribution of residual inhomogeneity from all ten studied subjects, assessed as 2SD over the masked breast region‐of‐interest, for prone (Figure 4A) and supine (Figure 4B) positioning. Overall, values for 2SD were $122\text{Hz}\pm 25\text{Hz}$ ($\text{mean}\pm \text{SD}$), ranging from 78 Hz to 178 Hz, for the prone position, and $152\text{Hz}\pm 15\text{Hz}$ ($\text{mean}\pm \text{SD}$), ranging from 127 Hz to 180 Hz, for the supine position, which was statistically significantly different ($p=0.027$). The 2SD increased from prone to supine positioning in nine out of ten subjects (Figure 4C).

FIGURE 4.

Residual ${\mathrm{B}}_0$ deviation (after linear shimming) in ten subjects in (A) prone and (B) supine positioning, with all values standardized to subject‐specific breast mask volume. Stronger ${\mathrm{B}}_0$ field inhomogeneities are reflected by a greater ${\mathrm{B}}_0$ standard deviation (SD) over the breast region of interest. (C) 2SD of ${\mathrm{B}}_0$ field inhomogeneity per subject in prone and supine positioning, showing statistically significant difference between the two positions. (D) Applied shim values in the three shim gradient directions.

Applied linear shim gradients (Figure 4D) were significantly different ($p\le 0.037$) between prone and supine positioning in all three shim gradient directions. Shim gradients ($\text{mean}\pm \text{SD}$ across all subjects) in prone positioning were $(-0.8\pm 2.2;29\pm 6.8;-14\pm 7.6)Hzc{m}^{-1}$ for (x, y, z) gradient directions. Shim gradients in supine positioning were $(-3.3\pm 1.6;-12.8\pm 3.9;-2.6\pm 2.3)Hzc{m}^{-1}$ for (x, y, z) gradient directions.

3.2. ${\mathrm{B}}_1^{+}$ variations

Data from six subjects was included in the ${\mathrm{B}}_1^{+}$ variation analysis, the remaining four were excluded from further analysis because of poor ${\mathrm{B}}_1^{+}$ map quality in either or both positions, typically because of cardiac pulsatility artifacts. We analyzed data with respect to both breasts as well as for the individual left and right breasts (Figures 5, 6). We note that, with all subjects positioned feet‐first, the subject's left breast was at $-\mathrm{x}$ scanner coordinates for prone positioning, but $\mathrm{+}\mathrm{x}$ coordinates for supine positioning; the subject's right breast was at $\mathrm{+}\mathrm{x}$ coordinates for prone positioning, but $-\mathrm{x}$ coordinates for supine positioning. We therefore use the terminology of $-\mathrm{x}$ and $\mathrm{+}\mathrm{x}$ to report breast‐side‐specific results with respect to scanner coordinates.

FIGURE 5.

${\mathrm{B}}_1^{+}$ mapping results. (A, B) Histograms of flip angle deviation (from prescribed) in six subjects for the prone and supine position, stratified by left and right breast. As the subject entered the scanner feet‐first in both positions, the left and right breasts swapped locations with respect to the scanner coordinate system. (C) 2SD of flip angle deviation per subject (left plus right breast region combined), without statistically significant difference between the two positions.

FIGURE 6.

${\mathrm{B}}_1^{+}$ mapping results in three subjects for (A–C) prone and (D–F) supine positioning. As the subject entered the scanner feet‐first for both positions, the left and right breasts swapped locations with respect to the scanner coordinate system.

For the combined left plus right breast, 2SD flip angle deviation increased in five out of six subjects (Figure 5C) in the supine position (2SD ranging $7^{\circ }\text{to}{13}^{\circ }$) when compared to the subject's prone position (2SD ranging $6^{\circ }\text{to}{8}^{\circ }$). But, prone‐to‐supine comparison showed no statistically significant difference ($p=0.093$).

For the individual breasts, $-\mathrm{x}$ mostly showed positive flip angle deviations, whereas $\mathrm{+}\mathrm{x}$ mostly showed negative flip angle deviations (Figure 5A, B). The regional shift in measured flip angle from $-\mathrm{x}$ to $\mathrm{+}\mathrm{x}$ was more pronounced in supine positioning than in prone positioning. The flip angle deviation in supine positioning was $5^{\circ }\pm 3^{\circ }$ at $-\mathrm{x}$ and $-2^{\circ }\pm 2^{\circ }$ at $\mathrm{+}\mathrm{x}$ ($\text{mean}\pm \text{SD}$ across six subjects), corresponding to 16.6% higher than prescribed flip angle at $-\mathrm{x}$, and 7% lower than prescribed flip angle at $\mathrm{+}\mathrm{x}$. The flip angle deviation in prone positioning was $0^{\circ }\pm 2^{\circ }$ at $-\mathrm{x}$ and $-3^{\circ }\pm 1^{\circ }$ at $\mathrm{+}\mathrm{x}$ ($\text{mean}\pm \text{SD}$ for six subjects). Figure 6 shows three representative subjects in the prone (A, B, and C) and supine (D, E, and F) position illustrating the quantitative results described above.

3.3. Respiratory‐induced motion

Breast tissue motion was assessed along 20 A‐P and 10 R‐L lines per subject (five axial locations, with four A‐P and two R‐L analysis lines per location) in all ten subjects. Analysis lines were defined as shown in Figure 7C. Overall, the median [IQR] peak‐to‐peak displacement across all subjects and analysis lines was 1.4 [0.5] mm for A‐P, and 1.9 [1.5] mm for R‐L (Figure 7A). The maximum measured peak‐to‐peak displacement was $6.1\text{mm}$.

FIGURE 7.

(A) Distribution of peak‐to‐peak displacement across all analysis lines and all subjects, separated into A‐P ($d_{A-P}$, blue) and R‐L ($d_{R-L}$, red) direction. Median [IQR] are shown in legend. (B) A‐P versus R‐L peak‐to‐peak displacement ($d_{A-P}/d_{R-L}$) at the five axial locations. Boxplots show median (bold line), interquartile range (IQR, box), extent from the box to the farthest data point lying within $1.5\times$ of the IQR (whiskers), and outliers (dots). (C) Location of A‐P and R‐L 1D analysis lines.

Figure 7B analyzes $d_{A-P}/d_{R-L}$ as a function of axial location across all ten subjects. By axial location, median $d_{A-P}/d_{R-L}$ was $<1$ for the most inferior (I2) and most superior (S2) location, and $\ge 1$ at locations I1, C, and S1.

Figure 8 shows the measured relative displacement data curves over time in one representative subject at the five axial locations, showing substantial tissue displacement of up to $d_{A-P}=3.3\text{mm}$ (A‐P), and $d_{R-L}=5.2\text{mm}$ (R‐L). Results show that periodic motion is mostly in‐phase between A‐P and R‐L directions, but peak‐to‐peak displacement along the two directions differs, depending on axial slice location. The A‐P versus R‐L ratio of peak‐to‐peak displacement ($d_{A-P}/d_{R-L}$) also differed between axial locations. This subject showed more motion along the A‐P direction in the two most inferior locations (I1, I2), and more motion along the R‐L direction in the three most superior slices (C, S1, S2). For this subject, the largest peak‐to‐peak displacement occurred at location S2 in the R‐L direction (Figure 8E).

FIGURE 8.

Motion analysis in one representative subject with substantial breast motion. (A‐E) At five axial locations, plots display calculated A‐P and R‐L relative displacement over multiple respiratory cycles, based on cross‐correlation of temporally resolved 1D signals. Data shows median (dots) and range (vertical bars) across the four A‐P (blue) and two R‐L (red) analysis lines, respectively. Values for $d_{A-P}$ and $d_{R-L}$ are measurements of peak‐to‐peak displacement in A‐P and R‐L direction, respectively. $d_{A-P}/d_{R-L}$ (bold‐faced text) defines the ratio between these two metrics, with value $>1$ indicating more motion along A‐P, and value $<1$ indicating more motion along R‐L.

3.4. Breast geometry

Figure 9 displays all central slice breast segmentation masks, and visualizes measurements for coverage ${\text{BR}}_{cov}$ (color‐coded perimeter) and breast thickness ${\text{BR}}_{th}$ (solid white line) in each subject. Figure 10 shows distributions across subjects of the six breast geometry metrics.

FIGURE 9.

Breast region segmentation masks at central S‐I location in 40 subjects, sorted by their ${\text{BR}}_{cov}$. For each subject, the maximum radial breast thickness (${\text{BR}}_{th}$) is shown (solid white line). The color‐coded line defines outer the breast perimeter with length ${\text{BR}}_{cov}$, and absolute ${\text{BR}}_{cov}$ values (in mm) are attached to each breast mask.

FIGURE 10.

Distribution of quantitative results for the six metrics of breast geometry in the supine position in 40 subjects. (A) overall extent of field of view (${\text{FOV}}_{S-I}$, ${\text{FOV}}_{A-P}$, and ${\text{FOV}}_{R-L}$); (B) breast tissue to iso‐center distance at central slice (${\text{BR}}_{iso}$) (median, 5th, and 95th percentile); (C) outer breast perimeter at central slice (${\text{BR}}_{cov}$); (D) maximum breast thickness at central slice (${\text{BR}}_{th}$).

The following results are given as $\text{mean}\pm \text{SD}$ across 40 subjects. The FOV extent to cover the subject's full axial throrax cross section and full breast region of interest was $148\text{mm}\pm 18\text{mm}$ (${\text{FOV}}_{S-I}$), $217\text{mm}\pm 28\text{mm}$ (${\text{FOV}}_{A-P}$), and $350\text{mm}\pm 41\text{mm}$ (${\text{FOV}}_{R-L}$). The median ${\text{BR}}_{iso}$, 5th percentile ${\text{BR}}_{iso}$, and 95th percentile ${\text{BR}}_{iso}$ (distance between breast pixels and isocenter) was $124\text{mm}\pm 17\text{mm}$, $72\text{mm}\pm 19\text{mm}$, and $167\text{mm}\pm 21\text{mm}$, respectively. ${\text{BR}}_{th}$ (maximum skin to chest wall breast thickness at central slice) was $48\text{mm}\pm 16\text{mm}$. ${\text{BR}}_{cov}$ (outer breast region perimeter at central slice) was $513\text{mm}\pm 71\text{mm}$.

4. DISCUSSION

This study characterized the imaging environment of supine breast MRI by assessing ${\mathrm{B}}_0$, ${\mathrm{B}}_1^{+}$, respiratory‐induced displacement of breast tissue, and breast geometry. Overall, results underline several patient‐specific technical challenges that should be considered to enable reliable high‐resolution, robust breast MRI in the supine position.

${\mathrm{B}}_0$ field maps in both prone and supine positioning presented substantial residual field inhomogeneities that were not correctable with linear shimming, indicating that higher‐order frequency offsets are present. Results obtained in the supine position showed greater residual field offsets, specifically in the lateral breast, that is at coordinates further away from isocenter. Results suggest an average 95% distribution width (i.e. $\pm$ 2SD) of 310 Hz. In contrast, in the prone position, this 95% distribution width decreased to 244 Hz. If one defines an acceptable 95% distribution width of 250 Hz (for example to achieve uniform fat‐saturation at 3T), the data presented here suggest that in none of the scanned supine‐positioned subjects were field inhomogeneities tolerable, whereas ${\mathrm{B}}_0$ inhomogeneities in the prone position were tolerable in six out of ten subjects. As uniform fat suppression is usually considered very important in breast MRI, this demonstrates a major technical challenge that needs to be addressed. Dixon fat‐water separation techniques may present an alternative solution in case of increased ${\mathrm{B}}_0$ inhomogeneity in supine. Further, acquisitions that specifically rely on ${\mathrm{B}}_0$ homogeneity, for example EPI diffusion‐weighted acquisitions, are likely to be more adversely affected in the supine position. Jordan et al. ²⁴ reported an average 80% distribution width of 233 Hz for prone breast MRI (after linear offset correction), which indicates a similar inhomogeneity to ${\mathrm{B}}_0$ mapping data in the prone position from the present study.

${\mathrm{B}}_0$ mapping results further showed that significantly stronger shims were applied in prone positioning when compared to supine positioning, both in the A‐P (y‐axis) direction as well as in the S‐I (z‐axis) direction. For the prone position, Han et al. ³⁰ reported values for shim gradients ($21.4\text{to}50.4\text{Hz}{\text{cm}}^{-1}$ along the y‐axis) that are in accordance with our results. The strong shimming difference between prone and supine positioning may be explained by the differences of the breast and overall thorax geometry, with the geometry being more irregular in the prone position. However, measured residual field inhomogeneities suggest that the linear shim approach is more effective in the prone position. We note that, for prone positioning, the prescribed 3D cuboid shim boxes mostly encompassed breast region, whereas, for supine positioning, shim boxes included a substantial portion of non‐breast region (lungs, liver, heart, as shown in Figure 2D). This possibility to closely fit the shim boxes around the actual breast region may explain the higher effectiveness of linear shimming for prone positioning. For the prone position, work by Hancu et al. ³¹ demonstrated effects of shim box placement with best results achieved when boxes were tightly fitted around the breast and chest wall region. To show the potential impact of 3D shim box prescription on linear shimming and residual ${\mathrm{B}}_0$ field inhomogeneities, we included ${\mathrm{B}}_0$ mapping data in a ‘supine’ phantom in Supporting Data 1. These exploratory results indicate moderate influences (2SD: 132 Hz to 153 Hz) on linear shimming performance, depending on how shim volumes are defined. We chose to use linear shimming rather than higher‐order shimming in this study, as higher‐order shims are not fully embedded in every clinical MRI system, and thus many times not used in the clinical routine. For example, the system used in our study requires running an additional sequence and plugin with tedious manual interaction. To investigate the best‐performing shimming approach for supine breast MRI, future phantom and in vivo technical studies should assess different linear and higher‐order shimming approaches, including shim box prescription.

${\mathrm{B}}_1^{+}$ mapping results for supine positioning suggest a higher than prescribed flip angle in the breast region positioned at $-\mathrm{x}$, and a lower than prescribed flip angle in the breast region positioned at $\mathrm{+}\mathrm{x}$. ${\mathrm{B}}_1^{+}$ data in the prone position followed this trend in most subjects, however, the difference between $-\mathrm{x}$ and $\mathrm{+}\mathrm{x}$ were not as large as in the supine position. ${\mathrm{B}}_1^{+}$ mapping results from other prone breast MRI studies ²⁶ , ³² , ³³ are in accordance, and reported the same trend in left to right flip angle variations. ${\mathrm{B}}_1^{+}$ variations become of particular importance when dynamic‐contrast‐enhanced T1‐weighted multi‐phase imaging data is used to obtain relative enhancement rates in lesions. ³² The observed ${\mathrm{B}}_1^{+}$ variation in supine identify a possible need to apply appropriate ${\mathrm{B}}_1^{+}$ correction or improved ${\mathrm{B}}_1^{+}$ transmit methods whenever homogeneous RF excitation is crucial.

While the prototype 60‐channel Rx coil was not directly tested for ${\mathrm{B}}_1^{+}$ distortion, the coil electronics and cables are almost identical to the commercially available AIR coils (GE HealthCare), which must not exceed maximum ${\mathrm{B}}_1^{+}$ distortion limits. We expect that the few differences in component values, element construction, and layout should not impact the passing results seen with the AIR coils as the prototype coil was designed such that the decoupling impedance meets the manufacturer's specifications of AIR coils. However, ${\mathrm{B}}_1^{+}$ distortion testing in prone and supine position with only the build‐in Tx/Rx body coil could be the subject of future investigations to fully demonstrate that ${\mathrm{B}}_1^{+}$ distortion is primarily affected by body shape and positioning. Ideally, this experiment should use a custom phantom that could morph to mimic natural prone and supine breast positions, and the Rx coil should not be present inside the Tx/Rx body coil.

Motion analysis in the supine position showed that respiratory‐induced breast tissue displacement is subject‐specific and on the order of several millimeters. This is important, given that sub‐millimeter pixel spacing of $\approx (0.5\times 0.5){\text{mm}}^2$ is targeted for high‐resolution breast MRI. There are a multitude of options that may address this, including retrospective motion‐correction, for example as proposed by Isaieva et al., ¹⁸ and/or motion‐robust acquisitions with center‐out non‐cartesian trajectories, for example radial XD‐GRASP ³⁴ or 3D Cones. ³⁵ , ³⁶ None of the subjects examined in the present work were instructed to breathe in a specific fashion, and the variations in the measured displacement indicate both belly‐breathing and chest‐breathing types. Results also emphasized that breast tissue displacement is present in both A‐P and R‐L dimensions. This should be considered, for example, when setting up sequences with phase‐encode along the dimension with seemingly lesser motion for the sake of mitigating motion‐induced artifacts (e.g., multi‐shot EPI for diffusion‐weighted breast MRI). Assessing breast tissue motion in the prone position was not within the scope of this study. Previous work by Clauser et al. ³⁷ presented the influence of patient motion between temporal phases in prone breast DCE MRI (spatial resolution: ($1.1\times 0.9\times 3){\text{mm}}^3$) on diagnostic estimates in 334 histologically confirmed breast lesions. Their results suggest a decrease in sensitivity but unchanged specificity and accuracy with increased motion artifacts. The study also reports that most of the data showed no (54%) or only mild (37%) motion artifacts in the DCE MRI data. Our quantitative assessment of motion in the supine position suggests substantial tissue displacement over the respiratory cycle, which could cause image blurring in addition to misregistration between temporal phases. Future investigations with a large patient cohort are warranted to discuss how potential motion‐related artifacts in supine breast MRI translate into impeding diagnostic estimates.

Substantial variations in breast geometry, as visualized in Figure 9 and supported by the six metrics (Figure 10), demonstrate a key challenge in RF coil design for supine breast MRI. Flexible surface array coils with extent to cover a wide range of breast sizes will be necessary. Further, to enable fast high‐resolution imaging, designs with high channel counts to enable parallel imaging will be beneficial. Trade‐offs need to be made regarding signal (with smaller elements) versus depth penetration (with larger elements). ${\text{BR}}_{th}$ measurements presented here provide valuable information about the distribution of breast tissue thickness. Obermann et al. ¹³ developed the so‐called ‘BraCoil’ with $(55\times 25){\text{cm}}^2$ array size and $8\text{cm}$ single‐element diameter, which would provide sufficient geometric surface coverage in 70% of our 40‐subject population. Though, the sensitive imaging area of a surface coil array likely exceeds the geometric coverage. On average, breast tissue was located as far as 16.7 cm (95th percentile of breast pixels) away from isocenter, which requires ${\mathrm{B}}_0$ shimming over a large region, and may likely contribute to the difficulties of achieving acceptable ${\mathrm{B}}_0$ homogeneity. Though not reported in the present study, this distance is expected to be considerably less for prone positioning, as the deformable breast tissue falls along the A‐P direction, and not away from isocenter.

For future supine breast MRI coil development, and based on results presented here, we would like to suggest few considerations to take into account: (i) Extending the sensitive area in R‐L direction rather than S‐I direction while considering overall measurements of breast geometry. (ii) Performing computational studies with patient‐specific sensitivity simulations of to inform decision making on optimal coil element size that is sufficient for a wide population with varying breast tissue thickness. (iii) Two cup‐shaped regions, as detailed by Vincent et al., ¹⁴ seem inefficacious, as breast tissue greatly varies in volume, often spreading across the entire cross‐sectional arch. Rather, a uniform element layout is more preferable for fitting different breast sizes and shapes.

The authors note several limitations of the present study. First, ${\mathrm{B}}_0$ and ${\mathrm{B}}_1^{+}$ field inhomogeneity analyses were done in only ten and six subjects, respectively, whose breast geometry did not fully represent the breadth of variation as seen in the 40‐subject dataset. The 10‐subject dataset may not reveal field inhomogeneity challenges that may occur in very large patients (e.g., bottom right case in Figure 9). Second, ${\mathrm{B}}_0$ mapping presents averaged free‐breathing results, so results do not present respiration‐induced effects. Bolan et al. ²⁵ studied effects of respiration‐induced ${\mathrm{B}}_0$ variations in the context of breast magnetic resonance spectroscopy (MRS) in the prone position, and showed substantial quantification errors of MRS if no correction was performed. Third, 2D+t GRE acquisitions for motion analysis were done sequentially and without using a respiratory signal to trigger the acquisition start to a specific timepoint of the respiratory cycle. Data from the five axial locations are therefore not ‘phase‐aligned’, so axial slices were instead analyzed separately. No information on inter‐slice displacement could be retrieved from the acquired data. Lastly, despite including 40 subjects in the breast geometry variations dataset, this dataset still may not represent the supraregional female population. Future studies should consider analyzing breast geometry based on supine data from a multi‐site patient cohort. Also, non‐MRI (e.g., computed tomography) data of the female thorax in the supine position could be exploited in a retrospective analysis.

5. CONCLUSION

In the present study, supine positioning for breast MRI led to greater ${\mathrm{B}}_0$ inhomogeneities and greater ${\mathrm{B}}_1^{+}$ variations compared with prone positioning. Breast tissue displacement was on the order of several mm, which translates to several pixels given that sub‐millimeter pixel spacing of $\approx (0.5\times 0.5){\text{mm}}^2$ is the target for high‐resolution supine breast MRI, specifically for cancer screening. Further, results underscored that breast geometry and size varies substantially across the female population, with the outer breast perimeter ranging from 34 to 68 cm across the 40 subjects included here. This characterization study should benefit future work in supine breast MRI, specifically breast‐dedicated coil development, motion correction, and correction strategies to mitigate effects arising from ${\mathrm{B}}_0$ and ${\mathrm{B}}_1^{+}$ field imperfections.

FUNDING INFORMATION

This work was supported by National Institute of Health R01‐EB009055 and R01‐CA249893 and GE HealthCare.

Supporting information

Supporting Data 1. Effect of dual 3D shim box placement on linear shims and residual B0 field inhomogeneities in a phantom. (A) The phantom consisted of a “posterior” rectangular ballistics gel block and and an “anterior” flexible ballistics gel pad with two refrigerant gel inlets (labelled with asterisk), with overall dimensions similar to the imaging region in supine breast MRI. The phantom spans 28 cm in S‐I direction (not shown in image). (B) Histogram of residual B0 field deviations after linear shimming for four different ways of placing shim boxes (“conventional”, “full”, “tight”, “tilted”). (C–F) B0 deviation maps with applied [x, y, z] shims (top) and definition of shim boxes (bottom). The same axial slice at z‐isocenter is shown for all four cases. “Conventional” (C) corresponds to how shim boxes were placed in our main study.

Zimmermann J., Vincent J., Robb F., Daniel B. L., Hargreaves B. A., and Moran C. J., “Characterizing the imaging environment for supine breast MRI,” Magnetic Resonance in Medicine 95, no. 1 (2026): 125–137, 10.1002/mrm.70029.