Introduction

During pregnancy and lactation the mammary gland undergoes a series of morphological and physiological changes that are regulated by key hormones and growth factors [1, 2]. The breast functional task during lactation, synthesizing and secreting milk, begins during pregnancy in the second trimester at the lactogenesis I stage. Within days of postpartum begins the actual secretion of milk in the hormonally regulated lactogenesis II stage, triggered by alteration in progesterone blood level [3]. The volume and composition of the secreted milk changes dramatically from the protein and electrolytes enriched colostrum in the first days of lactation, to mature stable mother milk with increased lipid content and decreased protein, sodium and chloride concentrations [4].

The returning of the breast to its original pre-conception state is also a gradual multi-step process. The mammary gland is experiencing involution characterized by a high degree of epithelial apoptosis, growth of the mammary adipose tissue and tissue re-modelling which shares similarities with inflammation characteristics [5, 6]. Ultimately, the composition of the breast revolves, inducing withdrawal of the lactation-typical prominent ductal tree which is replaced by connective tissue and fat [4].

Imaging of the lactating breast allows a particular view of the breast ductal tree architecture. Back in 1840, Sir Astley Cooper used injected colored wax to lactating breast cadavers in order to expose the underlying human breast anatomy [7]. Currently attempts are made to use ultra-sonography (US) in order to quantify the amount and appearance of the main glandular ducts in-vivo [8], and to reconstruct the complete three-dimensional ductal system structure, using manual tracing of the milk-laden ducts [9]. In these US studies it was possible to track milk laden ducts of diameters less than 1 mm, but it was not possible to delineate all the breast ducts with a lower diameter. Recently it was also demonstrated that intra-ductal injection of contrast agents allows non-invasive high resolution MRI of the 3 dimensional ductal architecture in the mouse mammary gland [10].

The development of Magnetic Resonance Imaging (MRI) led to the advancement of breast MRI in the 1980s [11, 12]. A unique characteristic of MRI is the ability to produce multiple contrasts between different tissues depending on the various water hydrogen nuclei properties, such as hydrogen nuclei density and relaxation rates (longitudinal T1 and transversal T2), as well as interaction with externally injected contrast agents and intrinsic contrast due to water diffusion resulting from random microscopic translational - motion (Brownian - motion). The water diffusion measurements in tissues are affected by hindrance and restriction caused by complex cellular and structural elements. The diffusion measurements may also be influenced by flow processes such as the intra-voxel incoherent motion (IVIM) in disordered capillary systems [13].

The drastic hormonal-induced physiological and structural changes during pregnancy and lactation create a challenge in the imaging of pregnancy-associated breast cancer (PABC) [14, 15] and references cited therein. Currently, US constitutes the modality of choice for evaluating breast disorders during pregnancy and lactation since the sensitivity of mammography in these women is low as a result of the high breast parenchymal density [15].

The application of the standard MRI technique for breast cancer detection, using injected Gadolinium-based contrast-agents [16,17,18] and references cited therein, is restricted in pregnant patients since the contrast agents can pass through the placental barrier and enter fetal circulation [19]. Consequently, it has been recommended to carefully assess the risk–benefit ratio of dynamic contrast enhanced (DCE) - MRI in each pregnant patient [19]. The use of breast DCE-MRI in lactating patients appears to be safe although a small amount is reaching the milk [14, 20]. Yet, DCE-MRI during lactation is limited because the normal lactating parenchyma shows fast and elevated contrast enhancement that mimics the appearance in malignancy [14, 21,22,23].

More recently, it has been shown that adding diffusion weighted imaging (DWI) to breast DCE-MRI improves the differentiation between cancers, benign lesions and normal breast tissue [24, 25]. DWI has been also applied by Sah et al. [26] to measure the Apparent Diffusion Coefficient (ADC) of normal lactating breast tissue as compared to breast cancer in non-lactating patients, showing a potential to aid in differentiating lactation from malignant transformation.

ADC values obtained by DWI reflect differences in tissue cell density [27, 28], however, it does not reflect tissue micro-structural features. In contrast, magnetic resonance Diffusion Tensor Imaging (DTI) extends the ability of diffusion measurements to investigate microstructural features in tissues [29]. The diffusion tensor components in the frame of the breast tissue enable modeling the diffusion in each voxel to an ellipsoid shape that characterizes the local structure in this voxel [29,30,31]. The direction of the three ellipsoid orthogonal axes represents the three tensor eigenvectors (ν1, ν2, ν3), and the size of the ellipsoid in these three directions indicates the value of the tensor eigenvalues, defining three directional diffusion coefficients, λ1, λ2, λ3, arranged from high to low values [29, 32]. Breast DTI studies have shown that parameters derived from DTI have a high potential to efficiently detect and diagnose breast cancer [32,33,34,35,36]. The diffusion tensor parameters were also determined in the lactating breast by Nissan et al. [37], showing that the directional diffusion coefficients, as well as the maximal anisotropy were significantly lower by 13%–21% than in the normal breasts of premenopausal volunteers.

In this study we describe the application of non-invasive MRI protocols and image processing tools to perform a longitudinal follow-up of the changes in the anatomic and microstructural features of the entire mammary gland from lactation to post-weaning.

Results

Anatomic Features of the Breast during Lactation and Post-Weaning Obtained from Analysis of T2-Weighted Images

The anatomic changes from time of lactation to post-weaning indicated a marked reduction in breast size, a reduction in the fibroglandular tissue fraction and a parallel increase in the fat fraction as shown in the T2-weighted images of Fig. 1 and summarized in Table 1. The bright signal in the fat suppressed T2-weighted images recoded during lactation (Fig.1) reveals the presence of ductal milk, primarily near the nipple. The mammary size and fibroglandular fraction of the post-weaning volunteers and of the age- matched control volunteers were not significantly different (Table 1), indicating that the gross morphological features of the mammary gland return close to their natural pre-conception state during the first year post-weaning.

Fig. 1
figure 1

T2 weighted transversal images at the center of the breasts demonstrating the anatomical changes in the mammary gland from lactating to post-weaning in a 41 YO volunteer. a, c: T2 weighted images; b, d: Fat suppressed T2 weighted images. a, b: 11 month of lactation, c, d: 5 month post-weaning. The area of the breast and the FG% decreased from lactation to post-weaning by 56% and 27%, respectively. The bright signal in b demonstrates clearly the presence of milk in ductal regions

Full size image
Table 1 Anatomic characteristics of the breast during lactation, post-weaning and in control group
Full size table

Diffusion Tensor Parameters of the Breast during Lactation and Post-Weaning

Analysis of the diffusion tensor datasets yielded for each pixel in the fibroglandular tissue a geometric illustration of an ellipsoid (Fig. 2). The ellipsoids modeled the diffusion tensor and defined the direction of the water diffusion in three orthogonal directions ν1, ν2, ν3, as well as the magnitude of the directional water diffusion coefficients, λ1, λ2, λ3. The color coding of the ellipsoids (Fig. 2c) reflects the direction of the prime eigenvector ν1 which is further illustrated in the vector map of ν1 (Fig. 2d). The posterior-anterior direction from the base of the breast towards the nipple appears in Fig 2c and d as the dominant direction of the ellipsoids and of ν1 allowing the tracking of the main ductal orientation in the mammary gland.

Fig. 2
figure 2

Ellipsoids and ν1-vector maps of a central transversal slice of a lactating left breast in a 31 YO volunteer. a T2 weighted image b Schematic Ellipsoid form c Ellipsoid’s map d ν1-vector map of the fibroglandular tissue shown in the T2 image (the colors of the ellipsoids reflect the direction as indicated by the color of the three axis

Full size image

Further image analysis of the DTI datasets yielded parametric maps of the three directional diffusion coefficients, the mean diffusivity (MD), also termed ADC, the fractional anisotropy (FA) index (see mathematical definition in Materials and Methods) and the maximal anisotropy index, λ1-λ3 (Fig. 3). In the lactating breast the distribution of all diffusion coefficients indicated faster diffusion in the anterior regions close to the nipple as compared to the posterior regions. The distribution of the two anisotropy indices indicated low anisotropy in the anterior regions close to the nipple that increased towards the posterior regions. A similar distribution of the diffusion coefficients and of the anisotropy indices from the anterior to the posterior regions was observed in the post-weaning breasts but the changes were much less pronounced than in the lactating breasts.

Fig. 3
figure 3

Breast diffusion tensor parametric maps during lactation and post-weaning. Transversal diffusion tensor parametric maps overlaid on T2 weighted images of the same volunteer as in Fig. 2. a, b: λ1, λ2, and λ3 during lactation and 11 month post- weaning respectively. c, d: MD, FA, and λ1- λ3 during lactation and 11 month post- weaning respectively. λ1, λ2, λ3, MD and λ1- λ3 are in units of ×10−3 mm2/s. Mean λ1 values for the lactating and post-weaning states are: 2.22 ± 0.42 × 10−3 mm2/s and 2.24 ± 0.42 × 10−3 mm2/s, respectively. Mean λ2, λ3, and MD values decreased from lactation to post-weaning from 1.86 ± 0.34 × 10−3 mm2/s to 1.72 ± 0.36 × 10−3 mm2/s, 1.55 ± 0.36 × 10−3 mm2/s to 1.28 ± 0.38 × 10−3 mm2/s and 1.88 ± 0.34 × 10−3 mm2/s to 1.74 ± 0.34 × 10−3 mm2/s, respectively. The mean values of the anisotropy indices, FA and λ1-λ3, increased from 0.18 ± 0.10 to 0.28 ± 0.12 and from 0.66 ± 0.36 × 10−3 mm2/s to 0.96 ± 0.42 × 10−3 mm2/s, respectively

Full size image

Table 2 summarizes the statistical evaluation of all the diffusion tensor parameters after confirming the normal distribution of each parameter during lactation and post-weaning using the Shapiro-Wilk normality test. Overall, the lactating and post-weaning fibroglandular tissue have similar mean values for the prime diffusion coefficient λ1 (p = 0.93), whereas the mean values of the other two radial coefficients, λ2 and λ3 decreased significantly (p = 0.016, p = 0.001, respectively) leading to a significant decrease in the mean value of MD as well (p = 0.022). In parallel, the mean values of the anisotropic indices significantly increased from lactation to post-weaning by 32.1 ± 20.2% for FA (p = 0.0005) and by (26.1 ± 17.9)% for the maximal anisotropy index (p = 0.001). The decrease in the mean values of the radial diffusion coefficients, λ2 and λ3, and the increase in the anisotropy indices from the lactating to the post-weaning state are directly related to the reduction in the radial size of the ductal/glandular system reflecting the ductal involution post-weaning.

Table 2 DTI parameters of the breast fibroglandular tissue in lactating, post-weaning and premenopausal control groups
Full size table

The results in Table 2 described above were further analyzed by performing histogram analysis [38] and confirming that the variable symmetry (skewness) and variable peakedness (kurtosis) of each diffusion tensor parameter in the lactating and post-weaning states were normally distributed, as well as by constructing box and whiskers plots for each diffusion tensor parameter (Fig. 4). The box and whiskers plots showed that the mean values of λ1 in the lactation and post-weaning states almost fully overlap, exhibiting very similar median and inter-quartile range (IQR) values. In contrast, the values of λ2, λ3 and MD, partially overlapped between the two states and the values of both FA and λ1-λ3 showed negligible overlap between lactation and post-weaning with higher values during post-weaning.

Fig. 4
figure 4

Plots of box (median ± interquartile range - IQR) and whiskers (1.5 IQR) of the diffusion parameters in lactation and post-weaning. The values of λ1 in lactation overlap with those of post-weaning. The values of λ2, λ3 and MD/ADC are slightly but significantly lower from lactation to post-weaning. The anisotropy indices, FA and λ1-λ3, post-weaning are appreciably and significantly higher than during lactation (for significance see Table 1)

Full size image

Table 2 also compares between the breasts’ DTI parameters of the post-weaning volunteers and of the non-lactating control volunteers. The values of the diffusion coefficients in the control group were slightly, yet significantly higher than in the post-weaning-group, but the maximal anisotropy index λ1- λ3 did not change significantly.

Discussion

The physical principles underlying diffusion tensor imaging are based on the capability to measure and quantify the extent of anisotropic water movement in an environment restricted and hindered by the presence of cells and microstructural elements. Diffusion of water molecules in the mammary ducts presents a particular example of restricted movement: free diffusion parallel to the ducts’ walls and restricted diffusion in the directions perpendicular to the walls. The experimental protocol applied to track this anisotropic motion involved measurements at 30 different magnetic field gradient directions at two diffusion gradient strengths. In voxels with anisotropic water diffusion the analysis generated ellipsoidal forms whereas in voxels with isotropic diffusion spherical forms were generated (Fig. 2).

In this study we followed the mammary gland of healthy volunteers from the time of lactation to post-weaning by measuring the intra-individual alterations in breast T2-weighted images and in DTI derived parametric images. These alterations appeared to be associated with the structural, physiological and biochemical changes that the mammary gland undergoes during lactation and post-weaning [1,Aff2,Aff3,4]. Structurally the highly enlarged breasts of the lactating women underwent a substantial intra-individual decrease in size and fraction of fibroglandular tissue during post-weaning, indicating reversal to its gross pre-conception anatomic features. The changes in the diffusion tensor parameters between lactation to post-weaning appeared to stem from the microstructural changes resulting from the involution of the ductal/glandular system that leads to a reduction in the diameter of the ducts [8, 39]. However, in this study IVIM changes caused by flow in blood capillaries were neglected as this pseudo-diffusion process is effective only at a much lower strength of the diffusion gradients than that used here. We also assumed that there were negligible changes due to convective motion occurring during breast feeding [40] as there were no signs of milk release or suckling during the short duration of the DTI scanning.

The changes in the diameter of the ducts are likely to primarily affect the diffusion coefficients which are restricted by the ductal walls, namely, the radial diffusion coefficients λ2 and λ3 but not in λ1, which is in the direction parallel to the ductal walls and close to free diffusion. Indeed, a similar λ1 value was found in the lactation and post-weaning period while λ2 and λ3 significantly decreased post-weaning. Moreover, the higher values of the anisotropy indices in the post-weaning breast also reflected a reduction in ductal diameter.

The diffusion coefficients in each voxel may be affected by the water diffusion in the ductal lumens, in the epithelial cells and in the stroma. In voxels of ducts with diameter ≥ 2 mm the intensity of the water signal in the diffusion weighted images is mainly due to the ductal lumen as the spatial resolution of the DTI scanning protocol is of the same order of magnitude. The epithelial cells fraction in these voxels is small and may affect very slightly the diffusion coefficients. In voxels composed solely of stroma the diffusion coefficients are relatively high and similar in all directions and are only slightly affected by the presence of the small volume faction occupied by the fibrous cells. In regions where the diameter of the ducts is less than 2 mm the diffusion coefficients can include contribution from the lumens of several ducts or from stroma but the fraction of epithelial cells and stromal cells will still be low, hence, λ1 will still be high and the anisotropy indices may be low due to partial volume with the stroma. The finding of an increase in the anisotropy indices in the post-weaning breast therefore suggests that the changes from lactation to post-weaning mainly reflect diffusion in the ductal lumen. Nevertheless, we cannot exclude the presence of minor contribution of partial volume effects by stromal tissue in the post-weaning breast.

Assessments of the diameter of the mammary ducts in-vivo by US is limited to measure dimeters ≥0.5 mm [8]. Hence, only the diameters of large ducts was measured and found to range from 0.6 mm to 4.4 mm in the non-lactating breast [39, 41]. US of the lactating breast showed that the mean diameter of the ducts in the base of the nipple was 2.0 ± 0.8 mm ranging from 1.0 mm to 4.4 mm and it reduced to a mean diameter at the first milk duct branch to 1.3 ± 0.6 mm [8], suggesting similar sizes for the large ducts in lactating and non-lactating mammary glands. In contrast, ex-vivo studies of histological specimens using computerized image analysis showed that the mean diameter of normal ducts was 0.09 mm with a range of 0.039 mm to 0.314 mm [42]. The discrepancy between the in-vivo and ex-vivo measurements in the non-lactating breast most likely stems from the fact that US measured only the large ducts close to the nipple whereas the ex-vivo study averaged the size of the ducts over the entire breast. In addition, processing of the ex-vivo tissue for histological evaluation, could cause shrinking of the ducts. According to Einstein’s equation that relates the diffusion coefficient to the displacement of the moving molecule and to the diffusion time, the mean displacement of free water diffusion for the diffusion time that we have selected (47 milliseconds) is ~0.03 mm. Thus, in the large ducts, only a small part of the water molecules, those close to the walls, will be restricted, and the average FA will be relative low. In the small ducts (less than 0.1 mm) the restriction in the ducts can yield high FA values, but due to partial volume effects of the stroma in the entire voxel, the FA will be also low. However, the higher mean FA values measured in post-weaning as compared to lactation (Table 2, Fig. 4) suggest a minor contribution from stromal partial volume effects and reflects a true reduction in the average diameter of the ductal system. It is anticipated from a preliminary report [43] that detailed tracking of the ductal system with advanced algorithms would enable in the future monitoring changes in individual ducts.

The fact that the least restricted water diffusion coefficient λ1 remained similar during lactation and post-weaning and did not return to the range of normal values for this coefficient in premenopausal control breasts suggests that the fluid in the ductal/glandular tissue in the first year post-weaning may still contain ingredients that reduce the ductal fluid viscosity [4, 44].

It has been previously shown that the directional diffusion coefficients and mean diffusivity are significantly lower in breast cancers as compared to normal breast tissue and can efficiently differentiate malignant from normal breast tissue and from benign breast lesions [31,32,33,34,35,36]. However, the clinical ability of diffusion MRI methods to detect cancer in pregnant or lactating patients has not yet been investigated. Currently, only few investigations evaluated the ability of the standard DCE- MRI protocol to detect cancer during lactation [15]. The results in these studies indicated potential difficulties stemming from increased permeability and hence rapid background enhancement of the lactating parenchyma that hamper MRI detection of malignancy. As DTI is not sensitive to hormonal induced changes in the micro-vascular system (37) the diffusion coefficients in the normal tissue of the lactating breasts should remain high and should not mask the presence of cancer tissue.

DTI is a safe, completely non-invasive and relatively fast method that exhibits high intrinsic contrast between normal and malignant breast tissue and shows high efficiency in detecting malignancy in dense breasts [45]. These advantages and the studies showing that pregnancy associated breast cancers (PABCs) share many histopathological similarities with breast carcinoma occurring in young women [46] and references cited therein, suggest a potential significant role for DTI in the detection of PABCs. Specifically, the steadiness and high values of λ1 during lactation and post-weaning and the previously reported low values of λ1 values in breast cancer lesions [31, 35] suggest a possible advantage of this parameter for evaluating malignant changes during these states and calls for specific testing of the DTI method for assisting in the diagnostic challenges of PABC [47].

In summary, the modifications that the breast undergoes from lactation to post-weaning affect the diffusion tensor parameters showing a stable λ1 during lactation and post-weaning but decreased values of λ2 and λ3 and increased anisotropy indices, λ1-λ3 and FA, during post weaning as compared to the lactation period. These results reflect the unique involution process of the post-weaning ductal glandular system. The relative stability of λ1 and its high values in the lactating and post-weaning breasts in comparison to values determined in breast cancer suggest a potential role for DTI in PABC.

Materials and Methods

Volunteers

All protocols were approved by Meir Hospital ethic committee and a signed informed consent was obtained from all participants. Overall, twenty-one healthy female volunteers participated in this longitudinal study between February 2013 to January 2015. Eleven of these volunteers (median age 36, range: 30 to 41) were prospectively recruited for a scan during lactation (5 to 24 months of lactation), and were then recalled for a second follow-up scan post-weaning (3–11 months after complete weaning). One lactating volunteer was excluded due to undisclosed suspicious findings in her examinations and she has been followed up until the submission of this paper. Additionally, ten healthy age matched volunteers (median age 36, range 29 to 40), that were not pregnant or lactating in the year prior to the study were scanned as a control group for normal breast tissue.

MRI Protocol

Images were acquired on a 3-T whole-body MRI scanner: MAGNETOM Trio, Tim System, (Siemens, Erlangen, Germany) equipped with a transmitting body coil and receiving 4-channel (Siemens, Erlangen, Germany) or 7 channel breast array-coils (Invivo, USA). Two scanning protocols were applied 1. Fast transversal, bi-lateral, 2D T2-weighted images with and without fat-suppression, using echo time/repetition time (TE/TR) 122/5500 ms, resolution of 0.8 × 0.6 × 2.5 mm3 and total acquisition time of 4.5 min. 2. Transversal bi-lateral 2D DTI images with fat-suppression, using spin-echo twice refocused echo-planar-imaging (EPI) sequence, diffusion gradients in 30 directions, b-values of 0 and 700 s/mm2, TE/TR of 120/10400 ms, diffusion time of 47 ms, resolution of 1.9 × 1.9 × 2.5 mm3 and total acquisition time of 6 min. In both protocols the field of view, number of slices and slice thickness were identical allowing matched comparisons of the anatomic features and parametric DTI maps. The long TE was selected so that the resulting diffusion time in the twice refocused EPI sequence will allow optimized detection of the ductal system anisotropy [31] and still provide sufficient signal intensity above the noise level. The signal to noise ratio (defined as the ratio between the average signal intensity in the ROI of the fibroglandular tissue and the average standard deviation of the intensity in noise ROIs of the b0 diffusion images) ranged from 28 to 40 during lactation and from 12 to 30 post-weaning, allowing straightforward processing of the data to yield the diffusion parameters as described below in the Image processing section.

Image Processing

Regions of interest (ROIs) of the breasts contour in three transversal central slices and ROIs of the fibroglandular tissue in these slices were manually delineated on non-fat-suppressed T2-weighted images assisted with fat suppressed T2-weighted images (Fig. 1). Calculation of the area in the ROIs of the breast contour reflected their size. The number of pixels in the ROIs of the fibroglandular tissue and in the ROIs of the breast contour served to calculate the percentages of the fibro-glandular tissue.

ROIs for calculating the mean and standard deviation (SD) values of the DTI parameters included the entire fibroglandular tissue of the two breasts in three transversal central slices. The diffusion tensor parameters and the anisotropy indices were calculated pixel by pixel for each given slice using a home-built software as previously described [32, 48]. The output of the software yielded a vector map of the main eigenvector (ν1) defining the diffusion direction with the highest diffusion coefficient, as well as, a 3D–ellipsoid map reflecting the tensor shape in each pixel defined by three eigenvectors in three orthogonal axes coinciding with the frame of the tissue (ν1, ν2, ν3) and their corresponding eigenvalues reflecting the values of the three diffusion coefficients (λ1, λ2, λ3) arranged from high to low. A three color coding display indicated the direction of the fastest diffusion vector ν1 in the vector and ellipsoids maps (Fig. 2). In addition the software yielded parametric maps at pixel resolution of λ1, λ2 and λ3, the mean diffusivity (averaging over 30 diffusion gradient directions) and two anisotropy indices, maximal anisotropy index, λ1-λ3, and fractional anisotropy (FA) index [29], normalized to range between 0 to 1.0, reflecting isotropic diffusion to diffusion in one direction, respectively.

Statistical Analysis

Statistical analysis included calculation of median and range of the anatomical features and mean and standard deviation (SD) of the DTI parameters in selected ROIs at the center of the two breasts delineated as described above. Two tailed paired Student’s t-test was applied for evaluating differences between measurements of the same volunteers at the time of lactation and post-weaning. Two tailed unpaired Student’s t-test was applied for comparison of the post-weaning group with the control group (Excel 2010, Microsoft). Histogram analysis (MedCalc software Version 16.8, 2016) yielded box and whiskers plots (MedCalc software Version 16.8, 2016). Statistical significance was defined as p < 0.05.