Dose-compatible grating-based phase-contrast mammography on mastectomy specimens using a compact synchrotron source

With the introduction of screening mammography, the mortality rate of breast cancer has been reduced throughout the last decades. However, many women undergo unnecessary subsequent examinations due to inconclusive diagnoses from mammography. Two pathways appear especially promising to reduce the number of false-positive diagnoses. In a clinical study, mammography using synchrotron radiation was able to clarify the diagnosis in the majority of inconclusive cases. The second highly valued approach focuses on the application of phase-sensitive techniques such as grating-based phase-contrast and dark-field imaging. Feasibility studies have demonstrated a promising enhancement of diagnostic content, but suffer from dose concerns. Here we present dose-compatible grating-based phase-contrast and dark-field images as well as conventional absorption images acquired with monochromatic x-rays from a compact synchrotron source based on inverse Compton scattering. Images of freshly dissected mastectomy specimens show improved diagnostic content over ex-vivo clinical mammography images at lower or equal dose. We demonstrate increased contrast-to-noise ratio for monochromatic over clinical images for a well-defined phantom. Compact synchrotron sources could potentially serve as a clinical second level examination.

inconclusive diagnosis after mammography and ultrasonography. The combination of absorption contrast and edge-enhancement resulting from phase shifts and subsequent free-space propagation enabled higher relative visibilities of abnormalities and the number of true-negative findings was significantly increased, while the dose level was lower or comparable to conventional mammography [4][5][6] . However, mammography with synchrotron radiation has the disadvantages of limited availability, remoteness with respect to clinics, and high costs.
A different approach to access phase information is x-ray imaging with a grating interferometer, which simultaneously provides attenuation, differential phase and dark-field signals 7,8 . Several studies have already successfully shown the application of phase-contrast imaging for mammography [9][10][11] and especially for the classification of microcalcifications 12 . The dark-field signal is related to small-angle scattering and has been shown to improve the visibility of microcalcifications 13 or could even help to distinguish between different types of calcifications 14 .
Various studies carried out with dissected cancerous breasts at conventional x-ray tubes have been conducted, showing significantly higher diagnostic content provided by the multimodal images [15][16][17] . However, initial investigations of this technique were performed with applied radiation doses far above radiological guidelines. Only recently, dose-compatible phase-contrast mammography images were presented 18 . In addition, the feasibility of installing a grating interferometer into a conventional digital mammography unit was successfully demonstrated 19,20 . Challenges are the low sensitivity of the interferometer due to the limited distance between the gratings, and dose compatibility.
In order to combine the advantages and avoid some disadvantages of the two approaches, we suggest to perform grating-based multimodal mammography at a compact synchrotron source based on inverse Compton scattering. These recently developed x-ray sources are compact enough to fit into regular laboratories 21 . They offer a monochromatic x-ray beam that is tunable in energy and partially coherent, allowing to obtain increased diagnostic content from grating-based phase-contrast and dark-field images while achieving equal image quality at lower dose compared to the aforementioned laboratory setups. In addition, monochromatic conventional absorption-based imaging or propagation-based phase-contrast imaging as done at SYRMEP is possible. Since the spatial and financial requirements of inverse Compton sources are relatively small compared to large-scale synchrotrons, the installation at medical centers appears feasible. The Munich Compact Light Source (MuCLS, Lyncean Technologies, USA) is the first commercial installation of an inverse Compton compact synchrotron source 22 .
Here, we present both absorption-only and grating-based multimodal images of freshly dissected cancerous mastectomy specimens acquired at the MuCLS at lower or equal dose compared to state-of-the-art clinical images. Furthermore, a dose study and analysis of the contrast-to-noise-ratio on a well defined sample, a mammographic accreditation phantom, was conducted. In addition, we present a comparison of the spatial resolution for experimental and clinical images. We believe that compact synchrotron sources like the MuCLS have great potential to bring benefits to clinical imaging, in particular for mammography, but also other fields, like coronary angiography 23 , could profit from the monochromatic, tunable x-ray beam.

Results
Four freshly dissected mastectomy specimens and a mammographic accreditation phantom were investigated with ex-vivo clinical mammography (cevAC-Mx) and both conventional absorption and grating-based multimodal monochromatic experimental mammography. Abbreviations used for the different contrast modalities in the following are given in Table 1. An overview of the CNR and the spatial resolution analysis is presented in Tables 2 and  3, respectively. The tumor characteristics for each sample are summarized in Table 4. The applied mean glandular dose (MGD) and the exposure times are also given in Table 4 for each sample and imaging modality. Improved delineation of tumor lesions in phase-contrast image. Specimen I presented with a palpable mass in the right breast and skin retraction. Conventional in-vivo mammography (civAC-Mx) showed an asymmetry in the respective region ( Fig. 1(e)). Ultrasound revealed inhomogeneities and several hypoechogenic lesions. Additional dynamic MRI showed an extensive infiltration with tumor branches extending close to the pectoralis muscle. Histopathology of the mastectomy specimen revealed a multicentric, invasive lobular carcinoma (G2) with extensive manifestations of a lobular carcinoma in situ. Figure 1(a-d) show monochromatic grating-based absorption-contrast (mgbAC-Mx), differential phase-contrast (mgbDPC-Mx), dark-field contrast (mgbDFC-Mx) and ex-vivo clinical absorption-contrast (cevAC-Mx) mammography images of the mastectomy specimen from specimen I, measured in anteroposterior  Specimen II had a history of breast cancer. While aftercare mammography and ultrasound showed a slight, ill-defined density in the scar area but no circumscribed lesion, dynamic MRI of the right breast revealed an extensive carcinoma. Histology of the mastectomy sample showed nodular infiltrations of a poorly differentiated (G3) invasive breast cancer of no special type. Figure 2(a-c) show cevAC-Mx, mgbAC-Mx, and mgbDPC-Mx measured in an AP orientation of the specimen. While the differentiation of tumor and scar tissue is difficult in the AC images, the mgbDPC-Mx (c) depicts the tumor spiculae (red arrows) as verified by histopathology (d). The H&E stained histology image shows the infiltration of the scar tissue (pink) with tumorous cells (purple).

Fibers
Calcifications Tumor Masses   Detection of microcalcifications at reduced dose. In Specimen III, civAC-Mx revealed a cluster of microcalcifications in the retromamillary area. Ultrasound found a hypoechoic lesion and MRI showed a corresponding suspicious mass with early contrast enhancement. Histology of the mastectomy sample revealed an invasive carcinoma of no special type, G2, and an adjacent intraductal carcinoma (DCIS), with comedonecrosis and extensive microcalcifications. Figure 3(a-e) display cevAC-Mx, mAC-Mx, mgbAC-Mx, mgbDPC-Mx and mgbDFC-Mx measured in AP orientation. A calcification cluster that had preoperatively been clip-marked (magenta arrows) is shown in magnification in the small inlets. The mAC-Mx image (b) equally shows the extent of the microcalcification cluster  Specimen IV was diagnosed with three microcalcification clusters in civAC-Mx, two of which were partially extracted in a biopsy. Histology of the mastectomy sample revealed a bifocal, well differentiated (G1) breast cancer of no special type, with an extensive intraductal carcinoma (DCIS) and microcalcifications. Dose-study with a mammographic accreditation phantom. In order to quantitatively compare monochromatic MuCLS mammography with a clinical mammography unit, a contrast-to-noise-ratio (CNR) analysis was performed for a mammographic accreditation phantom (Gammex, Model 156, www.sunnuclear. com/solutions/diagnostic/mammography/156phantom). The image of the accreditation phantom acquired by the conventional mammography system by automated exposure control is compared to both monochromatic absorption-contrast images and also grating-based trimodal images acquired at different mean glandular doses. A CNR analysis for different test objects is presented in Table 2. A selection of images is shown in Fig. 5.
It is clearly visible that the absorption images cevAC-Mx ( Fig. 5(a)), mAC-Mx ( Fig. 5(b,c)), and mgbAC-Mx (Fig. 5(d)) fulfill the standard criteria of clinical image quality set by the ACR 24 , which require a minimum of 4 fibrils, 3 groups of microcalcifications, and 3 tumor masses to be resolved. A quantitative analysis (Table 2) demonstrates that the CNR in the mAC-Mx (b) image considerably exceeds the CNR in the cevAC-Mx (a) image at equal dose. The mAC-Mx (c) still slightly outperforms the cevAC-Mx image (a) at 20% reduced dose.
For the grating-based images, the mgbAC-Mx (Fig. 5(d)) image at slightly lower dose than cevAC-Mx (a) provides a comparable CNR. Especially for the small tumor masses, the mgbDFC-Mx (f) CNR by far outperforms the clinical image (a), where the smallest tumor mass cannot be distinguished from the background (CNR < 1). This is still the case when the dose is reduced to 35% of the clinical value (i). In addition, the mgbDPC-Mx (e), for which a quantitative analysis is not possible due to the differential nature of the signal, allows to depict all Importantly, while the low-dose mgbAC-Mx (g) does not fulfill the ACR criteria, the combination with mgbDPC-Mx (h) and mgbDFC-Mx (i) meets or, in the case of the tumor masses, even exceeds the ACR criteria.

Resolution.
A power spectrum analysis was performed for all absorption and dark-field images cevAC-Mx, mAC-Mx, mgbAC-Mx and mgbDFC-Mx (not for DPC-Mx due to the differential nature of the signal) in order to compare their resolution 25 . As illustrated in Table 3, the resolution is mainly in the range between 3 and 4 linepairs (LP) per mm. The analysis shows that a higher resolution was achieved in the mAC-, mgbAC-and mgbDFC-MX than in the cevAC-Mx for all investigated samples, at lower or equal dose of the monochromatic images.

Discussion
We presented conventional absorption-contrast as well as grating-based multimodal mammography images of freshly dissected mastectomy samples acquired with a monochromatic x-ray beam at a compact synchrotron source. Comparison with clinical ex-vivo mammography images showed equal diagnostic quality at lower dose or superior diagnostic quality at equal dose of the monochromatic images. Tumorous lesions could be identified significantly better in the phase-contrast modality as verified by histopathology (Specimens I, II). Microcalcification clusters were revealed to an equal extent as in the clinical mammography image, at a significantly reduced dose for the monochromatic images (Specimens III, IV). A power spectrum analysis showed that the monochromatic absorption-contrast and dark-field-contrast images (mAC-Mx, mgbAC-Mx, mgbDFC-Mx) could achieve better resolution than the clinical image (cevAC-Mx) at lower or equal dose, proving that the small source size of the MuCLS is beneficial for the resolution.
A dose study conducted with a mammographic accreditation phantom showed a superior CNR of various test objects at equal mean glandular dose and a higher or comparable CNR at 20% reduced dose for the monochromatic mammography images compared to the clinical one. The grating-based phase-contrast and dark-field image yield an enhanced diagnostic content for the fibers and tumor masses, respectively. The dark-field image outperforms the clinical image for the tumor masses even at a 65% lower dose.
The presented results demonstrate that mammography at a compact synchrotron source like the MuCLS has great potential to improve the diagnostic quality of mammography. The compact size and limited financial requirements compared to large-scale synchrotron source allow for envisioning the installation at hospitals in the future. Thus, the implementation of a compact synchrotron could enable a second-level examination in the case of inconclusive diagnosis (as performed at the SYRMEP beamline [4][5][6] ), in addition with the possibility to choose between conventional, but monochromatic absorption imaging and grating-based trimodal imaging.
The main limitation of our study was the different orientation (anteroposterior instead of cranio-caudal or mediolateral-oblique) for three of the four specimens compared to clinical in-vivo mammography. The different orientation was necessary in order to avoid artifacts originating from surgical resection margins. Therefore, the ex-vivo images are not fully comparable to in-vivo images concerning orientation and sample thickness.
The limited sizes of beam and gratings required scanning of the samples, resulting in an increased scan time incompatible with clinical applications. Only recently, the feasibility of manufacturing stitched gratings in order to cover larger fields of view was demonstrated 26 . The field of compact synchrotron sources is constantly developing 22,27,28 and considerable improvements and further developments of the technology with regard to increased stability, flux and beam size can be expected. An increased field of view can be achieved by employing a larger exit aperture. The currently available 4 mrad opening angle is only limited by the diameter of the exit aperture. By simply replacing the exit aperture by one with a larger diameter, the beam size at the experimental station would increase. In addition, other potential clinical applications will profit especially from an extension of the energy range.
Concerning the mean glandular dose applied in mammography, our results show that dose-compatible grating-based mammography is feasible at a monochromatic source. The used setup even has the potential for further dose reduction by thinning the grating wafers from the currently used 500 μm thick Si substrate down to 200 μm and by modifying the setup to allow positioning of the sample to behind the phase grating.
In conclusion, we believe that monochromatic x-rays provided by compact synchrotron sources like the MuCLS, the first commercially installed source of this type, can enable significant improvements in diagnostic x-ray imaging. Mammography could benefit from enhanced diagnostic image content and improved resolution at a reduced dose. In addition, as the x-ray energy of these sources is tunable, also other applications, like for instance coronary angiography 23 , could benefit from the implementation of compact synchrotron sources in a clinical setting. Ex-vivo mammography. The mastectomy samples were intraoperatively marked with surgical sutures for 3D orientation. The samples were fixed within a metal-framed sample holder to afford adequate breast compression. The position was cranio-caudal, or anteroposterior in case a cranio-caudal position was impossible due to the shape of the dissected specimen. A digital ex-vivo mammography image was acquired at a clinical mammography unit (Hologic Selenia Dimensions) with a pixel size of 70 × 70 μm 2 .
Working principle of the MuCLS. The MuCLS is a compact synchrotron based on inverse Compton scattering and was developed and manufactured by Lyncean Technologies Inc., USA. As a relativistic electron bunch collides with a laser pulse, x-ray photons are produced. In good approximation for head-on collision and backscattering, the energy E x of the emitted x-rays is given by E x = 4γ 2 E L , where γ = E e /E 0 with E 0 the rest energy and E e the total energy of the electron, and E L is the energy of the laser photons. The electrons circulate in a miniature storage ring and collide with the laser pulse stored in a bow-tie enhancement cavity upon each revolution (cf. Fig. 6) 21,22,29 .
The produced x-rays are collimated to a 4 mrad cone. The x-ray beam is monochromatic, partially coherent, and the energy is tunable between 15 and 35 keV by adjusting the electron energy. The MuCLS currently offers -on a daily basis -a flux exceeding 2.4 · 10 10 photons per second (at 35 keV), and a source size of less than 50 × 50 μm 2 . At a distance of about 16 m from the interaction point, the beam has an elliptic shape of 62 mm × 74 mm 22 . Mammography at the MuCLS. Maintaining the position of the breast specimen within the sample holder in order to ensure comparability, monochromatic mammography images were acquired at the MuCLS. A Talbot interferometer was set up approximately 16 m from the interaction point of the MuCLS, as shown schematically in Fig. 6. The inter-grating distance was ~25 cm, the grating periods were p 1 = 4.9 μm and p 2 = 5.0 μm for phase and analyzer grating, respectively. The signals were retrieved from phase stepping as described in 8 . Conventional absorption-contrast images were acquired without the grating interferometer. A Dexela 1512 flatpanel detector (PerkinElmer Inc., USA) with a Gd 2 O 2 S scintillator was used, yielding an effective pixel size of 71 × 71 μm 2 . Scanning the sample and afterwards stitching of several images was required to cover the whole sample, using linear ramps to blend the overlapping areas. Table 4 summarizes the acquisition parameters for each scan. The exposure times in Table 4 are given for one Stitching-scan. The MuCLS x-ray energy was tuned to 25 keV for all samples. The interferometer visibility was approximately 45-50% for all measurements.

Dose calculation. The MGD automatically registered by the clinical mammography unit for clinical in vivo-
and ex vivo-images is given in Table 4 for each sample. It is calculated based on the x-ray tube settings and the compressed sample thickness. The MGD value for the cevAC-Mx images was corrected when an incorrect sample thickness was assumed due to the shape of the sample holder. A correction factor was calculated based on the half value layer of Aluminum for the respective acquisition settings and the assumed and correct breast thickness, from tabulated values by 30,31 .
For the monochromatic mammography performed at the MuCLS, the MGD was calculated using monoenergetic normalized glandular dose coefficients DgN(E) (with E the x-ray energy) as tabulated by Boone et al. 32 , taking into account the MuCLS spectrum 22 , and summing over all energy bins E: The formula for the MGD was adapted from 32 to be used with air kerma K instead of the older unit exposure, as the fitted equation for exposure given by 32 is incorrect 33 . The DgN (E) values were selected according to the compressed thickness of the dissected breasts, and assuming a 50%/50% distribution of glandular and adipose tissue. The air kerma for the MuCLS beam K per energy bin E can be calculated for known photon flux Φ at the sample position and x-ray spectrum 34 : en air where (μ en /ρ) air (E) is the mass energy attenuation coefficient of air 35 . The photon flux per energy bin Φ(E) was calculated from a single photon counting Pilatus 200 K detector (Dectris Ltd., Switzerland), taking into account the x-ray spectrum and the efficiency of the Si sensor of the detector 36 . The calculation of the air kerma was validated using a soft x-ray ionization chamber (Model 34013, PTW Freiburg GmbH, Germany) and values agreed within the measurement uncertainties of ±10%. A scintillation counter was calibrated to calculated air kerma prior to each sample measurement and the scintillation counts were logged for each measurement frame, thus yielding an exact measure of air kerma for each scan.
Image Analysis. The CNR was calculated according to the definition BG where S 1 and S 2 are the average signals in two regions of interest (ROI) which should be compared, and σ BG is the standard deviation within a larger ROI located in the background region. The resolution of the images was determined by analyzing their power spectra 25 . The images were Fourier transformed and a Gaussian filter was applied on the squared norm. The resolution of the images is then given by the maximal spatial frequency where the spectral power of the signal equals twice the spectral power of the noise baseline, taking into account the effective pixel size of clinical and experimental images.