Background and purpose: The recent emergence of neural networks models for the analysis of breast images has been a breakthrough in computer aided diagnostic. This approach was not yet developed in Contrast Enhanced Spectral Mammography (CESM) where access to large databases is complex. This work proposes a deep-learning-based Computer Aided Diagnostic development for CESM recombined images able to detect lesions and classify cases. Material and methods: A large CESM diagnostic dataset with biopsy-proven lesions was collected from various hospitals and different acquisition systems. The annotated data were split on a patient level for the training (55%), validation (15%) and test (30%) of a deep neural network with a state-of-the-art detection architecture. Free Receiver Operating Characteristic (FROC) was used to evaluate the model for the detection of 1) all lesions, 2) biopsied lesions and 3) malignant lesions. ROC curve was used to evaluate breast cancer classification. The metrics were finally compared to clinical results. Results: For the evaluation of the malignant lesion detection, at high sensitivity (Se<0.95), the false positive rate was at 0.61 per image. For the classification of malignant cases, the model reached an Area Under the Curve (AUC) in the range of clinical CESM diagnostic results. Conclusion: This CAD is the first development of a lesion detection and classification model for CESM images. Trained on a large dataset, it has the potential to be used for helping the management of biopsy decision and for helping the radiologist detecting complex lesions that could modify the clinical treatment.
KEYWORDS: Data modeling, Digital breast tomosynthesis, Breast, Model-based design, Sensors, Optical spheres, 3D modeling, Medical imaging, 3D image processing, Breast imaging
Model-based iterative reconstruction (MBIR) is implemented to process full clinical data sets of dedicated breast tomosynthesis (DBT) in a low dose condition and achieves less spreading of anatomical structure between slices. MBIR is a statistical based reconstruction which can control the trade-off between data fitting and image regularization. In this study, regularization is formulated with anisotropic prior weighting that independently controls the image regularization between in-plane and out-of-plane voxel neighbors. Studies at complete and partial convergence show that the appropriate formulation of data-fit and regularization terms along with anisotropic prior weighting leads to a solution with improved localization of objects within a more narrow range of slices. This result is compared with the solutions using simultaneous iterative reconstruction technique (SIRT), which is one of the state of art reconstruction in DBT. MBIR yields higher contrast-to-noise for medium and large size microcalcifications and diagnostic structures in volumetric breast images and supports opportunity for dose reduction for 3D breast imaging.
In breast X-ray images, texture has been characterized by a noise power spectrum (NPS) that has an inverse power-law shape described by its slope β in the log-log domain. It has been suggested that the magnitude of the power-law spectrum coefficient β is related to mass lesion detection performance. We assessed β in reconstructed digital breast tomosynthesis (DBT) images to evaluate its sensitivity to different typical reconstruction algorithms including simple back projection (SBP), filtered back projection (FBP) and a simultaneous iterative reconstruction algorithm (SIRT 30 iterations). Results were further compared to the β coefficient estimated from 2D central DBT projections. The calculations were performed on 31 unilateral clinical DBT data sets and simulated DBT images from 31 anthropomorphic software breast phantoms. Our results show that β highly depends on the reconstruction algorithm; the highest β values were found for SBP, followed by reconstruction with FBP, while the lowest β values were found for SIRT. In contrast to previous studies, we found that β is not always lower in reconstructed DBT slices, compared to 2D projections and this depends on the reconstruction algorithm. All β values estimated in DBT slices reconstructed with SBP were larger than β values from 2D central projections. Our study also shows that the reconstruction algorithm affects the symmetry of the breast texture NPS; the NPS of clinical cases reconstructed with SBP exhibit the highest symmetry, while the NPS of cases reconstructed with SIRT exhibit the highest asymmetry.
In breast X-ray imaging, breast texture has been characterized by a radial noise power spectrum (NPS) that has an inverse power-law shape with exponent β. The technique to estimate the radial power-law coefficient β is typically based on averaging 2-dimensional noise power spectra (NPS), calculated from partly overlapping image regions each weighted by a suitable window function. The linear regression applied over a selected frequency range to the logarithm of the 1- dimensional NPS as a function of the logarithm of the radial frequencies, gives β. For each step in this process, several alternative techniques have been proposed. This paper investigates the effect of image region of interest (ROI) size, image data windowing and alternative ways to determine radial frequency in terms of bias, variance and root mean square error (RMSE) in the estimated β. The effects of these three factors were analytically derived and evaluated using synthetic images with known β varying from 1 to 4 to cover the range of textures encountered in 2D and 3D breast X-ray imaging. Our results indicate that the RMSE in estimated β is smallest when the ROIs are multiplied with an appropriate window function and either no radial averaging or radial averaging with small frequency bins is applied. The ROI size yielding the smallest RMSE depends on several factors and needs to be validated with numerical simulations. In clinical practice however, there might be a need to compromise in the choice of the ROI size to balance between the RMSE magnitudes inherent to the applied β estimation technique and encompass the breast texture range so as to obtain an accurate shape of the NPS. When using 2.56 cm x 2.56 cm ROI sizes, applying a 2D Hann window and no radial frequency averaging, the RMSE in the estimated β ranges from 0.04 to 0.1 for true β values equal to 1 and 4. While many subtleties in real images were not modeled to simplify the mathematics in deriving our results, this work is illustrative in demonstrating the limits of commonly used algorithm steps to estimate accurate β values.
KEYWORDS: Signal attenuation, Digital breast tomosynthesis, Sensors, Reconstruction algorithms, Optical spheres, Model-based design, Breast, Data modeling, Computed tomography, Tissues
Model-based iterative reconstruction (MBIR) is an emerging technique for several imaging modalities and appli-
cations including medical CT, security CT, PET, and microscopy. Its success derives from an ability to preserve
image resolution and perceived diagnostic quality under impressively reduced signal level. MBIR typically uses a
cost optimization framework that models system geometry, photon statistics, and prior knowledge of the recon-
structed volume. The challenge of tomosynthetic geometries is that the inverse problem becomes more ill-posed
due to the limited angles, meaning the volumetric image solution is not uniquely determined by the incom-
pletely sampled projection data. Furthermore, low signal level conditions introduce additional challenges due to
noise. A fundamental strength of MBIR for limited-views and limited-angle is that it provides a framework for
constraining the solution consistent with prior knowledge of expected image characteristics. In this study, we
analyze through simulation the capability of MBIR with respect to prior modeling components for limited-views,
limited-angle digital breast tomosynthesis (DBT) under low dose conditions. A comparison to ground truth
phantoms shows that MBIR with regularization achieves a higher level of fidelity and lower level of blurring
and streaking artifacts compared to other state of the art iterative reconstructions, especially for high contrast
objects. The benefit of contrast preservation along with less artifacts may lead to detectability improvement of
microcalcification for more accurate cancer diagnosis.
KEYWORDS: Nonlinear filtering, Image filtering, Digital filtering, Image processing, Image quality, Linear filtering, Denoising, Gaussian filters, Statistical modeling, Signal to noise ratio
Non-linear image processing and reconstruction algorithms that reduced noise while preserving edge detail are currently being evaluated in medical imaging research literature. We have implemented a robust statistics analysis of four widely utilized methods. This work demonstrates consistent trends in filter impact by which such non-linear algorithms can be evaluated. We calculate observer model test statistics and propose metrics based on measured non-Gaussian distributions that can serve as image quality measures analogous to SDNR and detectability. The filter algorithms that vary significantly in their approach to noise reduction include median (MD), bilateral (BL), anisotropic diffusion (AD) and total-variance regularization (TV). It is shown that the detectability of objects limited by Poisson noise is not significantly improved after filtration. There is no benefit to the fraction of correct responses in repeated n-alternate forced choice experiments, for n=2-25. Nonetheless, multi-pixel objects with contrast above the detectability threshold appear visually to benefit from non-linear processing algorithms. In such cases, calculations on highly repeated trials show increased separation of the object-level histogram from the background-level distribution. Increased conspicuity is objectively characterized by robust statistical measures of distribution separation.
This work investigates a dual-energy subtraction technique for cone-beam breast CT combined with an iodinated
contrast agent. Simulations were performed to obtain optimally enhanced iodine-equivalent and morphological images.
The optimal x-ray beam energies and average glandular dose allocation between the LE and HE images were identified.
Cylindrical phantoms were simulated with 10, 14 and 18 cm diameters and composed of 50% fibroglandular breast tissue
equivalent material. They contained spherical lesion inserts composed of 0, 25, 75 and 100% fibroglandular equivalent
tissues, homogeneous mixtures of 50% fibroglandular equivalent tissue and 0.5, 1.0, 2.5 and 5.0 mg/cm3 iodine, as well
as pure calcium hydroxyapatite, emulating calcifications. An acquisition technique with 600 projection images is
proposed. Only primary x-ray photons were simulated and a perfect energy-integrating detector was considered. LE and
HE beams ranging from 20 keV to 80 keV were investigated. The LE and HE projections were reconstructed using a
filtered backprojection algorithm. The LE volume provided the morphological image while the iodine-equivalent volume
was obtained by recombining the LE and HE volumes. Contrast-to-noise ratio (CNR) between the spherical inserts and
background breast tissue normalized to the square root of the total AGD (CNRD) was used as figure-of-merit for lesion
detectability. Based on maximizing CNRD, a 30keV/34keV LE/HE pair and a ~50/50% LE/HE AGD allocation were
found to provide the best possible performance for iodine and morphological imaging for an average size breast.
Dual-energy contrast-enhanced digital breast tomosynthesis (DE
CE-DBT) image quality is affected by a large parameter
space including the tomosynthesis acquisition geometry, imaging technique factors, the choice of reconstruction
algorithm, and the subject breast characteristics. The influence of most of these factors on reconstructed image quality is
well understood for DBT. However, due to the contrast agent uptake kinetics in CE imaging, the subject breast
characteristics change over time, presenting a challenge for optimization . In this work we experimentally evaluate the
sensitivity of the reconstructed image quality to timing of the
low-energy and high-energy images and changes in iodine
concentration during image acquisition. For four contrast uptake patterns, a variety of acquisition protocols were tested
with different timing and geometry. The influence of the choice of reconstruction algorithm (SART or FBP) was also
assessed. Image quality was evaluated in terms of the lesion
signal-difference-to-noise ratio (LSDNR) in the central slice
of DE CE-DBT reconstructions. Results suggest that for maximum image quality, the low- and high-energy image
acquisitions should be made within one x-ray tube sweep, as separate low- and high-energy tube sweeps can degrade
LSDNR. In terms of LSDNR per square-root dose, the image quality is nearly equal between SART reconstructions with
9 and 15 angular views, but using fewer angular views can result in a significant improvement in the quantitative
accuracy of the reconstructions due to the shorter imaging time interval.
Dual-energy contrast-enhanced digital breast tomosynthesis (CE-DBT) using an iodinated contrast agent is an imaging
technique providing 3D functional images of breast lesion vascularity and tissue perfusion. The iodine uptake in the
breast is very small and causes only small changes in x-ray transmission; typically less than 5%. This presents
significant technical challenges on the imaging system performance. The purpose of this paper was to characterize
image lag and scattered radiation and their effects on image quality for dual-energy CE-DBT using a CsI(Tl) phosphor-based
detector. Lag was tested using typical clinical acquisition sequences and exposure parameters and under various
detector read-out modes. The performance of a prototype anti-scatter grid and its potential benefit on the magnitude and
range of the cupping artifact were investigated. Analyses were performed through phantom experiments. Our results
illustrate that the magnitude of image lag is negligible and breast texture cancelation is almost perfect when the detector
is read out several times between x-ray exposures. The anti-scatter grid effectively reduces scatter and the cupping
artifact.
KEYWORDS: Digital breast tomosynthesis, Breast, 3D modeling, Tissues, Binary data, 3D image processing, Biopsy, 3D acquisition, Reconstruction algorithms, X-rays
Needle insertion planning for digital breast tomosynthesis (DBT) guided biopsy has the potential to improve patient
comfort and intervention safety. However, a relevant planning should take into account breast tissue deformation and
lesion displacement during the procedure. Deformable models, like finite elements, use the elastic characteristics of the
breast to evaluate the deformation of tissue during needle insertion. This paper presents a novel approach to locally
estimate the Young's modulus of the breast tissue directly from the DBT data. The method consists in computing the
fibroglandular percentage in each of the acquired DBT projection images, then reconstructing the density volume.
Finally, this density information is used to compute the mechanical parameters for each finite element of the deformable
mesh, obtaining a heterogeneous DBT based breast model. Preliminary experiments were performed to evaluate the
relevance of this method for needle path planning in DBT guided biopsy. The results show that the heterogeneous DBT
based breast model improves needle insertion simulation accuracy in 71% of the cases, compared to a homogeneous
model or a binary fat/fibroglandular tissue model.
Dual-energy imaging with the injection of an iodinated contrast medium has the potential to depict cancers in the breast, by the demonstration of tumour angiogenesis and the suppression of the breast structure noise. The present study investigates the optimum monoenergetic beam parameters for this application. First, a theoretical study of the beam parameters was performed to find the best compromise between the quality of the dualenergy
recombined image and the patient dose. The result of this analysis was then validated by phantom experiments using synchrotron monoenergetic radiation at the European Synchrotron Radiation Facility (ESRF, Grenoble, France). For an average breast of 5cm thickness and 50% glandularity, the theoretical simulations
show an optimum at 20 keV for the low energy and 34 keV for the high energy, for a CsI detector of a standard mammography system. The SDNR variations with the low energy, the high energy or the dose repartition are very similar between the measurements on images acquired with synchrotron radiation and the simulated values.
This ensures the accuracy of our theoretical optimization and the validity of the optimal beam parameters found in this study. The aim of this work is to demonstrate the potential of Dual-Energy CEDM (Contrast Enhanced Digital Mammography) with ideal monoenergetic sources, in order to provide an indicator of how to shape the
polyenergetic spectra produced with classical X-ray sources for this application.
KEYWORDS: Digital breast tomosynthesis, Medical imaging, Computer aided diagnosis and therapy, Current controlled current source, Architectural distortion, Databases, Breast
We propose a new method to detect architectural distortions and spiculated masses in digital breast tomosynthesis volumes. To achieve this goal, an a contrario approach is used. In this approach, an event, corresponding to a minimal number of structures converging toward the same location, is defined such that its expectation of occurrence within a random image is very low. Occurrences of this event in real images are then detected and considered as possible lesion locations. During the last step, the number of false positives is reduced through classification using attributes computed on histograms of structure orientations.
The approach was tested using the leave-one-out method on a database composed of 38 breasts (10 containing a lesion and 28 containing no lesion). A sensitivity of 0.8 at 1.68 false positives/breast was achieved.
KEYWORDS: Iodine, Breast, Sensors, Point spread functions, Signal attenuation, Image quality, X-rays, Digital breast tomosynthesis, Tissues, Digital mammography
Dual-Energy Contrast Enhanced Digital Breast Tomosynthesis (DE CEDBT) is a promising technique for breast
cancer detection, which combines the strengths of functional and 3D imaging. In the present study, we first
focused on the optimization of the acquisition parameters for the low and high-energy projections, which leads
to a trade-off between image quality in the recombined slices and the Average Glandular Dose (AGD) delivered
to the patient. Optimized parameters were found and experimentally validated on phantom images. Then, we
addressed the problem of iodine quantification in the recombined slices. In DE CEDBT, iodine quantification is
limited by the z-resolution, due to the restricted angle acquisition inherent to tomosynthesis. We evaluated the
lesion thickness above which determination of iodine volumetric concentration is possible. For lesions below this
thickness, estimation of iodine concentration is possible if a priori information or a model on the shape of the
lesion is available. Iodine quantification for lesions located near the breast boundary is also challenging, due to
scatter border effects and variation of the breast thickness in this region. A scatter correction algorithm based
on a deconvolution scheme and a thickness compensation algorithm were applied on the low and high-energy
projections. Corrected images showed a more accurate quantification of iodine.
In this paper, we present a fast method for microcalcification detection in Digital Breast Tomosynthesis. Instead of
applying the straight-forward reconstruction/filtering/thresholding approach, the filtering is performed on projections
before simple back-projection reconstruction. This leads to a reduced computation time since the number of projections
is generally much smaller than the number of slices. For an average breast thickness and a typical number of
projections, the number of operations is reduced by a factor in the range of 2 to 4. At the same time, the approach yields
a negligible decrease of the contrast to noise ratio in the reconstructed slices. Image segmentation results are presented
and compared to the previous method as visual performance assessment.
KEYWORDS: Iodine, Reconstruction algorithms, Breast, Digital breast tomosynthesis, Tomography, Manganese, X-rays, Sensors, Signal to noise ratio, Visibility
In this paper, we present the development of dual-energy Contrast-Enhanced Digital Breast Tomosynthesis
(CEDBT). A method to produce background clutter-free slices from a set of low and high-energy projections
is introduced, along with a scheme for the determination of the optimal low and high-energy techniques. Our
approach consists of a dual-energy recombination of the projections, with an algorithm that has proven its performance
in Contrast-Enhanced Digital Mammography1 (CEDM), followed by an iterative volume reconstruction.
The aim is to eliminate the anatomical background clutter and to reconstruct slices where the gray level is
proportional to the local iodine volumetric concentration. Optimization of the low and high-energy techniques
is performed by minimizing the total glandular dose to reach a target iodine Signal Difference to Noise Ratio
(SDNR) in the slices. In this study, we proved that this optimization could be done on the projections, by
consideration of the SDNR in the projections instead of the SDNR in the slices, and verified this with phantom
measurements. We also discuss some limitations of dual-energy CEDBT, due to the restricted angular range
for the projection views, and to the presence of scattered radiation. Experiments on textured phantoms with
iodine inserts were conducted to assess the performance of dual-energy CEDBT. Texture contrast was nearly
completely removed and the iodine signal was enhanced in the slices.
In this study, we propose a novel approach to dual-energy
contrast-enhanced digital mammography, with the development of a
dual-energy recombination algorithm based on an image chain model
and the determination of the associated optimal low and high-energy
techniques. Our method produces clutter-free iodine-equivalent
images and includes thickness correction near the breast border.
After the algorithm description, the optimal low and high-energy
acquisition techniques are determined to obtain a compromise between
image quality and glandular dose. The low and high-energy techniques
were chosen to minimize the glandular dose for a target Signal
Difference to Noise Ratio (SDNR) in the dual-energy recombined
image. The theoretical derivation of the iodine SDNR in the
recombined image allowed the prediction of the optimal low and
high-energy techniques. Depending on the breast thickness and
glandular percentage, the optimal low-energy kVp and mAs ranged from
24kVp (Mo/Mo or Mo/Rh) to 35kVp (Rh/Rh), and from 60 to 90mAs
respectively, and the high-energy kVp and mAs ranged from 40kVp to
47kVp (Mo/Cu), and from 80mAs to 290mAs. We proved the better
performance of our algorithm compared to the classic weighted
logarithmic subtraction method in terms of patient dose and also in terms of texture
cancelation, through the use of artificial textured images. Values of iodine contrast measured on phantom were
close to the expected iodine thickness. Good correlation was found
between the measured and theoretical iodine SDNR in the dual-energy
images, which validates our theoretical optimization of the
acquisition techniques.
In this paper we present a novel approach for mass detection in Digital Breast Tomosynthesis (DBT) datasets. A
reconstruction-independent approach, working directly on the projected views, is proposed. Wavelet filter responses on
the projections are thresholded and combined to obtain candidate masses. For each candidate, we create a fuzzy contour
through a multi-level thresholding process. We introduce a fuzzy set definition for the class mass contour that allows the
computation of fuzzy membership values for each candidate contour. Then, an aggregation operator is presented that
combines information over the complete set of projected views, resulting in 3D fuzzy particles. A final decision is made
taking into account all available information. The performance of the presented algorithm was evaluated on a database of 11 one-breast-cases resulting in a sensitivity (Se) of 0.86 and a false positive
rate (FPR) of 3.5 per case.
Ideally, the gray level changes in a Contrast-Enhanced Digital Mammography (CEDM) sequence reflect the uptake and wash-out of contrast medium in the breast. While insignificant in standard mammography, gray level variations with time caused both by patient and system related factors, have been observed in clinical CEDM sequences.
We have acquired phantom image series on digital mammography systems using a Mo/Cu anode-filter combination and a tube voltage between 45 and 49 kVp, in order to derive a model for gray level change with time as a function of system parameters. The gray level variation exhibits a fair degree of inter-series repeatability, and strongly depends on the dose received by the detector and timing of the image acquisition series. Moreover, for tissue-equivalent compositions, the relative gray level change with respect to the first image does not depend on the composition.
We designed a calibration procedure that can be used to compensate for the tiny system-dependent signal variation that has been observed. A global reduction of 80-93% of the variation has been demonstrated in sequences acquired on a breast shaped phantom. Local improvement is effective across the whole field of view. When imaging iodine inserts (0.5-2 mg/cm2 concentration), the calibration increases the constancy with time of iodine signal on subtracted sequences by a factor of 4 (median value).
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.