Beverly A. Lau

A statistically defined anthropomorphic software breast phantom

PURPOSE Digital anthropomorphic breast phantoms have emerged in the past decade because of recent advances in 3D breast x-ray imaging techniques. Computer phantoms in the literature have incorporated power-law noise to represent glandular tissue and branching structures to represent linear components such as ducts. When power-law noise is added to those phantoms in one piece, the simulated fibroglandular tissue is distributed randomly throughout the breast, resulting in dense tissue placement that may not be observed in a real breast. The authors describe a method for enhancing an existing digital anthropomorphic breast phantom by adding binarized power-law noise to a limited area of the breast. METHODS Phantoms with (0.5 mm)(3) voxel size were generated using software developed by Bakic et al. Between 0% and 40% of adipose compartments in each phantom were replaced with binarized power-law noise (β = 3.0) ranging from 0.1 to 0.6 volumetric glandular fraction. The phantoms were compressed to 7.5 cm thickness, then blurred using a 3 × 3 boxcar kernel and up-sampled to (0.1 mm)(3) voxel size using trilinear interpolation. Following interpolation, the phantoms were adjusted for volumetric glandular fraction using global thresholding. Monoenergetic phantom projections were created, including quantum noise and simulated detector blur. Texture was quantified in the simulated projections using power-spectrum analysis to estimate the power-law exponent β from 25.6 × 25.6 mm(2) regions of interest. RESULTS Phantoms were generated with total volumetric glandular fraction ranging from 3% to 24%. Values for β (averaged per projection view) were found to be between 2.67 and 3.73. Thus, the range of textures of the simulated breasts covers the textures observed in clinical images. CONCLUSIONS Using these new techniques, digital anthropomorphic breast phantoms can be generated with a variety of glandular fractions and patterns. β values for this new phantom are comparable with published values for breast tissue in x-ray projection modalities. The combination of conspicuous linear structures and binarized power-law noise added to a limited area of the phantom qualitatively improves its realism.

An anthropomorphic software breast phantom for tomosynthesis simulation: power spectrum analysis of phantom projections

We have performed spectral analysis of simulated tomosynthesis projections generated using an anthropomorphic software breast phantom Twenty phantoms were generated: ten 450 ml phantoms with 40% glandular fraction and ten 1500 ml phantoms with 20% glandular fraction Simulated mammographic compression and acquisition was performed using monoenergetic ray-tracing ROIs were extracted and the modulus-squared 2D FFT was applied to each ROI to obtain periodograms Radially-averaged periodograms were compared between phantom and clinical images We observed a good agreement between the spectral power law exponents (β) calculated from phantom projections and clinical images.

Effect of Scan Angle and Reconstruction Algorithm on Model Observer Performance in Tomosynthesis

The goal of this work is to develop task-based figures of merit for the assessment of tomosynthesis imaging. Towards this aim, we have computed performance of a prewhitening model observer for a SKE detection task in uniform background, in the in-focus plane of a the signal in the reconstructed tomosynthesis volume, as well as signal contrast. Test images were computer generated and reconstructed using both iterative ML-EM and filtered backprojection. We found that in noiseless images, signal contrast was substantially higher for the FBP reconstruction. For ML-EM reconstruction, model observer performance decreased with iteration number. At 10 iterations, performance was similar to that of FBP reconstruction. We found an increase of model observer performance with scan angle in the ML-EM reconstructed images. FBP showed a slight decrease of observer performance with scan angle. The results of this study are limited by the task, which is not realistic because backgrounds do not contain structure noise. This will be addressed in the future.

TH‐D‐201B‐08: An Anthropomorphic Software Breast Phantom for Tomosynthesis Simulation: Power Spectrum Analysis of Phantom Reconstructions

Purpose: To validate an anthropomorphic software breast phantom for simulation studies by analyzing breast structure in projection and reconstructed images.Method and Materials: Twenty computer breast phantoms were generated: ten 450 ml phantoms with 40% glandular tissue compressed to 5‐cm thickness and ten 1500 ml phantoms with 20% glandular tissue compressed to 7.5‐cm thickness. Monoenergetic ray tracing was done to create synthetic projection view images and maximum likelihood expectation maximization (MLEM) with 8 iterations was used to reconstruct the images. Regions of interest (ROIs) were extracted and the squared modulus 2D FFT was taken to obtain periodograms for each ROI. The radial average of each periodogram was taken and the power‐law exponent of this approximated 1D power spectrum is β a description of the amount of structure in the ROI. This was repeated for both projection view images and reconstructed slices and the average β was calculated for each projection view and for each reconstructed slice. Results: For these twenty phantoms we found that in the reconstructed images were lower than the values in the projection view images. For the 450 ml phantoms was measured to be 3.09 (σ = 0.25) in the projection and 2.82 (0.48) in the reconstruction. For the 1500 ml phantoms was measured to be 2.86 (0.43) in the projection and 2.73 (0.66) in the reconstruction. Published data show that decreases from 3.06 to 2.87 in clinical tomosynthesisimages.Conclusion: We found that the changes in between projection views and tomosynthesisreconstructed slices are comparable for the anthropomorphic phantom and clinical breast images. This supports the fact that this phantom can provide realistic breast texture. Conflict of Interest: Research sponsored by Hologic Inc. and Dexela Ltd.

A directional small-scale tissue model for an anthropomorphic breast phantom

Mammographic tissue structure has been shown to exhibit directionality, with a preferred orientation towards the nipple. However, this property is absent in the small-scale tissue model of current breast phantoms. To improve existing breast phantoms, a model for simulating oriented breast tissue has been developed, and has been included into an existing anthropomorphic breast phantom. Within this model, directionality was introduced by filling compartments with binarized power-law noise that was oriented towards the nipple. Mammograms were simulated based on the original and the new directional phantom. Tissue orientation was measured in the simulated mammogram. Visually, the appearance of the enhanced phantom was more realistic. Further, the distribution of the orientation measure computed from the enhanced phantom was more similar to that in actual mammograms. In conclusion, the use of a directional model to simulate fibroglandular tissue greatly improves the realism of the breast phantom.

Effect of non-isotropic detector blur on microcalcification detectability in tomosynthesis

We have investigated the effect of non-isotropic blur in an indirect x-ray conversion screen in tomosynthesis imaging. To study this effect, we have implemented a screen model for angle-dependent x-ray incidence, and have validated the model using experimental as well as Monte-Carlo simulations reported in the literature. We investigated detector characteristics such as MTF, NPS and DQE, and we estimated system performance in a signal-known exactly detection task. We found that for such a screen, the frequency dependence of the MTF varies with x-ray source angle, while the frequency dependence of the NPS does not. Furthermore, as the x-ray source angle is increased, the DQE becomes more narrow and DQE(f=0) grows. We found that for a tomosynthesis scan angle of 90 degrees and a conversion screen thickness of 130 microns, detectability for small signals (radius=0.125 mm) was decreased by 13%, compared to signal radii above 0.5 mm. The magnitude of the degradation is expected to vary for different tomosynthesis configurations, such as scan angle and conversion screen thickness.

Algorithmic scatter correction in dual-energy digital mammography for calcification imaging

X-ray scatter leads to erroneous calculations of dual-energy digital mammography (DEDM). The purpose of this work is to design an algorithmic method for scatter correction in DEDM without extra exposures or lead sheet. The method was developed based on the knowledge that scatter radiation in mammograms varies slowly spatially and most pixels in mammograms are non-calcification pixels, and implemented on a commercial full-field digital mammography system with a phantom of breast tissue equivalent material. The pinhole-array interpolation scatter correction method was also implemented on the system. We compared the background dual-energy (DE) calcification signals in the DE calcification images. Results show that the background signal in the DE calcification image can be reduced. The rms of background DE calcification image signal of 1105μm with scatter-uncorrected data was reduced to 187μm and 253μm after scatter correction, using our algorithmic method and pinhole-array interpolation method, respectively. The range of background DE calcification signals using scatter-uncorrected data was reduced by ~80% with scatter-corrected data using algorithmic method. The proposed algorithmic scatter correction method is effective; it has similar or even better performance than pinhole-array interpolation method in scatter correction for DEDM.

Issues in characterizing anatomic structure in digital breast tomosynthesis

Normal mammographic backgrounds have power spectra that can be described using a power law P(f) = c/fβ, where β ranges from 1.5 to 4.5. Anatomic noise can be the dominant noise source in a radiograph. Many researchers are characterizing anatomic noise by β, which can be measured from an image. We investigated the effect of sampling distance, offset, and region of interest (ROI) size on β. We calculated β for tomosynthesis projection view and reconstructed images, and we found that ROI size affects the value of β. We evaluated four different square ROI sizes (1.28, 2.56, 3.2, and 5.12 cm), and we found that the larger ROI sizes yielded larger β values in the projection images. The β values change rapidly across a single projection view; however, despite the variation across the breast, different sampling schemes (which include a variety of sampling distances and offsets) produced average β values with less than 5% variation. The particular location and number of samples used to calculate β does not matter as long as the whole image is covered, but the size of the ROI must be chosen carefully.

TH‐C‐332‐09: The Effect of Variable Exposure Distribution On Microcalcification Detectability in Tomosynthesis

Purpose: To investigate the fundamental limitation of tomosynthesis acquisition on the detectability of microcalcifications (MCs) and to investigate the effects of employing unequal dose distribution (variable exposure) across the tomosynthesis projections. Method and Materials:Ray tracing was used through a 5‐cm thick homogeneous slab phantom, which contained 150‐, 280‐, and 400‐micron diameter spheres embedded at the center. In this study we accounted for acquisition geometry and x‐ray quantum noise. Detectability was calculated using a nonprewhitening observer. In the sinogram data, detectability was computed by integrating observer response over all projection views. In the reconstructed image, observer response was computed using a 2D matched template at the in‐focus slice of the sphere. To test variable distribution of exposure, 50% of the exposure was concentrated in the center three projection views, and the rest was divided equally among the remaining projection views. Results: We determined that the detectability of MCs is reduced for larger projection angles because of the increased pathlength through the phantom and larger source‐to‐detector distance. The spheres are projected on the detector as ellipses at larger angles, which blurs the MCs and decreases detectability. Detectability of MCs using the variable exposure method is approximately 10–20% higher than detectability in the equal exposure distribution method. Conclusion: The detectability of MC is reduced in a tomosynthesis acquisition because of the acquisition geometry if all projections are made with the same exposure. A variable exposure method can improve MC detectability in DBT systems. In future work we will examine the effect of breast structure noise on the variable exposure method. Conflict of Interest: Research sponsored by Hologic, Inc. and Dexela Ltd.

Microcalcification detectability in tomosynthesis

Microcalcifications (MCs) are an important early sign of breast cancer. In conventional mammography, MC detectability is limited primarily due to quantum noise. In tomosynthesis, a dose comparable to that delivered in one projection mammogram is divided across a number of projection views (typically ranging between 10 and 30). This potentially will reduce the detectability of MCs, if detector noise is not very low. The purpose of this study is to explore the relationship between MC detectability in the projection views and in the reconstructed image. The effect of angular range and number of angles on detectability will also be evaluated for an ideal detector. Microcalcification detectability is shown to be greater in the sinogram than in the reconstructed images. Further, the detectability is reduced when the MC is located far from the center of the breast. Also, the detectability in the projection images is dependent on the projection angle.