Malmö Breast ImaginG database: objectives and development

Purpose

We describe the design and implementation of the Malmö Breast ImaginG (M-BIG) database, which will support research projects investigating various aspects of current and future breast cancer screening programs. Specifically, M-BIG will provide clinical data to:1.investigate the effect of breast cancer screening on breast cancer prognosis and mortality;2.develop and validate the use of artificial intelligence and machine learning in breast image interpretation; and3.develop and validate image-based radiological breast cancer risk profiles.

Approach

The M-BIG database is intended to include a wide range of digital mammography (DM) and digital breast tomosynthesis (DBT) examinations performed on women at the Mammography Clinic in Malmö, Sweden, from the introduction of DM in 2004 through 2020. Subjects may be included multiple times and for diverse reasons. The image data are linked to extensive clinical, diagnostic, and demographic data from several registries.

Results

To date, the database contains a total of 451,054 examinations from 104,791 women. During the inclusion period, 95,258 unique women were screened. A total of 19,968 examinations were performed using DBT, whereas the rest used DM.

Conclusions

We describe the design and implementation of the M-BIG database as a representative and accessible medical image database linked to various types of medical data. Work is ongoing to add features and curate the existing data.

Simulation of breast lesions based upon fractal Perlin noise

PURPOSE

Steadily increasing use of computational/virtual phantoms in medical physics has motivated expanding development of new simulation methods and data representations for modelling human anatomy. This has emphasized the need for increased realism, user control, and availability. In breast cancer research, virtual phantoms have gained an important role in evaluating and optimizing imaging systems. For this paper, we have developed an algorithm to model breast abnormalities based on fractal Perlin noise. We demonstrate and characterize the extension of this approach to simulate breast lesions of various sizes, shapes, and complexity.

MATERIALS AND METHOD

Recently, we developed an algorithm for simulating the 3D arrangement of breast anatomy based on Perlin noise. In this paper, we have expanded the method to also model soft tissue breast lesions. We simulated lesions within the size range of clinically representative breast lesions (masses, 5-20 mm in size). Simulated lesions were blended into simulated breast tissue backgrounds and visualized as virtual digital mammography images. The lesions were evaluated by observers following the BI-RADS assessment criteria.

RESULTS

Observers categorized the lesions as round, oval or irregular, with circumscribed, microlobulated, indistinct or obscured margins. The majority of the simulated lesions were considered by the observers to have a realism score of moderate to well. The simulation method provides almost real-time lesion generation (average time and standard deviation: 1.4 ± 1.0 s).

CONCLUSION

We presented a novel algorithm for computer simulation of breast lesions using Perlin noise. The algorithm enables efficient simulation of lesions, with different sizes and appearances.

Finite element model of mechanical imaging of the breast

Purpose: Malignant breast lesions can be distinguished from benign lesions by their mechanical properties. This has been utilized for mechanical imaging in which the stress distribution over the breast is measured. Mechanical imaging has shown the ability to identify benign or normal cases and to reduce the number of false positives from mammography screening. Our aim was to develop a model of mechanical imaging acquisition for simulation purposes. To that end, we simulated mammographic compression of a computer model of breast anatomy and lesions. Approach: The breast compression was modeled using the finite element method. Two finite element breast models of different sizes were used and solved using linear elastic material properties in open-source virtual clinical trial (VCT) software. A spherical lesion (15 mm in diameter) was inserted into the breasts, and both the location and stiffness of the lesion were varied extensively. The average stress over the breast and the average stress at the lesion location, as well as the relative mean pressure over lesion area (RMPA), were calculated. Results: The average stress varied 6.2-6.5 kPa over the breast surface and 7.8-11.4 kPa over the lesion, for different lesion locations and stiffnesses. These stresses correspond to an RMPA of 0.80 to 1.46. The average stress was 20% to 50% higher at the lesion location compared with the average stress over the entire breast surface. Conclusions: The average stress over the breast and the lesion location corresponded well to clinical measurements. The proposed model can be used in VCTs for evaluation and optimization of mechanical imaging screening strategies.

Development and evaluation of a method for tumor growth simulation in virtual clinical trials of breast cancer screening

Purpose: Image-based analysis of breast tumor growth rate may optimize breast cancer screening and diagnosis by suggesting optimal screening intervals and guide the clinical discussion regarding personalized screening based on tumor aggressiveness. Simulation-based virtual clinical trials (VCTs) can be used to evaluate and optimize medical imaging systems and design clinical trials. This study aimed to simulate tumor growth over multiple screening rounds. Approach: This study evaluates a preliminary method for simulating tumor growth. Clinical data on tumor volume doubling time (TVDT) was used to fit a probability distribution ("clinical fit") of TVDTs. Simulated tumors with TVDTs sampled from the clinical fit were inserted into 30 virtual breasts ("simulated cohort") and used to simulate mammograms. Based on the TVDT, two successive screening rounds were simulated for each virtual breast. TVDTs from clinical and simulated mammograms were compared. Tumor sizes in the simulated mammograms were measured by a radiologist in three repeated sessions to estimate TVDT. Results: The mean TVDT was 297 days (standard deviation, SD, 169 days) in the clinical fit and 322 days (SD, 217 days) in the simulated cohort. The mean estimated TVDT was 340 days (SD, 287 days). No significant difference was found between the estimated TVDTs from simulated mammograms and clinical TVDT values ( p > 0.5 ). No significant difference ( p > 0.05 ) was observed in the reproducibility of the tumor size measurements between the two screening rounds. Conclusions: The proposed method for tumor growth simulation has demonstrated close agreement with clinical results, supporting potential use in VCTs of temporal breast imaging.

Computer Simulations of Case Difficulty in Digital Breast Tomosynthesis Using Virtual Clinical Trials

PURPOSE

Virtual clinical trials (VCTs) require computer simulations of representative patients and images to evaluate and compare changes in performance of imaging technologies. The simulated images are usually interpreted by model observers whose performance depends upon the selection of imaging cases used in training evaluation models. This work proposes an efficient method to simulate and calibrate soft tissue lesions, which matches the detectability threshold of virtual and human readings.

METHODS

Anthropomorphic breast phantoms were used to evaluate the simulation of four mass models (I-IV) that vary in shape and composition of soft tissue. Ellipsoidal (I) and spiculated (II-IV) masses were simulated using composite voxels with partial volumes. Digital breast tomosynthesis projections and reconstructions of a clinical system were simulated. Channelized Hotelling observers (CHOs) were evaluated using reconstructed slices of masses that varied in shape, composition, and density of surrounded tissue. The detectability threshold of each mass model was evaluated using receiver operating characteristic (ROC) curves calculated with the CHO's scores.

RESULTS

The area under the curve (AUC) of each calibrated mass model were within the 95% confidence interval (mean AUC [95% CI]) reported in a previous reader study (0.93 [0.89, 0.97]). The mean AUC [95% CI] obtained were 0.94 [0.93, 0.96], 0.92 [0.90, 0.93], 0.92 [0.90, 0.94], 0.93 [0.92, 0.95] for models I to IV, respectively. The mean AUC results varied substantially as a function of shape, composition, and density of surrounded tissue.

CONCLUSIONS

For successful VCTs, lesions composed of soft tissue should be calibrated to simulate imaging cases that match the case difficulty predicted by human readers. Lesion composition, shape, and size are parameters that should be carefully selected to calibrate VCTs. This article is protected by copyright. All rights reserved.

Computational Breast Anatomy Simulation Using Multi-Scale Perlin Noise

Virtual clinical trials (VCTs) of medical imaging require realistic models of human anatomy. For VCTs in breast imaging, a multi-scale Perlin noise method is proposed to simulate anatomical structures of breast tissue in the context of an ongoing breast phantom development effort. Four Perlin noise distributions were used to replace voxels representing the tissue compartments and Cooper's ligaments in the breast phantoms. Digital mammography and tomosynthesis projections were simulated using a clinical DBT system configuration. Power-spectrum analyses and higher-order statistics properties using Laplacian fractional entropy (LFE) of the parenchymal texture are presented. These objective measures were calculated in phantom and patient images using a sample of 140 clinical mammograms and 500 phantom images. Power-law exponents were calculated using the slope of the curve fitted in the low frequency [0.1, 1.0] mm-1 region of the power spectrum. The results show that the images simulated with our prior and proposed Perlin method have similar power-law spectra when compared with clinical mammograms. The power-law exponents calculated are -3.10, -3.55, and -3.46, for the log-power spectra of patient, prior phantom and proposed phantom images, respectively. The results also indicate an improved agreement between the mean LFE estimates of Perlin-noise based phantoms and patients than our prior phantoms and patients. Thus, the proposed method improved the simulation of anatomic noise substantially compared to our prior method, showing close agreement with breast parenchyma measures.

VIRTUAL CLINICAL TRIALS IN MEDICAL IMAGING SYSTEM EVALUATION AND OPTIMISATION

Virtual clinical trials (VCTs) can be used to evaluate and optimise medical imaging systems. VCTs are based on computer simulations of human anatomy, imaging modalities and image interpretation. OpenVCT is an open-source framework for conducting VCTs of medical imaging, with a particular focus on breast imaging. The aim of this paper was to evaluate the OpenVCT framework in two tasks involving digital breast tomosynthesis (DBT). First, VCTs were used to perform a detailed comparison of virtual and clinical reading studies for the detection of lesions in digital mammography and DBT. Then, the framework was expanded to include mechanical imaging (MI) and was used to optimise the novel combination of simultaneous DBT and MI. The first experiments showed close agreement between the clinical and the virtual study, confirming that VCTs can predict changes in performance of DBT accurately. Work in simultaneous DBT and MI system has demonstrated that the system can be optimised in terms of the DBT image quality. We are currently working to expand the OpenVCT software to simulate MI acquisition more accurately and to include models of tumour growth. Based on our experience to date, we envision a future in which VCTs have an important role in medical imaging, including support for more imaging modalities, use with rare diseases and a role in training and testing artificial intelligence (AI) systems.

Under-exploration of Three-Dimensional Images Leads to Search Errors for Small Salient Targets

Advances in 3D imaging technology are transforming how radiologists search for cancer1,2 and how security officers scrutinize baggage for dangerous objects.3 These new 3D technologies often improve search over 2D images4,5 but vastly increase the image data. Here, we investigate 3D search for targets of various sizes in filtered noise and digital breast phantoms. For a Bayesian ideal observer optimally processing the filtered noise and a convolutional neural network processing the digital breast phantoms, search with 3D image stacks increases target information and improves accuracy over search with 2D images. In contrast, 3D search by humans leads to high miss rates for small targets easily detected in 2D search, but not for larger targets more visible in the visual periphery. Analyses of human eye movements, perceptual judgments, and a computational model with a foveated visual system suggest that human errors can be explained by interaction among a target's peripheral visibility, eye movement under-exploration of the 3D images, and a perceived overestimation of the explored area. Instructing observers to extend the search reduces 75% of the small target misses without increasing false positives. Results with twelve radiologists confirm that even medical professionals reading realistic breast phantoms have high miss rates for small targets in 3D search. Thus, under-exploration represents a fundamental limitation to the efficacy with which humans search in 3D image stacks and miss targets with these prevalent image technologies.

Calculation of Radiomic Features to Validate the Textural Realism of Physical Anthropomorphic Phantoms for Digital Mammography

In this paper, radiomic features are used to validate the textural realism of two anthropomorphic phantoms for digital mammography. One phantom was based off a computational breast model; it was 3D printed by CIRS (Computerized Imaging Reference Systems, Inc., Norfolk, VA) under license from the University of Pennsylvania. We investigate how the textural realism of this phantom compares against a phantom derived from an actual patient's mammogram ("Rachel", Gammex 169, Madison, WI). Images of each phantom were acquired at three kV in 1 kV increments using auto-time technique settings. Acquisitions at each technique setting were repeated twice, resulting in six images per phantom. In the raw ("FOR PROCESSING") images, 341 features were calculated; i.e., gray-level histogram, co-occurrence, run length, fractal dimension, Gabor Wavelet, local binary pattern, Laws, and co-occurrence Laws features. Features were also calculated in a negative screening population. For each feature, the middle 95% of the clinical distribution was used to evaluate the textural realism of each phantom. A feature was considered realistic if all six measurements in the phantom were within the middle 95% of the clinical distribution. Otherwise, a feature was considered unrealistic. More features were actually found to be realistic by this definition in the CIRS phantom (305 out of 341 features or 89.44%) than in the phantom derived from a specific patient's mammogram (261 out of 341 features or 76.54%). We conclude that the texture is realistic overall in both phantoms.

Evaluation of Convolutional Neural Networks for Search in 1/f 2.8 Filtered Noise and Digital Breast Tomosynthesis Phantoms

With the advent of powerful convolutional neural networks (CNNs), recent studies have extended early applications of neural networks to imaging tasks thus making CNNs a potential new tool for assessing medical image quality. Here, we compare a CNN to model observers in a search task for two possible signals (a simulated mass and a smaller simulated micro-calcification) embedded in filtered noise and single slices of Digital Breast Tomosynthesis (DBT) virtual phantoms. For the case of the filtered noise, we show how a CNN can approximate the ideal observer for a search task, achieving a statistical efficiency of 0.77 for the microcalcification and 0.78 for the mass. For search in single slices of DBT phantoms, we show that a Channelized Hotelling Observer (CHO) performance is affected detrimentally by false positives related to anatomic variations and results in detection accuracy below human observer performance. In contrast, the CNN learns to identify and discount the backgrounds, and achieves performance comparable to that of human observer and superior to model observers (Proportion Correct for the microcalcification: CNN = 0.96; Humans = 0.98; CHO = 0.84; Proportion Correct for the mass: CNN = 0.98; Humans = 0.83; CHO = 0.51). Together, our results provide an important evaluation of CNN methods by benchmarking their performance against human and model observers in complex search tasks.

Technical Note: Noise models for virtual clinical trials of digital breast tomosynthesis

PURPOSE

To investigate the use of an affine-variance noise model, with correlated quantum noise and spatially dependent quantum gain, for the simulation of noise in virtual clinical trials (VCT) of digital breast tomosynthesis (DBT).

METHODS

Two distinct technologies were considered: an amorphous-selenium (a-Se) detector with direct conversion and a thallium-doped cesium iodide (CsI(Tl)) detector with indirect conversion. A VCT framework was used to generate noise-free projections of a uniform three-dimensional simulated phantom, whose geometry and absorption match those of a polymethyl methacrylate (PMMA) uniform physical phantom. The noise model was then used to generate noisy observations from the simulated noise-free data, while two clinically available DBT units were used to acquire projections of the PMMA physical phantom. Real and simulated projections were then compared using the signal-to-noise ratio (SNR) and normalized noise power spectrum (NNPS).

RESULTS

Simulated images reported errors smaller than 4.4% and 7.0% in terms of SNR and NNPS, respectively. These errors are within the expected variation between two clinical units of the same model. The errors increase to 65.8% if uncorrelated models are adopted for the simulation of systems featuring indirect detection. The assumption of spatially independent quantum gain generates errors of 11.2%.

CONCLUSIONS

The investigated noise model can be used to accurately reproduce the noise found in clinical DBT. The assumption of uncorrelated noise may be adopted if the system features a direct detector with minimal pixel crosstalk.

Evaluation of non-Gaussian statistical properties in virtual breast phantoms

Images derived from a "virtual phantom" can be useful in characterizing the performance of imaging systems. This has driven the development of virtual breast phantoms implemented in simulation environments. In breast imaging, several such phantoms have been proposed. We analyze the non-Gaussian statistical properties from three classes of virtual breast phantoms and compare them to similar statistics from a database of breast images. These include clustered-blob lumpy backgrounds (CBLBs), truncated binary textures, and the UPenn virtual breast phantoms. We use Laplacian fractional entropy (LFE) as a measure of the non-Gaussian statistical properties of each simulation procedure. Our results show that, despite similar power spectra, the simulation approaches differ considerably in LFE with very low scores for the CBLB to high values for the UPenn phantom at certain frequencies. These results suggest that LFE may have value in developing and tuning virtual phantom simulation procedures.

Analysis of Volume Overestimation Artifacts in the Breast Outline Segmentation in Tomosynthesis

In digital breast tomosynthesis (DBT), the reconstruction is calculated from x-ray projection images acquired over a small range of angles. One step in the reconstruction process is to identify the pixels that fall outside the shadow of the breast, to segment the breast from the background (air). In each projection, rays are back-projected from these pixels to the focal spot. All voxels along these rays are identified as air. By combining these results over all projections, a breast outline can be determined for the reconstruction. This paper quantifies the accuracy of this breast segmentation strategy in DBT. In this study, a physical phantom modeling a breast under compression was analyzed with a prototype next-generation tomosynthesis (NGT) system described in previous work. Multiple wires were wrapped around the phantom. Since the wires are thin and high contrast, their exact location can be determined from the reconstruction. Breast parenchyma was portrayed outside the outline defined by the wires. Specifically, the size of the phantom was overestimated along the posteroanterior (PA) direction; i.e., perpendicular to the plane of conventional source motion. To analyze how the acquisition geometry affects the accuracy of the breast outline segmentation, a computational phantom was also simulated. The simulation identified two ways to improve the segmentation accuracy; either by increasing the angular range of source motion laterally or by increasing the range in the PA direction. The latter approach is a unique feature of the NGT design; the advantage of this approach was validated with our prototype system.

Interactions of lesion detectability and size across single-slice DBT and 3D DBT

Three dimensional image modalities introduce a new paradigm for visual search requiring visual exploration of a larger search space than 2D imaging modalities. The large number of slices in the 3D volumes and the limited reading times make it difficult for radiologists to explore thoroughly by fixating with their high resolution fovea on all regions of each slice. Thus, for 3D images, observers must rely much more on their visual periphery (points away from fixation) to process image information. We previously found a dissociation in signal detectability between 2D and 3D search tasks for small signals in synthetic textures evaluated with non-radiologist trained observers. Here, we extend our evaluation to more clinically realistic backgrounds and radiologist observers. We studied the detectability of simulated microcalcifications (MCALC) and masses (MASS) in Digital Breast Tomosynthesis (DBT) utilizing virtual breast phantoms. We compared the lesion detectability of 8 radiologists during free search in 3D DBT and a 2D single-slice DBT (center slice of the 3D DBT). Our results show that the detectability of the microcalcification degrades significantly in 3D DBT with respect to the 2D single-slice DBT. On the other hand, the detectability for masses does not show this behavior and its detectability is not significantly different. The large deterioration of the 3D detectability of microcalcifications relative to masses may be related to the peripheral processing given the high number of cases in which the microcalcification was missed and the high number of search errors. Together, the results extend previous findings with synthetic textures and highlight how search in 3D images is distinct from 2D search as a consequence of the interaction between search strategies and the visibility of signals in the visual periphery.

Method for simulating dose reduction in digital mammography using the Anscombe transformation

PURPOSE

This work proposes an accurate method for simulating dose reduction in digital mammography starting from a clinical image acquired with a standard dose.

METHODS

The method developed in this work consists of scaling a mammogram acquired at the standard radiation dose and adding signal-dependent noise. The algorithm accounts for specific issues relevant in digital mammography images, such as anisotropic noise, spatial variations in pixel gain, and the effect of dose reduction on the detective quantum efficiency. The scaling process takes into account the linearity of the system and the offset of the detector elements. The inserted noise is obtained by acquiring images of a flat-field phantom at the standard radiation dose and at the simulated dose. Using the Anscombe transformation, a relationship is created between the calculated noise mask and the scaled image, resulting in a clinical mammogram with the same noise and gray level characteristics as an image acquired at the lower-radiation dose.

RESULTS

The performance of the proposed algorithm was validated using real images acquired with an anthropomorphic breast phantom at four different doses, with five exposures for each dose and 256 nonoverlapping ROIs extracted from each image and with uniform images. The authors simulated lower-dose images and compared these with the real images. The authors evaluated the similarity between the normalized noise power spectrum (NNPS) and power spectrum (PS) of simulated images and real images acquired with the same dose. The maximum relative error was less than 2.5% for every ROI. The added noise was also evaluated by measuring the local variance in the real and simulated images. The relative average error for the local variance was smaller than 1%.

CONCLUSIONS

A new method is proposed for simulating dose reduction in clinical mammograms. In this method, the dependency between image noise and image signal is addressed using a novel application of the Anscombe transformation. NNPS, PS, and local noise metrics confirm that this method is capable of precisely simulating various dose reductions.

Description and characterization of a novel method for partial volume simulation in software breast phantoms

A modification to our previous simulation of breast anatomy is proposed to improve the quality of simulated x-ray projections images. The image quality is affected by the voxel size of the simulation. Large voxels can cause notable spatial quantization artifacts; small voxels extend the generation time and increase the memory requirements. An improvement in image quality is achievable without reducing voxel size by the simulation of partial volume averaging in which voxels containing more than one simulated tissue type are allowed. The linear x-ray attenuation coefficient of voxels is, thus, the sum of the linear attenuation coefficients weighted by the voxel subvolume occupied by each tissue type. A local planar approximation of the boundary surface is employed. In the two-material case, the partial volume in each voxel is computed by decomposition into up to four simple geometric shapes. In the three-material case, by application of the Gauss-Ostrogradsky theorem, the 3D partial volume problem is converted into one of a few simpler 2D surface area problems. We illustrate the benefits of the proposed methodology on simulated x-ray projections. An efficient encoding scheme is proposed for the type and proportion of simulated tissues in each voxel. Monte Carlo simulation was used to evaluate the quantitative error of our approximation algorithms.

Estimation of scattered radiation in digital breast tomosynthesis

Digital breast tomosynthesis (DBT) is a promising technique to overcome the tissue superposition limitations found in planar 2D x-ray mammography. However, as most DBT systems do not employ an anti-scatter grid, the levels of scattered radiation recorded within the image receptor are significantly higher than that observed in planar 2D x-ray mammography. Knowledge of this field is necessary as part of any correction scheme and for computer modelling and optimisation of this examination. Monte Carlo (MC) simulations are often used for this purpose, however they are computationally expensive and a more rapid method of calculation is desirable. This issue is addressed in this work by the development of a fast kernel-based methodology for scatter field estimation using a detailed realistic DBT geometry. Thickness-dependent scatter kernels, which were validated against the literature with a maximum discrepancy of 4% for an idealised geometry, have been calculated and a new physical parameter (air gap distance) was used to estimate more accurately the distribution of scattered radiation for a series of anthropomorphic breast phantom models. The proposed methodology considers, for the first time, the effects of scattered radiation from the compression paddle and breast support plate, which can represent more than 30% of the total scattered radiation recorded within the image receptor. The results show that the scatter field estimator can calculate scattered radiation images in an average of 80 min for projection angles up to 25° with equal to or less than a 10% error across most of the breast area when compared with direct MC simulations.

A virtual trial framework for quantifying the detectability of masses in breast tomosynthesis projection data

PURPOSE

Digital breast tomosynthesis (DBT) is a promising breast cancer screening tool that has already begun making inroads into clinical practice. However, there is ongoing debate over how to quantitatively evaluate and optimize these systems, because different definitions of image quality can lead to different optimal design strategies. Powerful and accurate tools are desired to extend our understanding of DBT system optimization and validate published design principles.

METHODS

The authors developed a virtual trial framework for task-specific DBT assessment that uses digital phantoms, open-source x-ray transport codes, and a projection-space, spatial-domain observer model for quantitative system evaluation. The authors considered evaluation of reconstruction algorithms as a separate problem and focused on the information content in the raw, unfiltered projection images. Specifically, the authors investigated the effects of scan angle and number of angular projections on detectability of a small (3 mm diameter) signal embedded in randomly-varying anatomical backgrounds. Detectability was measured by the area under the receiver-operating characteristic curve (AUC). Experiments were repeated for three test cases where the detectability-limiting factor was anatomical variability, quantum noise, or electronic noise. The authors also juxtaposed the virtual trial framework with other published studies to illustrate its advantages and disadvantages.

RESULTS

The large number of variables in a virtual DBT study make it difficult to directly compare different authors' results, so each result must be interpreted within the context of the specific virtual trial framework. The following results apply to 25% density phantoms with 5.15 cm compressed thickness and 500 μm(3) voxels (larger 500 μm(2) detector pixels were used to avoid voxel-edge artifacts): 1. For raw, unfiltered projection images in the anatomical-variability-limited regime, AUC appeared to remain constant or increase slightly with scan angle. 2. In the same regime, when the authors fixed the scan angle, AUC increased asymptotically with the number of projections. The threshold number of projections for asymptotic AUC performance depended on the scan angle. In the quantum- and electronic-noise dominant regimes, AUC behaviors as a function of scan angle and number of projections sometimes differed from the anatomy-limited regime. For example, with a fixed scan angle, AUC generally decreased with the number of projections in the electronic-noise dominant regime. These results are intended to demonstrate the capabilities of the virtual trial framework, not to be used as optimization rules for DBT.

CONCLUSIONS

The authors have demonstrated a novel simulation framework and tools for evaluating DBT systems in an objective, task-specific manner. This framework facilitates further investigation of image quality tradeoffs in DBT.

Analyzing tree-shape anatomical structures using topological descriptors of branching and ensemble of classifiers

The analysis of anatomical tree-shape structures visualized in medical images provides insight into the relationship between tree topology and pathology of the corresponding organs. In this paper, we propose three methods to extract descriptive features of the branching topology; the asymmetry index, the encoding of branching patterns using a node labeling scheme and an extension of the Sholl analysis. Based on these descriptors, we present classification schemes for tree topologies with respect to the underlying pathology. Moreover, we present a classifier ensemble approach which combines the predictions of the individual classifiers to optimize the classification accuracy. We applied the proposed methodology to a dataset of x-ray galactograms, medical images which visualize the breast ductal tree, in order to recognize images with radiological findings regarding breast cancer. The experimental results demonstrate the effectiveness of the proposed framework compared to state-of-the-art techniques suggesting that the proposed descriptors provide more valuable information regarding the topological patterns of ductal trees and indicating the potential of facilitating early breast cancer diagnosis.

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.

Optimized generation of high resolution breast anthropomorphic software phantoms

PURPOSE

The authors present an efficient method for generating anthropomorphic software breast phantoms with high spatial resolution. Employing the same region growing principles as in their previous algorithm for breast anatomy simulation, the present method has been optimized for computational complexity to allow for fast generation of the large number of phantoms required in virtual clinical trials of breast imaging.

METHODS

The new breast anatomy simulation method performs a direct calculation of the Cooper's ligaments (i.e., the borders between simulated adipose compartments). The calculation corresponds to quadratic decision boundaries of a maximum a posteriori classifier. The method is multiscale due to the use of octree-based recursive partitioning of the phantom volume. The method also provides user-control of the thickness of the simulated Cooper's ligaments and skin.

RESULTS

Using the proposed method, the authors have generated phantoms with voxel size in the range of (25-1000 μm)(3)∕voxel. The power regression of the simulation time as a function of the reciprocal voxel size yielded a log-log slope of 1.95 (compared to a slope of 4.53 of our previous region growing algorithm).

CONCLUSIONS

A new algorithm for computer simulation of breast anatomy has been proposed that allows for fast generation of high resolution anthropomorphic software phantoms.

Development and characterization of an anthropomorphic breast software phantom based upon region-growing algorithm

PURPOSE

We present a novel algorithm for computer simulation of breast anatomy for generation of anthropomorphic software breast phantoms. A realistic breast simulation is necessary for preclinical validation of volumetric imaging modalities.

METHODS

The anthropomorphic software breast phantom simulates the skin, regions of adipose and fibroglandular tissue, and the matrix of Cooper's ligaments and adipose compartments. The adipose compartments are simulated using a seeded region-growing algorithm; compartments are grown from a set of seed points with specific orientation and growing speed. The resulting adipose compartments vary in shape and size similar to real breasts; the adipose region has a compact coverage by adipose compartments of various sizes, while the fibroglandular region has fewer, more widely separated adipose compartments. Simulation parameters can be selected to cover the breadth of variations in breast anatomy observed clinically.

RESULTS

When simulating breasts of the same glandularity with different numbers of adipose compartments, the average compartment volume was proportional to the phantom size and inversely proportional to the number of simulated compartments. The use of the software phantom in clinical image simulation is illustrated by synthetic digital breast tomosynthesis images of the phantom. The proposed phantom design was capable of simulating breasts of different size, glandularity, and adipose compartment distribution. The region-growing approach allowed us to simulate adipose compartments with various size and shape. Qualitatively, simulated x-ray projections of the phantoms, generated using the proposed algorithm, have a more realistic appearance compared to previous versions of the phantom.

CONCLUSIONS

A new algorithm for computer simulation of breast anatomy has been proposed that improved the realism of the anthropomorphic software breast phantom.

Analysis of parenchymal texture with digital breast tomosynthesis: comparison with digital mammography and implications for cancer risk assessment

PURPOSE

To correlate the parenchymal texture features at digital breast tomosynthesis (DBT) and digital mammography with breast percent density (PD), an established breast cancer risk factor, in a screening population of women.

MATERIALS AND METHODS

This HIPAA-compliant study was approved by the institutional review board. Bilateral DBT images and digital mammograms from 71 women (mean age, 54 years; age range, 34-75 years) with negative or benign findings at screening mammography were retrospectively collected from a separate institutional review board-approved DBT screening trial (performed from July 2007 to March 2008) in which all women had given written informed consent. Parenchymal texture features of skewness, coarseness, contrast, energy, homogeneity, and fractal dimension were computed from the retroareolar region. Principal component analysis (PCA) was applied to obtain orthogonal texture components. Mammographic PD was estimated with software. Correlation analysis and multiple linear regression with generalized estimating equations were performed to determine the association between texture features and breast PD. Regression was adjusted for age to determine the independent association of texture to breast PD when age was also considered as a predictor variable.

RESULTS

Texture feature correlations to breast PD were stronger with DBT than with digital mammography. Statistically significant correlations (P < .001) were observed for contrast (r = 0.48), energy (r = -0.47), and homogeneity (r = -0.56) at DBT and for contrast (r = 0.26), energy (r = -0.26), and homogeneity (r = -0.33) at digital mammography. Multiple linear regression analysis of PCA texture components as predictors of PD also demonstrated significantly stronger associations with DBT. The association was strongest when age was also considered as a predictor of PD (R² = 0.41 for DBT and 0.28 for digital mammography; P < .001).

CONCLUSION

Parenchymal texture features are more strongly correlated to breast PD in DBT than in digital mammography. The authors' long-term hypothesis is that parenchymal texture analysis with DBT will result in quantitative imaging biomarkers that can improve the estimation of breast cancer risk.

Breast percent density: estimation on digital mammograms and central tomosynthesis projections

PURPOSE

To evaluate inter- and intrareader agreement in breast percent density (PD) estimation on clinical digital mammograms and central digital breast tomosynthesis (DBT) projection images.

MATERIALS AND METHODS

This HIPAA-compliant study had institutional review board approval; all patients provided informed consent. Breast PD estimation was performed on the basis of anonymized digital mammograms and central DBT projections in 39 women (mean age, 51 years; range, 31-80 years). All women had recently detected abnormalities or biopsy-proved cancers. PD was estimated by three experienced readers on the mediolateral oblique views of the contralateral breasts by using software; each reader repeated the estimation after 2 months. Spearman correlations of inter- and intrareader and intermodality PD estimates, as well as kappa statistics between categoric PD estimates, were computed.

RESULTS

High correlation (rho = 0.91) was observed between PD estimates on digital mammograms and those on central DBT projections, averaged over all estimations; the corresponding kappa coefficient (0.79) indicated substantial agreement. Mean interreader agreement for PD estimation on central DBT projections (rho = 0.85 +/- 0.05 [standard deviation]) was significantly higher (P < .01) than that for PD estimation on digital mammograms (rho = 0.75 +/- 0.05); the corresponding kappa coefficients indicated substantial (kappa = 0.65 +/- 0.12) and moderate (kappa = 0.55 +/- 0.14) agreement for central DBT projections and digital mammograms, respectively.

CONCLUSION

High correlation between PD estimates on digital mammograms and those on central DBT projections suggests the latter could be used until a method for PD estimation based on three-dimensional reconstructed images is introduced. Moreover, clinical PD estimation is possible with reduced radiation dose, as each DBT projection was acquired by using about 22% of the dose for a single mammographic projection.

A representation and classification scheme for tree-like structures in medical images: analyzing the branching pattern of ductal trees in X-ray galactograms

We propose a multistep approach for representing and classifying tree-like structures in medical images. Tree-like structures are frequently encountered in biomedical contexts; examples are the bronchial system, the vascular topology, and the breast ductal network. We use tree encoding techniques, such as the depth-first string encoding and the PrUfer encoding, to obtain a symbolic string representation of the tree's branching topology; the problem of classifying trees is then reduced to string classification. We use the tf-idf text mining technique to assign a weight of significance to each string term (i.e., tree node label). Similarity searches and k-nearest neighbor classification of the trees is performed using the tf-idf weight vectors and the cosine similarity metric. We applied our approach to characterize the ductal tree-like parenchymal structure in X-ray galactograms, in order to distinguish among different radiological findings. Experimental results demonstrate the effectiveness of the proposed approach with classification accuracy reaching up to 86%, and also indicate that our method can potentially aid in providing insight to the relationship between branching patterns and function or pathology.

Parenchymal texture analysis in digital breast tomosynthesis for breast cancer risk estimation: a preliminary study

RATIONALE AND OBJECTIVES

Studies have demonstrated a relationship between mammographic parenchymal texture and breast cancer risk. Although promising, texture analysis in mammograms is limited by tissue superposition. Digital breast tomosynthesis (DBT) is a novel tomographic x-ray breast imaging modality that alleviates the effect of tissue superposition, offering superior parenchymal texture visualization compared to mammography. The aim of this study was to investigate the potential advantages of DBT parenchymal texture analysis for breast cancer risk estimation.

MATERIALS AND METHODS

DBT and digital mammographic (DM) images of 39 women were analyzed. Texture features, shown in previous studies with mammograms to correlate with cancer risk, were computed from the retroareolar breast region. The relative performances of the DBT and DM texture features were compared in correlating with two measures of breast cancer risk: (1) the Gail and Claus risk estimates and (2) mammographic breast density. Linear regression was performed to model the association between texture features and increasing levels of risk.

RESULTS

No significant correlation was detected between parenchymal texture and the Gail and Claus risk estimates. Significant correlations were observed between texture features and breast density. Overall, the DBT texture features demonstrated stronger correlations with breast percent density than DM features (P < or = .05). When dividing the study population into groups of increasing breast percent density, the DBT texture features appeared to be more discriminative, having regression lines with overall lower P values, steeper slopes, and higher R(2) estimates.

CONCLUSION

Although preliminary, the results of this study suggest that DBT parenchymal texture analysis could provide more accurate characterization of breast density patterns, which could ultimately improve breast cancer risk estimation.

Breast dosimetry using high-resolution voxel phantoms

A computer model of X-ray mammography has been developed, which uses quasi-realistic high-resolution voxel phantoms to simulate the breast. The phantoms have 400 microm voxels and simulate the three-dimensional distributions of adipose and fibro-glandular tissues, Cooper's ligaments, ducts and skin and allow the estimation of dose to individual tissues. Calculations of the incident air kerma to mean glandular dose conversion factor, g, were made using a Mo/Mo spectrum at 28 kV for eight phantoms in the thickness range 40-80 mm and of varying glandularity. The values differed from standard tabulations used for breast dosimetry by up to 43%, because of the different spatial distribution of glandular tissue within the breast. To study this further, additional voxel phantoms were constructed, which gave variations of between 9 and 59% compared with standard values. For accurate breast dosimetry, it is therefore very important to take the distribution of glandular tissues into account.

Calculation of the properties of digital mammograms using a computer simulation

A Monte Carlo computer model of mammography has been developed to study and optimise the performance of digital mammographic systems. The program uses high-resolution voxel phantoms to model the breast, which simulate the adipose and fibroglandular tissues, Cooper's ligaments, ducts and skin in three dimensions. The model calculates the dose to each tissue, and also the quantities such as energy imparted to image pixels, noise per image pixel and scatter-to-primary (S/P) ratios. It allows studies of the dependence of image properties on breast structure and on position within the image. The program has been calibrated by calculating and measuring the pixel values and noise for a digital mammographic system. The thicknesses of two components of this system were unknown, and were adjusted to obtain a good agreement between measurement and calculation. The utility of the program is demonstrated with the calculations of the variation of the S/P ratio with and without a grid, and of the image contrast across the image of a 50-mm-thick breast phantom.

Evaluation of a novel method of noise reduction using computer-simulated mammograms

A novel method of noise reduction has been tested for mammography using computer-simulated images for which the truth is known exactly. This method is based on comparing two images. The images are compared at different scales, using a cross-correlation function as a measure of similarity to define the image modifications in the wavelet domain. The computer-simulated images were calculated for noise-free primary radiation using a quasi-realistic voxel phantom. Two images corresponding to slightly different geometry were produced. Gaussian noise was added with certain properties to simulate quantum noise. The added noise could be reduced by >70% using the proposed method without any noticeable corruption of the structures. It is possible to save 50% dose in mammography by producing two images (each 25% of the dose for a standard mammogram). Additionally, a reduction of the anatomical noise and, therefore, better detection rates of breast cancer in mammography are possible.

Effects of quantum noise and binocular summation on dose requirements in stereoradiography

In the case of a quantum-noise limited detector, signal detection theory suggests that stereoradiographic images can be acquired with one half of the per-image dose needed for a standard radiographic projection, as information from the two stereo images can be combined. Previously, film-screen stereoradiography has been performed using the same per-image dose as in projection radiography, i.e., doubling the total dose. In this paper, the assumption of a possible decrease in dose for stereoradiography was tested by a series of contrast-detail experiments, using phantom images acquired over a range of exposures. The number of visible details, the effective reduction of the dose, and the effective decrease in the threshold signal-to-noise ratio were determined using human observers under several display and viewing conditions. These results were averaged over five observers and compared with multiple readings by a single observer and with the results of an additional observer with limited stereoscopic acuity. Experimental results show that the total dose needed to produce a stereoradiographic image pair is approximately 1.1 times the dose needed for a single projection in standard radiography, indicating that under these conditions the human visual system demonstrates almost ideal binocular summation.

Mammogram synthesis using a three-dimensional simulation. III. Modeling and evaluation of the breast ductal network

A method is proposed for realistic simulation of the breast ductal network as part of a computer three-dimensional (3-D) breast phantom. The ductal network is simulated using tree models. Synthetic trees are generated based upon a description of ductal branching by ramification matrices (R matrices), whose elements represent the probabilities of branching at various levels of a tree. We simulated the ductal network of the breast, consisting of multiple lobes, by random binary trees (RBT). Each lobe extends from the ampulla and consists of branching ductal segments of decreasing size, and the associated terminal ductal-lobular units. The lobes follow curved paths that project from the nipple toward the chest wall. We have evaluated the RBT model by comparing manually-traced ductal networks from 25 projections of ductal lobes in clinical galactograms and manually-traced networks from 23 projections of synthetic RBTs. A root-mean-square (rms) fractional error of 41%, between the R-matrix elements corresponding to clinical and synthetic images, was computed. This difference was influenced by projection and segmentation artifacts and by the limited number of available images. In addition, we analyzed 23 synthetic trees generated using R matrices computed from clinical images. A comparison of these synthetic and clinical images yielded a rms fractional error of 11%, suggesting the possibility that a more appropriate model of the ductal branching morphology may be developed. Rejection of the RBT model also suggests the existence of a relationship between ductal branching morphology and the state of mammary developmentand pathology.

Classification of galactograms with ramification matrices: preliminary results

RATIONALE AND OBJECTIVES

The poor specificity of galactography, the imaging modality generally indicated in cases of nipple discharge, has led to a large number of biopsies with negative results. A quantitative scheme for classifying galactographic findings might help reduce the number of such biopsies in the future. As a first step toward that goal, the authors have studied one quantitative method for describing the branching of ducts by using ramification matrices (R matrices), and the correlation of the values of the matrix elements with clinical findings.

MATERIALS AND METHODS

The ductal trees were manually segmented for 25 galactographic views from 15 patients, and corresponding R matrices were calculated. Patients were divided into two groups: those with no reported galactographic findings (NF) and those with reported findings (RF) of ductal ectasia, cysts, or papilloma. In a leave-one-out fashion, the authors evaluated a classification scheme that was based on R-matrix coefficients and used a Bayesian decision rule. The effects of segmentation were tested by successively removing each of the terminal ducts and computing the corresponding matrices of the pruned trees.

RESULTS

With use of a single R-matrix element, 92% and 62% of NF and RF cases were correctly classified, respectively (P = .007). With use of two elements, 83% and 77% of NF and RF cases were correctly classified, but this result was not statistically significant (P = .108). In a test of robustness, an analysis of pruned trees yielded an average root-mean-square fractional difference of 9.7% between the elements of the original and the R matrix averaged over all pruned trees.

CONCLUSION

The preliminary analysis indicates that it may be possible to identify cases with reported galactographic findings by using R matrices.

Mammogram synthesis using a 3D simulation. I. Breast tissue model and image acquisition simulation

A method is proposed for generating synthetic mammograms based upon simulations of breast tissue and the mammographic imaging process. A computer breast model has been designed with a realistic distribution of large and medium scale tissue structures. Parameters controlling the size and placement of simulated structures (adipose compartments and ducts) provide a method for consistently modeling images of the same simulated breast with modified position or acquisition parameters. The mammographic imaging process is simulated using a compression model and a model of the x-ray image acquisition process. The compression model estimates breast deformation using tissue elasticity parameters found in the literature and clinical force values. The synthetic mammograms were generated by a mammogram acquisition model using a monoenergetic parallel beam approximation applied to the synthetically compressed breast phantom.

Mammogram synthesis using a 3D simulation. II. Evaluation of synthetic mammogram texture

We have evaluated a method for synthesizing mammograms by comparing the texture of clinical and synthetic mammograms. The synthesis algorithm is based upon simulations of breast tissue and the mammographic imaging process. Mammogram texture was synthesized by projections of simulated adipose tissue compartments. It was hypothesized that the synthetic and clinical texture have similar properties, assuming that the mammogram texture reflects the 3D tissue distribution. The size of the projected compartments was computed by mathematical morphology. The texture energy and fractal dimension were also computed and analyzed in terms of the distribution of texture features within four different tissue regions in clinical and synthetic mammograms. Comparison of the cumulative distributions of the mean features computed from 95 mammograms showed that the synthetic images simulate the mean features of the texture of clinical mammograms. Correlation of clinical and synthetic texture feature histograms, averaged over all images, showed that the synthetic images can simulate the range of features seen over a large group of mammograms. The best agreement with clinical texture was achieved for simulated compartments with radii of 4-13.3 mm in predominantly adipose tissue regions, and radii of 2.7-5.33 and 1.3-2.7 mm in retroareolar and dense fibroglandular tissue regions, respectively.

Aliasing effects in digital images of line-pair phantoms

Line-pair phantoms are commonly used for evaluating screen-film systems. When imaged digitally, aliasing effects give rise to additional periodic patterns. This paper examines one such effect that medical physicists are likely to encounter, and which can be used as an indicator of super-resolution.