Several clinical image databases are currently available to support scientific research in the medical field. These images are generally used to validate studies based on measuring the sensitivity and specificity of a particular clinical task. In the case of digital mammography, the radiation dose directly influences the quality of the image and consequently the performance of radiologists. Therefore, it is important to conduct studies to find a balance between image quality and radiation dose. Image processing methods are typically employed to optimize this relationship. For the evaluation of these methods, it is crucial to have a mammographic image database with specific characteristics, currently unavailable for scientific use. For example, this image database should contain sets of images from the same patient acquired at different radiation doses with breast lesions in known locations. This is achievable using computational methods for noise and microcalcification insertion into pre-acquired clinical images. In this context, the present work aims to present a cloud-based application for on-demand generation of a clinical mammographic image database with different radiation doses and breast lesions. From a set of pre-acquired clinical digital mammograms, it is possible to create N databases with different characteristics. This technique can also be considered as data augmentation.
In digital mammography, the physics of the acquisition system and post-processing algorithms can cause image noise to be spatially correlated. Noise correlation is characterized by non-constant noise power spectral density and can negatively affect image quality. Although the literature explores ways to quantify the frequency dependence of noise in digital mammography, there is still a lack of studies that explore the effect of this phenomenon on clinical tasks. Thus, the aim of this work is to evaluate the impact of noise correlation on the quality of digital mammography and the detectability of lesions using a virtual clinical trial (VCT) tool. Considering the radiographic factors of a standard full-dose acquisition, VCT was used to generate two sets of images: one containing mammograms corrupted with correlated noise and the other with uncorrelated (white) noise. Clusters of five to seven microcalcifications of different sizes and shapes were computationally inserted into the images at regions of dense tissue. We then designed a human observer study to investigate performance on a clinical task of locating microcalcifications on digital mammography from both image sets. In addition, nine objective image quality metrics were calculated on mammograms. The results obtained with four medical physicists showed that the average performance in localization was 72% for images with correlated noise and 95% with uncorrelated noise. Thus, our results suggest that correlated noise promotes a greater reduction in the conspicuity of subtle microcalcifications than uncorrelated noise. Furthermore, only four of the nine objective quality metrics calculated in this work were consistent with the results of the human observer study, highlighting the importance of using appropriate metrics to assess the quality of corrupted images with correlated noise. The source code for our framework is publicly available at https://github.com/LAVI-USP/SPIE2023-ImageQuality.
It is well-known that x-ray systems featuring indirect detectors are affected by noise spatial correlation. In the case of digital breast tomosynthesis (DBT), this phenomenon might affect the perception of small details in the image, such as microcalcifications. In this work, we propose the use of a deep convolutional neural network (CNN) to restore DBT projections degraded with correlated noise using the framework of a cycle generative adversarial network (cycle-GAN). To generate pairs of images for the training procedure, we used a virtual clinical trial (VCT) system. Two approaches were evaluated: in the first one, the network was trained to perform noise decorrelation by changing the frequency-dependency of the noise in the input image, but keeping the other characteristics. In the second approach, the network was trained to perform denoising and decorrelation, with the objective of generating an image with frequency-independent (white) noise and with characteristics equivalent to an acquisition with a radiation exposure four times greater than the input image. We tested the network with virtual and clinical images and we found that in both training approaches the model successfully corrected the power spectrum of the input images.
KEYWORDS: Iterative methods, Interference (communication), Digital breast tomosynthesis, Calibration, Systems modeling, Denoising, X-rays, Transform theory, Data analysis
The majority of the denoising algorithms available in the literature are designed to treat signal-independent Gaussian noise. However, in digital breast tomosynthesis (DBT) systems, the noise model seldom presents signal-independence. In this scenario, variance-stabilizing transforms (VSTs) may be used to convert the signaldependent noise into approximately signal-independent noise, enabling the use of ‘off-the-shelf’ denoising techniques. The accurate stabilization of the noise variance requires a robust estimation of the system’s noise coefficients, usually obtained using calibration data. However, practical issues often arise when calibration data are required, impairing the clinical deployment of algorithms that rely on variance stabilization. In this work, we present a practical method to achieve variance stabilization by approaching it as an optimization task, with the stabilized noise variance dictating the cost function. An iterative method is used to implicitly optimize the coefficients used in the variance stabilization, leveraging a single set of raw DBT projections. The variance stabilization achieved using the proposed method is compared against the stabilization achieved using noise coefficients estimated from calibration data, considering two commercially available DBT systems and a prototype DBT system. The results showed that the average error for variance stabilization achieved using the proposed method is comparable to the error achieved through calibration data. Thus, the proposed method can be a viable alternative for achieving variance stabilization when calibration data are not easily accessible, facilitating the clinical deployment of algorithms that rely on variance stabilization.
KEYWORDS: Denoising, Image processing, Digital mammography, Mammography, Image quality, Signal to noise ratio, Cancer, Breast imaging, Signal processing, Interference (communication)
Noise negatively impacts the detection and characterization of lesions in mammography. While denoising filters may be used to suppress noise, they might also negatively affect the conspicuity of small lesions due to signal blurring and smearing. In previous works, we designed and validated a denoising pipeline, dedicated to mammography, capable of suppressing noise and avoiding excessive blur and smear. This is achieved by a fine-tuned noisy-denoised image blending step, which leverages a Poisson-Gaussian noise model. In the current work, we investigate the impact of the denoising pipeline on the localization of low contrast microcalcification clusters. To this end, a human observers study was conducted with a team of five medical physicists with experience in breast imaging. First, in the pilot study, we defined the limit of contrast for the localization task with and without the application of the denoising pipeline. Next, we investigated the effect of the denoising on the localization of microcalcification clusters. Clinical patient cases with dense breasts and simulated microcalcification clusters were used throughout this study to emulate challenging cases and to guarantee fine control over the lesion’s contrast. The results from six readers show that the limit of localization occurred at the contrasts 0.090 and 0.079 without and with denoising, respectively. The average correct localization rate was 77% and 81% without and with denoising, respectively. Thus, the results show that the readers were able to correctly locate significantly less conspicuous lesions (p<0.05), and also performed significantly better localizing microcalcification clusters (p<0.05) when the denoising pipeline was applied.
KEYWORDS: Digital breast tomosynthesis, Signal to noise ratio, Sensors, Reconstruction algorithms, Image processing, Breast, X-rays, Systems modeling, X-ray detectors, Fluctuations and noise
In this work, we investigated and measured the noise in Digital Breast Tomosynthesis (DBT) slices considering the back-projection (BP) algorithm for image reconstruction. First, we presented our open-source DBT reconstruction toolbox and validated with a freely available virtual clinical trials (VCT) software, comparing our results with the reconstruction toolbox available at the Food and Drug Administration's (FDA) repository. A virtual anthropomorphic breast phantom was generated in the VCT environment and noise-free DBT projections were simulated. Slices were reconstructed by both toolboxes and objective metrics were measured to evaluate the performance of our in-house reconstruction software. For the noise analysis, commercial DBT systems from two vendors were used to obtain x-ray projections of a uniform polymethyl methacrylate (PMMA) physical phantom. One system featured an indirect thallium activated cesium iodide (CsI(TI)) scintillator detector and the other a direct amorphous selenium (a-Se) detector. Our in-house software was used to reconstruct raw projections into tomographic slices, and the mean pixel value, noise variance, signal-to-noise ratio (SNR) and the normalized noise power spectrum (NNPS) were measured. In addition, we investigated the adequacy of a heteroskedastic Gaussian model, with an affine variance function, to describe the noise in the reconstruction domain. The measurements show that the variance and SNR from reconstructed slices report similar spatial and signal dependency from previously reported in the projection domain. NNPS showed that the reconstruction process correlates the noise of the DBT slices in the case of projections degraded with almost uncorrelated noise.
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.