The new techniques of three-dimensional visualization of the breast that have been developed in recent years have the potential to markedly improve diagnostic breast imaging. Many experts even expect the new 3D techniques such as digital breast tomosynthesis (DBT) or breast CT to completely replace “conventional” digital mammography in the intermediate term as they have definitive advantages over 2D mammography. Various approaches are theoretically conceivable and have to be evaluated in the clinical setting, including the use of different reconstruction algorithms, the use of different DBT angles, slit-scan DBT vs. flat-panel DBT vs. breast CT, different dose settings, and DBT in one plane vs. two planes. Finally, DBT might also have advantages in combination with other tools for improving the diagnostic accuracy of breast imaging such as computer-aided diagnosis and contrast mammography.
Digital mammography overcomes several technical limitations associated with screen-film mammog-raphy. An essential feature of digital mammography is that both the intensity and the spatial distribution of the X-ray transmission pattern are sampled to form the image. In the spatial domain, the interval between samples (pitch) and the response profile of the detector element (del) largely determine the spatial resolution of the imaging system. The dynamic range of the detector and the number of bits used to digitize the image determine the ability to image all parts of the breast with acceptable contrast and signal-to-noise ratio. Depending on the system design, it is possible to eliminate much of the structural or fixed-pattern noise associated with the detector and the X-ray beam to approach a quantum noise limited situation. In digital mam-mography systems, it is often possible to design detectors that allow efficient use of the incident X-rays without excessive loss of spatial resolution. This permits a substantial reduction in the radiation dose to the breast when compared with film mammography without sacrifice of image quality. Because of the differences in technology, the optimum exposure conditions may shift toward the use of higher energy spectra than would be used with film, particularly for dense or thick breasts.
Softcopy reading of mammographic images has many aspects in common with traditional screen-film viewing but also many aspects that are very different. Viewing images is the core diagnostic task, and one can consider it from two perspectives. On the one hand, it is the technology used to display the images and how technical factors such as luminance and display noise affect the quality of the image and, hence, the perception and interpretation of features in that image. On the other hand, there are the human observers relying on their perceptual and cognitive systems to process the information presented to them to render a diagnostic decision. Therefore, to maintain low error rates, careful consideration of both sides of the reading equation is required to optimize the interpretation of softcopy mammographic images.
Computer-aided detection (CADe) and computer-aided diagnosis (CADx) are emerging technologies to help radiologists interpret medical images. In screening mammography, CADe can help radiologists avoid overlooking a cancer, while CADx can help radiologists decide whether a biopsy is warranted when reading a diagnostic mammogram. Even though there is much commonality in the techniques used in CADe and CADx algorithms, there are important differences in the input data and in the output of the algorithms. In particular, CADe outputs the location of potential cancers, while CADx outputs the likelihood that a known lesion is malignant. These differences affect the metrics used to evaluate their performance. Commercial CADe systems have been developed and clinical studies of CADe have indicated the ability to increase radiologists' sensitivity by approximately 10% with a comparable increase in the recall rate. Commercial CADx systems do not exist till date, but observer study results are very compelling. CADe and CADx schemes continue to evolve in terms of accuracy and user interface. It is expected that CADe and eventually CADx will play an increasingly important role in breast imaging in the future.
The transition from a film-based to a completely digital organization cannot be accomplished by merely replacing individual components. In contrast, the whole flow of operations needs reengineering. This affects not only the way individual physicians or radiographers interact with workstations or imaging modalities, but also the overall process of medical cooperation.The individual components of PACS, RIS, and other information systems must be integrated to a higher degree than one may expect. This requires new strategic decisions regarding the deployment of information systems and regarding the relationship of the radiology department to the technology infrastructure.Information technology opens up completely new applications such as telemammography. Merely introducing technology will not provide the expected results, as concepts and technology fundamentally depend on the organization of telecooperation.
A number of studies have demonstrated that radiologists can become fatigued after reading cases for a number of hours Some of these studies have demonstrated that there are associated decreases in observer performance (i.e., reduced accuracy), but mammographic reading has not been a focus of any of these studies Based on these recent findings regarding decreases in performance as a function of time of day and/or number of hours reading, this retrospective study examined data from a variety of mammography studies in which readers participated in two sessions – once in the morning and once in the afternoon The ROC Az data from these studies were compared for statistical differences between morning and afternoon reading Overall there was a small yet significant (t = 2.365, p = 0.0277) between morning and afternoon diagnostic performance, with performance being degraded in the afternoon These data suggest that reader fatigue may impact mammography interpretation performance, although more formal studies are required to verify these findings with a prospective study since this retrospective analysis did have limitations.
There have been two major clinical trials of digital mammography in North America. The Colorado-Massachusetts trial was groundbreaking in its design, as the first trial to test the modalities head-to-head, and to consider findings detected by each modality equally. This trial showed a significant decrease in the recall rate for digital and a nonsignificant trend for film in increased cancer detection. As an early trial, it was limited by technical factors that would be improved shortly after the trial. The most apparent of these was the digital workstation used for interpretation. The DMIST trial built on the Colorado-Massachusetts trial and expanded on it with markedly larger numbers of subjects and institutions. It also looked at machines from multiple vendors. This trial found a significant advantage for digital in cancer detection rate and overall performance, as measured by ROC analysis, for young women with dense breasts. For the entire cohort, however, there was no significant difference between the modalities.
The purpose of this paper is to compare subregional breast density and whole breast density and their association with breast cancer risk. The film mammograms of 278 cases and 834 age and ethnicity-matched controls were digitized and analyzed using single-energy x-ray absorptiometry (SXA). The subregion was a 3-cm diameter circle centered in the breast. The whole and subregional densities are found to be highly correlated (r2=0.7). The 4:1 quartile odds ratio after controlling for other significant risk factors (age, BMI, family history and age at first live birth) was 3.6 (95 CI 2.1-5.4) and the 2.4 (95 CI 1.5-3.7) for the whole and subregional breast density, respectively. Further studies are underway to optimize the placement of the ROI and combined multiple regions.
An effective quality control system for digital mammography needs to evaluate the status of each stage of image formation — acquisition, processing and display. Such quality control benefits greatly from the ability to make more precise and reproducible measurements than was possible with film-screen systems. On the other hand, the greater variety of system designs and general lack of experience with different digital systems has complicated the introduction of quality control (QC) procedures. Those with extensive experience of QC in digital mammography have stressed the importance of checking regularly for artifacts in images of uniform test blocks for the early detection of any problems arising in the image acquisition stage, e.g. the detector. Although the tests for the subsequent stages of image processing and display are less well developed, they are of considerable importance and will be the focus of further work. Digital technology makes possible the automation of routine QC procedures and a method of doing this is described.
Digital breast tomosynthesis (DBT) is an imaging modality in which tomographic sections of the breast are generated from a limited range of x-ray tube angles One drawback of DBT is resolution loss in the oblique projection images The purpose of this work is to extend Swank's formulation of the transfer functions of turbid granular phosphors to oblique x-ray incidence, using the diffusion approximation to the Boltzmann equation to model the spread of light in the phosphor As expected, the modulation transfer function (MTF) and noise power spectra (NPS) are found to decrease with projection angle regardless of frequency By contrast, the dependence of detective quantum efficiency (DQE) on projection angle is frequency dependent DQE increases with projection angle at low frequencies, and only decreases with projection angle at high frequencies Importantly, the x-ray quantum detection efficiency (AQ) and the Swank information factor (AS) are also found to be angularly dependent.
The performance of a new double layer amorphous Selenium detector for digital mammography is compared to the FDA approved CR system for digital mammography in a two step study. A study comparing radiation between both systems is first done to obtain the best settings for a clinical study. In a second step the results in terms of quality are evaluated by three readers comparing the final images. A minimal reduction of dose (20%) is obtained and a better definition of glandular structures is demonstrated with the new device.
The use of digital mammography systems has become widespread recently However, the optimal exposure parameters are uncertain in clinical practice We need to optimize the exposure parameter in digital mammography while maximizing image quality and minimizing patient dose The purpose of this study was to evaluate the most beneficial exposure variable—tube voltage for each compressed breast thickness—with these indices: noise power spectrum, noise equivalent quanta, detective quantum efficiency, and signal-to-noise ratios (SNR) In this study, the SNRs were derived from the perceived statistical decision theory model with the internal noise of eye-brain system (SNRi), contrived and studied by Loo LN [1], Ishida M et al [2] These image quality indices were obtained under a fixed average glandular dose (AGD) and a fixed image contrast Our results indicated that when the image contrast and AGD was constant, for phantom thinner than 5 cm, an increase of the tube voltage did not improve the noise property of images very much The results also showed that image property with the target/filter Mo/Rh was better than that with Mo/Mo for phantom thicker than 4 cm In general, it is said that high tube voltage delivers improved noise property Our result indicates that this common theory is not realized with the x-ray energy level for mammography.
Comparing the clinical performance of digital mammography technologies is challenging. The aim of this work is to develop and test a methodology for adjusting mammographic images taken on a given imaging system to simulate their appearance as if taken on a different system. Such methodology would be very useful for a wide range of system performance and design studies using both phantom and clinical images. The process involves changing the image blurring in accordance with the measured modulation transfer functions and adding noise (electronic, quantum and structure). The method has been tested by adapting flat field images acquired using an amorphous selenium detector and a computed radiography (CR) detector to different dose levels and comparing the resultant simulated NPSs with directly measured NPSs. For the detectors used in this work the NPSs at different dose levels are well predicted. This could be a powerful tool for studies of clinical image quality.
This paper deals with learning spiculation scores of masses in a supervised manner. Three spiculation score prediction models treating the score either as a continuous or ordinary variable are presented. These models were compared on a data-set of 255 masses.
A common metric used to optimise digital mammography image acquisition is contrast-to-noise ratio. Using the standard attenuation rate (SAR), a quantitative normalised representation of breast tissue for image analysis applications, we demonstrate that the image contrast may be completely separated from the acquisition parameters, in particular the beam quality, used for acquisition. Optimising the contrast-to-noise ratio at acquisition is therefore suboptimal, since the contrast may be manipulated by post processing. A tissue equivalent phantom is used to investigate the variation in both signal-to-noise ratio, and image sharpness within the SAR images. The results show that the primary effect of varying the acquisition parameters through the various automated optimisation of parameter modes, and hence the mean glandular dose, is to vary the global contrast of the acquired image, an effect successfully mapped to a common normalised basis using the SAR. The signal-to-noise ratio and image sharpness are second order effects, and are therefore dominated by the global image contrast when image acquisition is optimised using the contrast-to-noise ratio.
A computer-based training tool was developed through a collaborative design process. The tool allows trainee radiologists to access a large number of suspicious lesions. The tool employs a certainty-based scoring system in which trainees’ responses are scored not just as right or wrong but according to their confidence. Different approaches to providing trainees with feedback were considered: one based on a histogram and one using a line graph of cumulative scores. Following an initial assessment by radiologists, a revised scheme was introduced in which disagreements between trainee and expert are rated according the clinical or pedagogical significance of the error.
In the investigation of emerging tomographic breast imaging methods using flat-panel detectors (FPDs), digital breast object models are useful tools. These models are also commonly referred to as digital phantoms. We have created an ensemble of digital breast object phantoms based on CT scans of surgical mastectomy specimens. Early versions of the phantoms have been used in our published research. Recently we have improved some of our processing tools such as the use of 3-D anisotropic diffusion filtering (ADF) prior to segmentation, and we have evaluated breast object models generated with different methods including power spectral analysis, ROI statistics and an observation study.
Digital breast tomosynthesis (DBT) is being investigated to overcome the obscuring effect of overlapping breast tissue in projection mammography. To quantify the effectiveness of DBT in reducing overlapping breast structures, it is important to investigate how breast structural noise propagates during the reconstruction process. Others have found that breast structure may be characterized as power law noise of the form κ/ f β . We investigate how the power law exponent, β, varies as a function of reconstruction methods. Clinical DBT data sets were used to analyze breast structural noise in both projection and reconstructed domains using different filter schemes of a filtered back projection (FBP) reconstruction algorithm. The dependence on filter settings was compared with cascaded linear system theory. The goal this work is to combine frequency domain analysis of breast structural noise with previous work on quantum noise in DBT and develop a generalized framework to optimize DBT for breast lesion detection.
Quality assurance in digital mammography, especially in screening environments, needs a lot thoughtfulness. As the images are digital, it provides the possibility of the use of computer assisted tests. This requires adequate computer readable phantoms and powerful software tools, which should support the medical technical assistant (MTA) in carrying out the required constancy tests. An added value of software assisted quality assurance is the ability to generate reports and statistical evaluations of the collected results. The national mammography screening program in Luxembourg has established a sophisticated quality assurance program including automatic reading, ready to use reporting and tailored formation at a national scope during the last years. This paper reports our experiences during the implementation of this common national wide quality control system and discuss the challenges involved.
