Ramona W. Bouwman

An alternative method for noise analysis using pixel variance as part of quality control procedures on digital mammography systems

According to the European Guidelines for quality assured breast cancer screening and diagnosis, noise analysis is one of the measurements that needs to be performed as part of quality control procedures on digital mammography systems. However, the method recommended in the European Guidelines does not discriminate sufficiently between systems with and without additional noise besides quantum noise. This paper attempts to give an alternative and relatively simple method for noise analysis which can divide noise into electronic noise, structured noise and quantum noise. Quantum noise needs to be the dominant noise source in clinical images for optimal performance of a digital mammography system, and therefore the amount of electronic and structured noise should be minimal. For several digital mammography systems, the noise was separated into components based on the measured pixel value, standard deviation (SD) of the image and the detector entrance dose. The results showed that differences between systems exist. Our findings confirm that the proposed method is able to discriminate systems based on their noise performance and is able to detect possible quality problems. Therefore, we suggest to replace the current method for noise analysis as described in the European Guidelines by the alternative method described in this paper.

Average glandular dose in digital mammography and digital breast tomosynthesis: comparison of phantom and patient data

For the evaluation of the average glandular dose (AGD) in digital mammography (DM) and digital breast tomosynthesis (DBT) phantoms simulating standard model breasts are used. These phantoms consist of slabs of polymethyl methacrylate (PMMA) or a combination of PMMA and polyethylene (PE). In the last decades the automatic exposure control (AEC) increased in complexity and became more sensitive to (local) differences in breast composition. The question is how well the AGD estimated using these simple dosimetry phantoms agrees with the average patient AGD. In this study the AGDs for both dosimetry phantoms and for patients have been evaluated for 5 different x-ray systems in DM and DBT modes. It was found that the ratios between patient and phantom AGD did not differ considerably using both dosimetry phantoms. These ratios averaged over all breast thicknesses were 1.14 and 1.15 for the PMMA and PMMA-PE dosimetry phantoms respectively in DM mode and 1.00 and 1.02 in the DBT mode. These ratios were deemed to be sufficiently close to unity to be suitable for dosimetry evaluation in quality control procedures. However care should be taken when comparing systems for DM and DBT since depending on the AEC operation, ratios for particular breast thicknesses may differ substantially (0.83-1.96). Although the predictions of both phantoms are similar we advise the use of PMMA  +  PE slabs for both DM and DBT to harmonize dosimetry protocols and avoid any potential issues with the use of spacers with the PMMA phantoms.

Can the non-pre-whitening model observer, including aspects of the human visual system, predict human observer performance in mammography?

PURPOSE In mammography, images are processed prior to display. Current methodologies based on physical image quality measurements are however not designed for the evaluation of processed images. Model observers (MO) might be suitable for this evaluation. The aim of this study was to investigate whether the non-pre-whitening (NPW) MO can be used to predict human observer performance in mammography-like images by including different aspects of the human visual system (HVS). METHODS The correlation between human and NPW MO performance has been investigated for the detection of disk shaped objects in simulated white noise (WN) and clustered lumpy backgrounds (CLB), representing quantum noise limited and mammography-like images respectively. The images were scored by the MO and five human observers in a 2-alternative forced choice experiment. RESULTS For WN images it was found that the log likelihood ratio (RLR2), which expresses the goodness of fit, was highest (0.44) for the NPW MO without addition of HVS aspects. For CLB the RLR2 improved from 0.46 to 0.65 with addition of HVS aspects. The correlation was affected by object size and background. CONCLUSIONS This study shows that by including aspects of the HVS, the performance of the NPW MO can be improved to better predict human observer performance. This demonstrates that the NPW MO has potential for image quality assessment. However, due to the dependencies found in the correlation, the NPW MO can only be used for image quality assessment for a limited range of object sizes and background variability.

Daily quality control for breast tomosynthesis

Breast tomosynthesis is an imaging modality that recently became available for breast examination. For conventional projection mammography quality control procedures are well described. For breast tomosynthesis, on the other hand, such procedures have not yet been established. In this paper we propose a simple method and phantom for daily quality control (DQC). With DQC image quality problems arising after acceptance of the system should be detected. Therefore, the DQC procedure needs to monitor the stability of the most critical components of the system over time. For breast tomosynthesis we assume that the most critical items are the image receptor, X-ray tube and the tomosynthesis motion. In the proposed procedure the image receptor homogeneity and system stability are evaluated using an image of a homogeneous block of PMMA. The z-resolution is assumed to be dependent on the tomosynthesis motion. To monitor this motion the nominal z-resolution using the slice sensitive profile is measured. Shading artefacts that arise due to objects with high attenuation are also typical for tomosynthesis systems. Analysing those artefacts may provide additional information about the tomosynthesis motion. The proposed DQC procedure has been evaluated on two different breast tomosynthesis systems: A multi slit scanning system and a system using a stationary a-Se detector. Preliminary results indicate that the proposed method is useful for DQC, although some minor changes to the phantoms are advised. To verify that this method detects image quality problems sufficiently, more experience with different DBT systems, over longer periods of time are needed.

Image Quality Measurements in Breast Tomosynthesis

Breast tomosynthesis systems are developed by a number of manufacturers. It is desirable to have methods for image quality evaluation that can be used for quality control and minimum performance. In this study methods of measurement of three aspects of image quality (homogeneity, artifacts and detail detection on a homogeneous background) are evaluated on a prototype tomosynthesis system. The homogeneity evaluation showed that pixel value and standard deviation varied by up to respectively 6% and 30% in the tomosynthesis image. Artifacts have been quantified using an aluminium sphere of 2 mm diameter. In the plane of the object higher pixel value artefacts were visible. In all other planes artefacts were also visible. The detail detection did not fulfil the limiting values of threshold contrast for projection mammography, but image quality might have been underestimated. Homogeneity, detail detection and artifacts should be evaluated in future quality control procedures for tomosynthesis.

Noise analysis of full field digital mammography systems

In digital mammography noise characteristics are measured in quality control procedures. In the European Guidelines a method of measurement to investigate noise in digital mammography systems was proposed to evaluate the presence of additional noise beside quantum noise. However this method of noise analysis does not discriminate sufficiently between systems with and without additional noise. Therefore a different noise analysis is proposed. In this analysis the noise of a digital system is subdivided into three components: electronic, quantum and structured noise and the noise dose dependency of these components is studied. The usefulness of this analysis in both the frequency and spatial domain is investigated on a number of DR and CR systems. The results show that large differences between digital mammography systems exists. Some systems do have a large range in detector dose for which quantum noise is the largest noise component. For one system however, electronic and structured noise are more dominant. In addition to the differences between systems smaller differences in noise characteristics exist between different target-filter combinations on a particular system. These differences might be attributed to the limited flatfield calibration, the heel effect and difference in sensitivity. The noise analysis in both the frequency and spatial domain give useful information about the noise characteristics of systems. The analysis in the spatial domain is relatively easy to perform and to interpret. This analysis might be suitable for QC purposes. The analysis in the frequency domain does give additional information and might be used for thorough investigations.

New reconstruction algorithm for digital breast tomosynthesis: better image quality for humans and computers:

Background The image quality of digital breast tomosynthesis (DBT) volumes depends greatly on the reconstruction algorithm. Purpose To compare two DBT reconstruction algorithms used by the Siemens Mammomat Inspiration system, filtered back projection (FBP), and FBP with iterative optimizations (EMPIRE), using qualitative analysis by human readers and detection performance of machine learning algorithms. Material and Methods Visual grading analysis was performed by four readers specialized in breast imaging who scored 100 cases reconstructed with both algorithms (70 lesions). Scoring (5-point scale: 1 = poor to 5 = excellent quality) was performed on presence of noise and artifacts, visualization of skin-line and Cooper’s ligaments, contrast, and image quality, and, when present, lesion visibility. In parallel, a three-dimensional deep-learning convolutional neural network (3D-CNN) was trained (n = 259 patients, 51 positives with BI-RADS 3, 4, or 5 calcifications) and tested (n = 46 patients, nine positives), separately with FBP and EMPIRE volumes, to discriminate between samples with and without calcifications. The partial area under the receiver operating characteristic curve (pAUC) of each 3D-CNN was used for comparison. Results EMPIRE reconstructions showed better contrast (3.23 vs. 3.10, P = 0.010), image quality (3.22 vs. 3.03, P < 0.001), visibility of calcifications (3.53 vs. 3.37, P = 0.053, significant for one reader), and fewer artifacts (3.26 vs. 2.97, P < 0.001). The 3D-CNN-EMPIRE had better performance than 3D-CNN-FBP (pAUC-EMPIRE = 0.880 vs. pAUC-FBP = 0.857; P < 0.001). Conclusion The new algorithm provides DBT volumes with better contrast and image quality, fewer artifacts, and improved visibility of calcifications for human observers, as well as improved detection performance with deep-learning algorithms.

A model observer study using acquired mammographic images of an anthropomorphic breast phantom

PURPOSE To study the feasibility of a task-based framework composed of an anthropomorphic breast phantom and mathematical model observers (MOs) for the evaluation of system-processed mammographic images. METHODS A prototype anthropomorphic breast phantom with inserted gold discs of 0.1 mm and 0.25 mm diameter was imaged with two digital mammography systems (system A and B) at four different dose levels. From the acquired processed and unprocessed images, signal-present and signal-absent regions of interest (ROIs) were extracted. The ROIs were evaluated by a non-pre-whitening MO with eye filter (NPWE) and by three human observers in a two-alternative forced-choice experiment. We compared the human and the MO performance on a simple detection task of the calcification-like discs in ROIs with and without postprocessing. Proportion of correct responses of the human (PCH ) and NPWE (PCNPWE ) experiments was calculated and the correlation between the two was analyzed using a mixed-effect regression model. Correlation results including the goodness of fit (r2 ) of PCH and PCNPWE for all different parameters investigated were evaluated to determine whether NPWE MO can be used to predict human observer performance. RESULTS PCH and PCNPWE increased with dose for all conditions investigated (signal size, processing status, and different system). In case of the 0.1 mm discs, for system A, r2 between PCH with PCNPWE was 0.81. For system B, r2 was 0.93. In case of the 0.25 mm discs, r2 in system A was 0.79 and for system B, r2 was 0.82. For the combined parameters investigated, and after excluding the 0.1 mm discs on system A because the results were influenced by aliasing, the overall r2 was 0.81. Image processing did not affect the detectability of calcification-like signals. No significant difference (P > 0.05) was found between the predicted PCH(pred) by the MO and the PCH for all different conditions. CONCLUSIONS The framework seems promising to be used in objective image quality assessment. It was found to be relatively robust for the range of parameters investigated. However, further optimization of the anthropomorphic breast phantom and investigation of other MOs for a broader range of image quality assessment tasks is needed.

Signal template generation from acquired mammographic images for the non-prewhitening model observer with eye-filter

Model observers (MOs) are being investigated for image quality assessment in full-field digital mammography (FFDM). Signal templates for the non-prewhitening MO with eye filter (NPWE) were formed using acquired FFDM images. A signal template was generated from acquired images by averaging multiple exposures resulting in a low noise signal template. Noise elimination while preserving the signal was investigated and a methodology which results in a noise-free template is proposed. In order to deal with signal location uncertainty, template shifting was implemented. The procedure to generate the template was evaluated on images of an anthropomorphic breast phantom containing microcalcification-related signals. Optimal reduction of the background noise was achieved without changing the signal. Based on a validation study in simulated images, the difference (bias) in MO performance from the ground truth signal was calculated and found to be <1%. As template generation is a building stone of the entire image quality assessment framework, the proposed method to construct templates from acquired images facilitates the use of the NPWE MO in acquired images.

Can the channelized Hotelling observer including aspects of the human visual system predict human observer performance in mammography

PURPOSE In mammography, images are processed prior to display. Model observers (MO) are candidates to objectively evaluate processed images if they can predict human observer performance for detail detection. The aim of this study was to investigate if the channelized Hotelling observer (CHO) can be configured to predict human observer performance in mammography like images. METHODS The performance correlation between human observers and CHO has been evaluated using different channel-sets and by including aspects of the human visual system (HVS). The correlation was investigated for the detection of disk-shaped details in simulated white noise (WN) and clustered lumpy backgrounds (CLB) images, representing respectively quantum noise limited and mammography like images. The images were scored by the MO and five human observers in 2-alternative forced choice experiments. RESULTS For WN images the most useful formulation of the CHO to predict human observer performance was obtained using three difference of Gaussian channels without adding HVS aspects (RLR2=0.62). For CLB images the most useful formulation was the partial least square channel-set without adding HVS aspects (RLR2=0.71). The correlation was affected by detail size and background. CONCLUSIONS This study has shown that the CHO can predict human observer performance. Due to object size and background dependency it is important that the range of object sizes and allowed variability in background are specified and validated carefully before the CHO can be implemented for objective image quality assessment.