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

Convolutional neural network -based phantom image scoring for mammography quality control

BACKGROUND

Visual evaluation of phantom images is an important, but time-consuming part of mammography quality control (QC). Consistent scoring of phantom images over the device's lifetime is highly desirable. Recently, convolutional neural networks (CNNs) have been applied to a wide range of image classification problems, performing with a high accuracy. The purpose of this study was to automate mammography QC phantom scoring task by training CNN models to mimic a human reviewer.

METHODS

Eight CNN variations consisting of three to ten convolutional layers were trained for detecting targets (fibres, microcalcifications and masses) in American College of Radiology (ACR) accreditation phantom images and the results were compared with human scoring. Regular and artificially degraded/improved QC phantom images from eight mammography devices were visually evaluated by one reviewer. These images were used in training the CNN models. A separate test set consisted of daily QC images from the eight devices and separately acquired images with varying dose levels. These were scored by four reviewers and considered the ground truth for CNN performance testing.

RESULTS

Although hyper-parameter search space was limited, an optimal network depth after which additional layers resulted in decreased accuracy was identified. The highest scoring accuracy (95%) was achieved with the CNN consisting of six convolutional layers. The highest deviation between the CNN and the reviewers was found at lowest dose levels. No significant difference emerged between the visual reviews and CNN results except in case of smallest masses.

CONCLUSION

A CNN-based automatic mammography QC phantom scoring system can score phantom images in a good agreement with human reviewers, and can therefore be of benefit in mammography QC.

Optimization of the difference-of-Gaussian channel sets for the channelized Hotelling observer

The channelized-Hotelling observer (CHO) was investigated as a surrogate of human observers in task-based image quality assessment. The CHO with difference-of-Gaussian (DoG) channels has shown potential for the prediction of human detection performance in digital mammography (DM) images. However, the DoG channels employ parameters that describe the shape of each channel. The selection of these parameters influences the performance of the DoG CHO and needs further investigation. The detection performance of the DoG CHO was calculated and correlated with the detection performance of three humans who evaluated DM images in 2-alternative forced-choice experiments. A set of DM images of an anthropomorphic breast phantom with and without calcification-like signals was acquired at four different dose levels. For each dose level, 200 square regions-of-interest (ROIs) with and without signal were extracted. Signal detectability was assessed on ROI basis using the CHO with various DoG channel parameters and it was compared to that of the human observers. It was found that varying these DoG parameter values affects the correlation (r 2 ) of the CHO with human observers for the detection task investigated. In conclusion, it appears that the the optimal DoG channel sets that maximize the prediction ability of the CHO might be dependent on the type of background and signal of ROIs investigated.

Determination of contrast-detail curves in mammography image quality assessment by a parametric model observer

A novel approach is proposed for the determination of contrast-detail curves in mammography image quality assessment. The approach is compared with current practice using virtual mammography. A binary parametric model observer is applied to images of the CDMAM phantom. The observer accounts for the simple disc shaped objects in the phantom and is applied separately to each cell of the phantom. For each of these applications, the area under the ROC curve (AUC) of the model observer is determined. The different AUCs, calculated from different applications of the parametric model observer, are then combined to a single contrast-detail curve quantifying the ability of the observer to detect details in the images. Virtual mammography is developed as a tool to simulate X-ray images of single CDMAM cells and to quantitatively assess the approach in comparison with current practice. It is shown that the proposed approach can lead to similar contrast-detail curves as current practice. The precision of the estimated contrast-detail curves is increased, i.e. using 5 images yields about the same precision for the proposed approach as 16 images when applying current practice. We conclude that contrast-detail curves in mammography image quality assessment can also be determined through the AUC of a binary parametric model observer. Since the proposed approach has higher precision than current practice, it is a promising candidate for contrast-detail analysis in mammography image quality assessment.

Systematic approach to a channelized Hotelling model observer implementation for a physical phantom containing mass-like lesions: Application to digital breast tomosynthesis

PURPOSE

to develop a channelized model observer (CHO) that matches human reader (HR) scoring of a physical phantom containing breast simulating structure and mass lesion-like targets for use in quality control of digital breast tomosynthesis (DBT) imaging systems.

METHODS

A total of 108 DBT scans of the phantom was acquired using a Siemens Inspiration DBT system. The detectability of mass-like targets was evaluated by human readers using a 4-alternative forced choice (4-AFC) method. The percentage correct (PC) values were then used as the benchmark for CHO tuning, again using a 4-AFC method. Three different channel functions were considered: Gabor, Laguerre-Gauss and Difference of Gaussian. With regard to the observer template, various methods for generating the expected signal were studied along with the influence of the number of training images used to form the covariance matrix for the observer template. Impact of bias in the training process on the observer template was evaluated next, as well as HR and CHO reproducibility.

RESULTS

HR performance was most closely matched by 8 Gabor channels with tuned phase, orientation and frequency, using an observer template generated from training image data. Just 24 DBT image stacks were required to give robust CHO performance with 0% bias, although a bias of up to 33% in the training images also gave acceptable performance. CHO and HR reproducibility were similar (on average 3.2 PC versus 3.4 PC).

CONCLUSIONS

The CHO algorithm developed matches human reader performance and is therefore a promising candidate for automated readout of phantom studies.

Can a channelized Hotelling observer assess image quality in acquired mammographic images of an anthropomorphic breast phantom including image processing?

PURPOSE

To study the feasibility of a channelized Hotelling observer (CHO) to predict human observer performance in detecting calcification-like signals in mammography images of an anthropomorphic breast phantom, as part of a quality control (QC) framework.

METHODS

A prototype anthropomorphic breast phantom with inserted gold disks of 0.25 mm diameter was imaged with two different digital mammography x-ray systems at four different dose levels. Regions of interest (ROIs) were extracted from the acquired processed and unprocessed images, signal-present and signal-absent. The ROIs were evaluated by a CHO using four different formulations of the difference of Gaussian (DoG) channel sets. Three human observers scored the ROIs in a two-alternative forced-choice experiment. We compared the human and the CHO performance on the simple task to detect calcification-like disks in ROIs with and without postprocessing. The proportion of correct responses of the human reader (PC_H ) and the CHO (PC_CHO ) was calculated and the correlation between the two was analyzed using a mixed-effect regression model. To address the signal location uncertainty, the impact of shifting the DoG channel sets in all directions up to two pixels was evaluated. Correlation results including the goodness of fit (r² ) of PC_H and PC_CHO for all different parameters were evaluated.

RESULTS

Subanalysis by system yielded strong correlations between PC_H and PC_CHO , with r² between PC_H and PC_CHO was found to be between 0.926 and 0.958 for the unshifted and between 0.759 and 0.938 for the shifted channel sets, respectively. However, the linear fit suggested a slight system dependence. PC_CHO with shifted channel sets increased CHO performance but the correlation with humans was decreased. These correlations were not considerably affected by of the DoG channel set used.

CONCLUSIONS

There is potential for the CHO to be used in QC for the evaluation of detectability of calcification-like signals. The CHO can predict the PC of humans in images of calcification-like signals of two different systems. However, a global model to be used for all systems requires further investigation.

A computer-aided diagnosis scheme of breast lesion classification using GLGLM and shape features: Combined-view and multi-classifiers

PURPOSE

To address high false-positive results of FFDM issue, we make the first effort to develop a computer-aided diagnosis (CAD) scheme to analyze and distinguish breast lesions.

METHOD

The breast lesion regions were first segmented and depicted on FFDM images from 106 patients. In this work, 11 gray-level gap-length matrix texture features and 12 shape features were extracted form craniocaudal view and mediolateral oblique view, and then Student's t-test, Fisher-score and Relief-F were introduced to select features. We also investigated the effect of three factors, i.e., discretisation, selection methods and classifier methods, of the classification performance via analysis of variance. Finally, a classification model was constructed. Spearman's correlation coefficient analysis was conducted to assess the internal relevance of features.

RESULTS

The proposed scheme using Student's t-test achieved an area under the receiver operating characteristic curve (AUC) value of 0.923 at 512 bins. The AUC values are 0.884, 0.867, 0.874 and 0.901 for the low gray-level gaps emphasis (LGGE), solidity, extent, and the combined set, respectively. Solidity and extent depicts the correlation coefficient of 0.86 (P < 0.05).

CONCLUSIONS

We present a new CAD scheme based on the contribution of the significant factors. The experimental results demonstrate that the presented scheme can be used to successfully distinguish breast carcinoma lesions and benign fibroadenoma lesions in our FFDM dataset and the MIAS dataset, which may provide a CAD method to assist radiologists in diagnosing and interpreting screening mammograms. Moreover, we found that LGGE, solidity and extent features show great potential for breast lesion classification.

Optimization of the exposure parameters in digital mammography using contrast-detail metrics

PURPOSE

Optimization studies in digital mammography aid to assure the image quality and radiological protection of the patient. The aim of this work is to test effectiveness and applicability of a method based on a Figure of Merit (FOM=(IQF_inv)²/AGD) to improve all the exposure parameters (Target/Filter combination, kVp and mAs) in order to improve the image acquisition technique that will provide the best compromise between image quality and the average glandular dose (AGD).

METHODS

A contrast-detail analysis, employing the test object CDMAM, was carried out for the digital mammography unit manufactured by Lorad Hologic - model Selenia. We simulated two breast thicknesses using phantoms and a Figure of Merit as optimization tool, which includes an indicator of image quality, the IQF_inv and the average glandular dose. Images of the ACR and TORMAM phantoms were obtained with both, automatic and optimized exposure parameters. In order to compare the image quality, the SNR (Signal to Noise Ratio) was measured in each image.

RESULTS

In the two phantoms, for both 4.5 and 7.5cm thicknesses, the AGDs obtained with the optimized parameters show a reduction. In addition, the images obtained with the optimized exposure parameters, had the same or a better image quality when compared to the images obtained using the automatic mode.

CONCLUSIONS

The proposed optimization methodology proved to be an effective tool to improve the digital mammography unit, due to the use of objective metrics for evaluation and validation of the results.

Toward image quality assessment in mammography using model observers: Detection of a calcification-like object

PURPOSE

Model observers (MOs) are of interest in the field of medical imaging to assess image quality. However, before procedures using MOs can be proposed in quality control guidelines for mammography systems, we need to know whether MOs are sensitive to changes in image quality and correlations in background structure. Therefore, as a proof of principle, in this study human and model observer (MO) performance are compared for the detection of calcification-like objects using different background structures and image quality levels of unprocessed mammography images.

METHOD

Three different phantoms, homogeneous polymethyl methacrylate, BR3D slabs with swirled patterns (CIRS, Norfolk, VA, USA), and a prototype anthropomorphic breast phantom (Institute of Medical Physics and Radiation Protection, Technische Hochschule Mittelhessen, Germany) were imaged on an Amulet Innovality (FujiFilm, Tokyo, Japan) mammographic X-ray unit. Because the complexities of the structures of these three phantoms were different and not optimized to match the characteristics of real mammographic images, image processing was not applied in this study. In addition, real mammograms were acquired on the same system. Regions of interest (ROIs) were extracted from each image. In half of the ROIs, a 0.25-mm diameter disk was inserted at four different contrast levels to represent a calcification-like object. Each ROI was then modified, so four image qualities relevant for mammography were simulated. The signal-present and signal-absent ROIs were evaluated by a non-pre-whitening model observer with eye filter (NPWE) and a channelized Hotelling observer (CHO) using dense difference of Gaussian channels. The ROIs were also evaluated by human observers in a two alternative forced choice experiment. Detectability results for the human and model observer experiments were correlated using a mixed-effect regression model. Threshold disk contrasts for human and predicted human observer performance based on the NPWE MO and CHO were estimated.

RESULTS

Global trends in threshold contrast were similar for the different background structures, but absolute contrast threshold levels differed. Contrast thresholds tended to be lower in ROIs from simple phantoms compared with ROIs from real mammographic images. The correlation between human and model observer performance was not affected by the range of image quality levels studied.

CONCLUSIONS

The correlation between human and model observer performance does not depend on image quality. This is a promising outcome for the use of model observers in image quality analysis and allows for subsequent research toward the development of MO-based quality control procedures and guidelines.