Digital mammography (DM) and digital breast tomosynthesis, the gold standards for breast cancer screening, requires correct breast positioning to ensure accuracy. Improper positioning can result in missed cancers, or can lead to additional imaging. We propose an automated deep learning (DL) segmentation approach to perform multi-class identification of regions of interest (ROI) commonly used for identification of poor positioning in mediolateral oblique (MLO) breast views. We hypothesize that by leveraging the capabilities of DL through the use of the well-founded U-Net model architecture, multi-class DL-based segmentation approaches can accurately identify air, parenchyma, pectoralis, and nipple locations within MLO images. In this study, we employed model hyperparameter searches to determine optimal model parameters for our proposed DL architecture, including the optimal loss function configuration; our best model achieved an average Sørensen-Dice coefficient of 0.919 ± 0.061 on the held-out test set. We identified high levels of localization performance in the nipple ROI. We believe our proposed segmentation model can be a foundational step in further mammogram analysis, such as for breast positioning and localized image processing tools.
Noiseless digital mammograms (DM) are unobtainable in clinical screening environments, limiting the development of deep learning-based (DL) denoising applications. Virtual clinical trials (VCTs) allow the precise simulation of noise levels in DM images for controlled training of DL models. We evaluated a set of DL denoising models, trained using VCT data, that showcases the trade-offs between denoising strength and fine structure preservation. Our results show that metrics, such as peak signal-to-noise ratio (PSNR), are improved with the use of our trained residual convolutional neural network. This quantifiable improvement indicates that our proposed DL methodology can accurately denoise simulated mammograms.
Specimen x-ray imaging provides important information on the margin of surgically excised tissue as well as radiologic and pathologic correlation of the lesion. Similar to breast imaging, where mammograms are digitally processed to enhance readability and lesion conspicuity, specimen images are also processed and enhanced. However, specimen image processing is made challenging by the diversity of specimen containers that are commercially available, compounded by variations in specimen size. In this work, we demonstrate our specimen container and size classification system based on a simple convolutional neural network (CNN), trained to identify the container type. This system allows for automated image processing of the supported container types. A dataset consisting of 1428 HIPAA and IRB-complaint anonymized specimen images were collected. We prepared a simple CNN for image classification with 3 convolutional and 3 fully connected layers, and evaluated the performance based on three comparison metrics. Each network was analyzed in terms of accuracy, multi-class AUC, and via a confusion matrix. The best performing classifier, determined via cross validation, was then used for testing, and evaluated with the same three metrics. The results of training and tuning within cross validation showed that the specimen classes are easily differentiable with this simple convolutional neural network structure. During testing, the network was able to achieve an accuracy of 95.8±4.0%, and an AUC of 0.9763±0.0001.
Computed super-resolution (SR) is a method of reconstructing images with pixels that are smaller than the detector element size; superior spatial resolution is achieved through the elimination of aliasing and alteration of the sampling function imposed by the reconstructed pixel aperture. By comparison, magnification mammography is a method of projection imaging that uses geometric magnification to increase spatial resolution. This study explores the development and application of magnification digital breast tomosynthesis (MDBT). Four different acquisition geometries are compared in terms of various image metrics. High-contrast spatial resolution was measured in various axes using a lead star pattern. A modified Defrise phantom was used to determine the low-frequency spatial resolution. An anthropomorphic phantom was used to simulate clinical imaging. Each experiment was conducted at three different magnifications: contact (1.04x), MAG1 (1.3x), and MAG2 (1.6x). All images were taken on our next generation tomosynthesis system, an in-house solution designed to optimize SR. It is demonstrated that both computed SR and MDBT (MAG1 and MAG2) provide improved spatial resolution over non-SR contact imaging. To achieve the highest resolution, SR and MDBT should be combined. However, MDBT is adversely affected by patient motion at higher magnifications. In addition, MDBT requires more radiation dose and delays diagnosis, since MDBT would be conducted upon recall. By comparison, SR can be conducted with the original screening data. In conclusion, this study demonstrates that computed SR and MDBT are both viable methods of imaging the breast.
A method for geometric calibration of a next-generation tomosynthesis (NGT) system is proposed and tested. The NGT system incorporates additional geometric movements between projections over conventional DBT. These movements require precise geometric calibration to support magnification DBT and isotropic SR. A phantom was created to project small tungsten-carbide ball bearings (BB’s) onto the detector at four different magnifications. Using a bandpass filter and template matching, a MATLAB program was written to identify the centroid locations of each BB projection on the images. An optimization algorithm calculated an effective location for the source and detector that mathematically projected the BB’s onto the same locations on the detector as found on the projection images. The average distance between the BB projections on the image and the mathematically computed projections was 0.11 mm. The effective locations for the source and detector were encoded in the DICOM file for each projection; these were then used by the reconstruction algorithm. Tomographic image reconstructions were performed for three acquisition modes of the NGT system; these successfully demonstrated isotropic SR, magnified SR, and oblique reconstruction.
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.