Browsing by Subject "Mammography"

A 3D microwave imaging system suitable for clinical trials has been developed. The anatomy, histology, and pathology of breast cancer were all carefully considered in the development of this system. The central component of this system is a breast imaging chamber with an integrated 3D antenna array containing 36 custom designed bowtie patch antennas that radiate efficiently into human breast tissue. 3D full-wave finite element method models of this imaging chamber, complete with full antenna geometry, have been developed using Ansoft HFSS and verified experimentally. In addition, an electronic switching system using Gallium Arsenide (GaAs) absorptive RF multiplexer chips, a custom hardware control system with a parallel port interface utilizing TTL logic, and a custom software package with graphical user interface using Java and LabVIEW have all been developed. Finally, modeling of the breast (both healthy and malignant) was done using published data of the dielectric properties of human tissue, confirming the feasibility of cancer detection using this system.

Based on a biopsychosocial model of mammography pain, the current study assessed if specific biological and psychosocial factors were associated with higher reported mammography pain in early stage breast cancer survivors. One hundred and twenty-seven women completed questionnaires assessing demographic information, cancer treatment history, ongoing breast pain, mammography-related anxiety, and social support immediately prior to receiving a mammogram. They then completed questionnaires assessing mammography pain and mammography-related pain catastrophizing immediately following the mammogram. Using path modeling and mediation analyses, relations among these variables were examined. Results revealed that mammography-related pain catastrophizing was related to higher mammography pain directly, while ongoing breast pain, lower social support quantity, and lower perceived quality of social support related to higher mammography pain indirectly through mammography-related pain catastrophizing. Moderated mediation analyses found that the mediation effects of mammography-related pain catastrophizing were significantly different at varying levels of perceived quality of social support, with more pronounced negative effects for those with higher quality support than those with lower quality support. The theoretical, clinical, and research implications of these findings are discussed.

Breast cancer screening is currently performed by mammography, which is limited by overlying anatomy and dense breast tissue. Computer aided detection (CADe) systems can serve as a double reader to improve radiologist performance. Tomosynthesis is a limited-angle cone-beam x-ray imaging modality that is currently being investigated to overcome mammography's limitations. CADe systems will play a crucial role to enhance workflow and performance for breast tomosynthesis.

The purpose of this work was to develop unique CADe algorithms for breast tomosynthesis reconstructed volumes. Unlike traditional CADe algorithms which rely on segmentation followed by feature extraction, selection and merging, this dissertation instead adopts information theory principles which are more robust. Information theory relies entirely on the statistical properties of an image and makes no assumptions about underlying distributions and is thus advantageous for smaller datasets such those currently used for all tomosynthesis CADe studies.

The proposed algorithm has two 2 stages (1) initial candidate generation of suspicious locations (2) false positive reduction. Images were accrued from 250 human subjects. In the first stage, initial suspicious locations were first isolated in the 25 projection images per subject acquired by the tomosynthesis system. Only these suspicious locations were reconstructed to yield 3D Volumes of Interest (VOI). For the second stage of the algorithm false positive reduction was then done in three ways: (1) using only the central slice of the VOI containing the largest cross-section of the mass, (2) using the entire volume, and (3) making decisions on a per slice basis and then combining those decisions using either a linear discriminant or decision fusion. A 92% sensitivity was achieved by all three approaches with 4.4 FPs / volume for approach 1, 3.9 for the second approach and 2.5 for the slice-by-slice based algorithm using decision fusion.

We have therefore developed a novel CADe algorithm for breast tomosynthesis. The techniques uses an information theory approach to achieve very high sensitivity for cancer detection while effectively minimizing false positives.

We present a new x-ray imaging technique, Correlation Imaging (CI), for improved breast and lung cancer detection. In CI, multiple low-dose radiographic images are acquired along a limited angular arc. Information from unreconstructed angular projections is directly combined to reduce the effect of overlying anatomy - the largest bottleneck in diagnosing cancer with projection imaging. In addition, CI avoids reconstruction artifacts that otherwise limit the performance of tomosynthesis. This work involved assessing the feasibility of the CI technique, its optimization, and its implementation for breast and chest imaging.

First a theoretical model was developed to determine the diagnostic information content of projection images using a mathematical observer. The model was benchmarked for a specific application in assessing the impact of reduced dose in mammography. Using this model, a multi-factorial task-based framework was developed to optimize the image acquisition of CI using existing low-dose clinical data. The framework was further validated using a CADe processor. Performance of CI was evaluated on mastectomy specimens at clinically relevant doses and further compared to tomosynthesis. Finally, leveraging on the expected improvement in breast imaging, a new hardware capable of CI acquisition for chest imaging was designed, prototyped, evaluated, and experimentally validated.

The theoretical model successfully predicted diagnostic performance on mammographic backgrounds, indicating a possible reduction in mammography dose by as much as 50% without adversely affecting lesion detection. Application of this model on low-dose clinical data showed that peak CI performance may be obtained with 15-17 projections. CAD results confirmed similar trends. Mastectomy specimen results at higher dose revealed that the performance of optimized breast CI may exceed that of mammography and tomosynthesis by 18% and 8%, respectively. Furthermore, for both CI and tomosynthesis, highest dose setting and maximum angular span with an angular separation of 2.75o was found to be optimum, indicating a threshold in the number of projections per angular span for optimum performance.

Finally, for the CI chest imaging system, the positional errors were found to be within 1% and motion blur to have minimal impact on the system MTF. The clinical images had excellent diagnostic quality for potentially improved lung cancer detection. The system was found to be robust and scalable to enable advanced applications for chest radiography, including novel tomosynthesis trajectories and stereoscopic imaging.

Proper breast positioning in screening mammography minimizes the likelihood that tissue is obscured or missing from a mammogram and decreases the rate of technical repeats and misdiagnosis. Several factors contribute to improper positioning including technologist experience, patient condition, and patient body habitus. One way to help the technologist identify and improve improper positioning is through MQSA’s required periodic image quality performance evaluations known as the EQUIP review. While valuable for performance improvement, the number of cases evaluated during these reviews is relatively small compared to the overall number of studies a technologist performs and may not provide the nuanced feedback needed for process improvement. In this study, we have developed convolutional neural network models to detect several aspects of suboptimal positioning as defined by the ACR’s positioning standards towards providing technologists valuable feedback for measuring performance improvement.The dataset contained a total of 600 clinical screening mammograms with a variety of positioning-related anomalies. An experienced technologist labeled relevant anomalies and image features important to breast positioning assessment. Image review tasks were separated into one of two categories: classification tasks and segmentation tasks. Using a neural network approach, segmentation models were developed to identify the pectoralis muscle boundary, skin folds, the inframammary fold (IMF), and nipple location. Classification models were developed to determine if the nipple was in profile, if the breast was droopy, and if the IMF was open, not open, or not shown. The classification models had a final accuracy of 89.3% for classifying whether the IMF was shown or not, 81.3% for determining if the IMF was shown and open, shown and not open, or not shown, and 80.2% for identifying if the nipple was in profile. These algorithms will ultimately be implemented to evaluate all screening mammograms acquired at Duke Health to automatically identify breast positioning features for clinical feedback.

The overall goal of this dissertation is to optimize and evaluate the performance of the single photon emission computed tomography (SPECT) subsystem of a dedicated three-dimensional (3D) dual-modality breast imaging system for enhanced semi-automated, quantitative clinical imaging. This novel hybrid imaging system combines functional or molecular information obtained with a SPECT subsystem with high-resolution anatomical imaging obtained with a low dose x-ray Computed Tomography (CT) subsystem. In this new breast imaging paradigm, coined "mammotomography," the subject is imaged lying prone while the individual subsystems sweep 3-dimensionally about her uncompressed, pendant breast, providing patient comfort compared to traditional compression-based imaging modalities along with high fidelity and information rich images for the clinician.

System evaluation includes a direct comparison between dedicated 3D SPECT and dedicated 2D scintimammography imaging using the same high performance, semi-conductor gamma camera. Due to the greater positioning flexibility of the SPECT system gantry, under a wide range of measurement conditions, statistically significantly (p<0.05) more lesions and smaller lesion sizes were detected with dedicated breast SPECT than with compressed breast scintimammography. The importance of good energy resolution for uncompressed SPECT breast imaging was also investigated. Results clearly illustrate both visual and quantitative differences between the various energy windows, with energy windows slightly wider than the system resolution having the best image contrast and quality.

An observer-based contrast-detail study was performed in an effort to evaluate the limits of object detectability under various imaging conditions. The smallest object detail was observed using a 45-degree tilted trajectory acquisition. The complex 3D projected sine wave acquisition, however, had the most consistent combined intra and inter-observer results, making it potentially the best imaging approach for consistent clinical imaging.

Automatic ROR contouring is implemented using a dual-layer light curtain design, ensuring that an arbitrarily shaped breast is within ~1 cm of the camera face, but no closer than 0.5 cm at every projection angle of a scan. Autocontouring enables simplified routine scanning using complex 3D trajectories, and yields improved image quality. Absolute quantification capabilities are also integrated into the SPECT system, allowing the calculation of in vivo total lesion activity. Initial feasibility studies in controlled low noise experiments show promising results with total activity agreement within 10% of the dose calibrator values.

The SPECT system is integrated with a CT scanner for added diagnostic power. Initial human subject studies demonstrate the clinical potential of the hybrid SPECT-CT breast imaging system. The reconstructed SPECT-CT images illustrate the power of fusing functional SPECT information to localize lesions not easily seen in the anatomical CT images. Enhanced quantitative 3D SPECT-CT breast imaging, now with the ability to dynamically contour any sized breast, has high potential to improve detection, diagnosis, and characterization of breast cancer in upcoming larger-scale clinical testing.

Machine learning algorithms are being adopted for clinical use, assisting with difficult medical tasks previously limited to highly-skilled professionals. AI (artificial intelligence) performance on isolated tasks regularly exceeds that of human clinicians, spurring excitement about AI's potential to radically change modern healthcare. However, there remain major concerns about the uninterpretable (i.e., "black box") nature of commonly-used models. Black box models are difficult to troubleshoot, cannot provide reasoning for their predictions, and lack accountability in real-world applications, leading to a lack of trust and low rate of adoption by clinicians. As a result, the European Union (through the General Data Protection Regulation) and the US Food & Drug Administration have published new requirements and guidelines calling for interpretability and explainability in AI used for medical applications.

My thesis addresses these issues by creating interpretable models for the key clinical decisions of lesion analysis in mammography (Chapters 2 and 3) and pattern identification in EEG monitoring (Chapter 4). To create models with comparable discriminative performance to their uninterpretable counterparts, I constrain neural network models using novel neural network architectures, objective functions and training regimes. The resultant models are inherently interpretable, providing explanations for each prediction that faithfully represent the underlying decision-making of the model. These models are more than just decision makers; they are decision aids capable of explaining their predictions in a way medical practitioners can readily comprehend. This human-centered approach allows a clinician to inspect the reasoning of an AI model, empowering users to better calibrate their trust in its predictions and overrule it when necessary

Currently the most common and effective procedure of early detection of breast cancer is through screening mammography. Mammography detects not only invasive cancers, but also in situ lesions including ductal carcinoma in situ (DCIS), which is the most common non-invasive breast cancer. The focus of this dissertation is to investigate machine learning and deep learning technologies to improve detection and diagnosis of breast cancer from mammography, focusing on calcifications that represent a common type of suspicious lesions. This dissertation is composed of four main projects.

The first project developed a transfer learning framework to predict occult invasive disease in DCIS by a machine learning approach. Performance was improved by using forced labeling and domain adaptation based upon data from two related diseases, atypical ductal hyperplasia (ADH) and invasive ductal carcinoma (IDC).

The second project expanded the dataset fivefold, producing the largest cohort of DCIS cases to date. This rich dataset enabled rigorous studies to assess the performance and clinical utility for guiding patients with DCIS who are considering active surveillance or surgical planning.

The third project developed a one-class anomaly detection system to detect calcifications with deep autoencoder network and dissimilarity index. Calcifications were detected by learning negative-only images. Using a very large mammography dataset, this study was one of the first in medical imaging to show model capacity changes based on different architectures as well as different number or type of training images.

Finally, the fourth project broadened the diagnostic task to classify breast calcifications as benign or malignant using a deep multitask network. Unlike previous projects, algorithms here were designed to detect calcifications and classify cancers simultaneously. The multitask network model’s results showed improved performance compared to single classification networks. A correlation between two output losses’ weighting factors was also achieved by analyzing performance impacts from both weighting factors.

In conclusion, this dissertation provides three main contributions. First, it provides a complete classification framework to distinguish invasive disease from DCIS based on the largest single institution dataset. Second, it demonstrates the potential of an anomaly detection approach to detect mammographic calcifications that may facilitate new clinical applications such as computer-aided triage and quality improvement. Third, a unified approach for multitask learning was used to perform detection and classification simultaneously, which provided the highest performance for classification from our group to date.