The purpose of this project was to enhance breast cancer screening by identifying the factors that impact it. Breast cancer is the most common cancer diagnosed in women, and American women have a 12.3% risk of contracting it at some point in their life. Although screening has proven to be successful, controversies surrounding breast cancer screening programs remain. The screening start age, end age and frequency are subjects of the debate. This project aimed to address these controversies by developing a computer simulation of breast cancer screening. The simulation used rejection sampling to generate a cohort of 1 million women with 5 main attributes which are 1) age in the year 2016, 2) race, 3) age of death from non-breast cancer causes, 4) age at which the breast cancer was contracted, and 5) the growth rate of breast cancer. We then examined the generated cohort with different screening procedures. The study focused on mammography and breast CT screening modalities. To determine the most efficient breast cancer screening protocol several parameters were of interest. One of these parameters was years of life saved per number of tests performed. We concluded that screening every 2 years with CT starting at age 50 until 80 has the highest years of life saved per number of screenings for the entire population. The study also suggests that races benefit differently from each screening program, which should be taken into account for screening policies.

Purpose

Methods

Results

Conclusion

Breast computed tomography (CT) is a relatively new breast imaging modality based upon cone-beam CT geometry, and there are ongoing efforts to optimize breast CT for the detection of breast cancer lesions. Clinical trials are considered the ideal approach when conducting protocol optimization studies. However, the time, expense, and extensive database required to conduct these studies pose significant challenges and costs. As an alternative to clinical studies, simulation studies have been proposed, where synthetic images are generated using phantom imaging or computer simulations. In addition, model observers have been proposed in lieu of human observers to evaluate the simulated images. By using simulated images and model observers, thousands of images can be generated and evaluated in a short period of time. In this work, we developed methods to simulate “hybrid” breast CT images, where mathematically generated mass lesions and microcalcifications are inserted into actual patient breast CT volume data sets. Then, we used model observers, namely the pre-whitened matched filter (PWMF) and convolutional neural networks (CNNs), to detect the simulated lesions.First, we simulated contrast-enhanced mass lesions and explored the improvement in mass lesion detectability due to contrast enhancement across lesion diameter, section thickness, breast density, and view plane. An average 20% improvement was observed, and a larger improvement was observed for patients with dense breasts. Small lesions are generally harder to detect in dense breasts, but these results demonstrated that injected contrast can substantially improve detection performance in dense breasts. Next, we compared the PWMF model observer with the CNN model observer for detecting mathematically generated unenhanced mass lesions inserted into 1) breast CT background and 2) Gaussian background, where the PWMF is known to be an ideal observer. In Gaussian background, the CNN performed essentially identically to the PWMF across lesion sizes and section thicknesses. In breast CT background, the CNN outperformed the PWMF across lesion size, breast density, and most section thicknesses. These findings suggest that in breast CT images, CNNs capture more diagnostic information than PWMFs and may be a more pertinent observer when conducting optimal performance studies. Lastly, we simulated microcalcifications and microcalcification clusters. The loss of intensity owing to partial volume effects was modeled and used to mathematically insert microcalcifications into acquired patient breast CT images. 2D and 3D CNNs were used to evaluate the detectability of simulated calcifications across clinical parameters. Our results demonstrated the utility of the maximum intensity projection (MIP) for displaying image volumes containing microcalcification clusters. We found that there was no statistically significant difference in detection performance when using the MIP compared to all slices in the native section thickness, but that thicker sections led to reduced detection performance. The MIP procedure essentially compresses 3D images to 2D images, resulting in efficient and better detection for microcalcifications. Collectively, these studies elucidate the key factors affecting mass lesion and microcalcification detectability in unenhanced and contrast-enhanced breast CT and demonstrate the utility of model observers for examining breast CT images when human observers are unavailable. As breast CT advances towards translation to the clinic, these studies will be useful for optimizing breast CT protocols.

Objective: To generate low-dose Breast CT (LD-bCT) images from a cone-beam Breast CT (bCT) system and denoise them to achieve a comparable image quality to their standard-dose (SD) counterparts.Methods: LD-bCT phantom images were generated from standard-dose bCT (SD-bCT) phantom datasets for specific pixel binning and gain modes using four parameters unique to each flat-panel detector. The four parameters determined were F, the mean signal-variance relationship, σ2e, the variance of electronic noise, Kq, the autocorrelation kernel of quantum noise, and Ke, the autocorrelation kernel of electronic noise. A noise injection equation incorporating the four parameters was formulated and applied to SD phantom datasets. The noise power spectrum (NPS) was used as a similarity metric to validate the noise injection procedure for LD-bCT simulation. The mean squared error (MSE) and noise variance quantified the difference between the curves. After the verification, LD-bCT patient data was generated at 75 and 50% of the SD level. Transfer learning with a pretrained network called the Residual Encoder-Decoder CNN (RED-CNN) was utilized for the LD-bCT denoising task since it was previously trained on Low-Dose CT data. A variation of the Hyperbane hyperparameter tuning algorithm was performed to identify the optimal hyperparameters. The loss metric was chosen as MSE, and the evaluation metrics were selected as the root mean squared error (RMSE), peak signal-to-noise ratio (PSNR), and structural similarity index (SSIM). The model's overall performance was validated using adversarial inputs, the testing set, and another test set that appreciated for all 18 patient datasets. Results: The variance of electronic noise was calculated as 82.747ADU2, the F factor as 0.00125, Ke as a 2D delta function, and Kq as a 7x7 correlation matrix. The NPS curves had good agreement with MSE values of 2.22e-14, 3.56e-14, and 4.82e-13 at 90, 60 and 45mA, respectively. After applying transfer learning using a RED-CNN and deciding optimal parameters for the denoising task, the model was trained for 150 epochs and obtained a validation MSE loss of 5.11e-5. During the testing stage, the model obtained an RMSE of 0.0097, PSNR of 41.0308, and SSIM of 0.9673. Conclusion: The LD-bCT simulation supplies an accurate noise modeling technique for generating LD-bCT images with excellent NPS agreement with its corresponding SD-bCT. When LD-bCT patient data was fed to an extension of the pre-trained RED-CNN, the images are found to be effectively denoised without comprising on detail and while preserving anatomical structures. Clinically, this research finds applications in LD-bCT patient acquisitions and SD-bCT enhancement. Keywords: Computed Tomography, Denoising CNNs, Low-Dose Image Denoising

Purpose

Methods

Results

Conclusions

This dissertation is focused on the development and characterization of a novel x-ray tube prototype for digital breast tomosynthesis (DBT) and the application of this multisource approach to cone beam CT (CBCT) and dedicated breast CT (bCT). The design and testing of a prototype Multi-X-ray-source Array (MXA) for DBT is reported. The MXA is comprised of an array of tungsten filament cathodes with focus cup grid-controlled modulation and a common rotating anode housed in a single vacuum envelope. Prototypes consisting of arrays of three-source elements and eleven-source-elements were fabricated and evaluated. The prototype sources demonstrated focal spot sizes of 0.3 mm at 45 kV with 50 mA. Measured X-ray spectra were consistent with the molybdenum anode employed, and the tube output (air kerma) was between 0.6 mGy/100 mAs at 20 kV and 17 mGy/100 mAs at 45 kV with a distance of 100 cm. HVL measurements ranged from 0.5 mm Al at 30 kV to 0.8 mm Al at 45 kV, and X-ray pulse widths were varied from 20 ms to 110 ms at operating frequencies ultimately to be limited by source turn-on/off times of ~1 ms. Initial results of reconstructed tomographic data were presented.

Multisource configurations were applied to CBCT using phantom imaging and Monte Carlo simulations. Image quality, scatter, and dose were evaluated in both overlapping (large cone angle) and collimated (small cone angle) configurations for CBCT. Four x-ray tube configurations were considered: traditional one source, three source overlapping, six source overlapping, and six sources collimated. Image quality was evaluated on a prototype breast CT system using the following five phantoms: a Defrise phantom, a previously reported CBCT QA phantom (Corgi), a polyethylene cylinder, and two anthropomorphic phantoms (hand and knee). Scatter contamination and radiation dose were evaluated using Monte Carlo simulations of a voxelized polyethylene cylinder.

The modulation of the Defrise phantom disks on average was 2.7X greater for the six-source collimated configuration than the six-source overlapping configuration. The data lost from cone beam artifact (spatial domain) and the null cone (frequency domain) in the overlapping configuration were completely recovered using the collimated configuration. The maximum scatter to primary ratio (SPR) for the overlapping configuration was 0.81 and the maximum SPR for the collimated configuration was 0.26. The average dose and maximum dose was 4X less in the collimated six source configuration when compared with the overlapping configurations. The maximum dose for the overlapping configurations (one, three & six) remained constant, but the average dose for the multisource (three & six source) overlapping configurations increased 25% when compared to the one source configuration. Use of a collimated multisource x-ray tube configuration was shown to provide significant improvements in image quality throughout the cone-beam geometry field-of-view, reduction in scatter contamination, and more efficient use of dose in comparison to both traditional cone-beam CT geometry with a single source and the overlapping multisource configurations.

A phantom was developed to simulate microcalcification lesions and a non-prewhitening-matched filter (NPW) model was applied to evaluate detectability for various technique parameters and cone angle regions in bCT. The detectability of microcalcifications was improved with smaller pixel size (75 µm) and higher dose (6 mGy) when compared to larger pixel size (150 µm) and lower dose (3 mGy). The number of projections (N = 250 or N = 500) did not affect the detectability significantly. The detectability decreased from small to medium cone angle regions and from small to large cone angle regions but was not significantly affected from medium to large cone angle regions. The differences in detectability across cone angle regions was more prominent in the coronal plane than the sagittal and axial planes.

Purpose

Method

Result

Conclusion

Purpose

Methods

Results

Conclusions