Browsing by Subject "CT"

There are three main x-ray based modalities for imaging the thorax: radiography, tomosynthesis, and computed tomography (CT). CT perhaps provides the highest level of feature resolution but at notably higher radiation dose, which has increased the concern among radiation protection professionals. Being able to accurately assess the radiation dose patients receive during CT procedures is a crucial step in the management of CT dose. To identify the best imaging modality for patients, the American College of Radiology published the guiding principle of "The right exam, for the right reason, at the right time". To implement this principle in making an appropriate choice between standard chest projection imaging, tomosynthesis, and CT, the organ and effective dose for each modality should be accurately known. This thesis work attempted to explain the effect on dose results when choosing different types of computational phantoms used in CT dosimetry; this work also compared radiation dose across three main x-ray based modalities on one common platform for different body shape adults.

The first part of this thesis compared organ doses, effective doses, and risk indices from 13 representative adult CT protocols using four types of reference phantoms (XCAT, ICRP 110, ImPACT, and CT-Expo). Despite closely-matched organ mass, total body weight, and height, large differences in organ dose exist due to variation in organ location, spatial distribution, and dose approximation method. Dose differences for fully irradiated radiosensitive organs were much smaller than those for partially irradiated organs. Weighted dosimetry quantities including effective dose, male risk indices, k factors, and male q factors agreed well across phantoms. The female risk indices and q factors varied considerably across phantoms.

PurposeThe aim of this study was to address the limitations observed in past research, particularly the limited accuracy of individual deep learning or radiomics methods in small datasets. By developing a fusion approach that integrates the two techniques, we hypothesized that the performance in CT-based thyroid cancer diagnosis could be improved. Materials and Methods Eighty-five patients with thyroid tumors (58 malignant, 27 benign) who underwent CT scans were included in this study. The dataset was divided into training (70%) and testing (30%). A shallow CNN model, including five convolutional layers and two fully connected layers, was developed for tumor classification. Radiomics features were extracted and selected using the pyradiomics package and statistical tests (T-test, etc.). These features were then utilized to develop a Multiple Logistic Regression (MLR) model for tumor classification. The CNN and MLR models were combined using a fusion method that calculates the weighted sum of each diagnostic output for classification. The accuracy of the diagnostic methods was evaluated for both the individual and combined fusion models. The statistical significance of the weighted combination model was examined using the Wilcoxon-Test. Results The CNN model achieved an accuracy of 82.713%, and the MLR model achieved an accuracy of 76.596%. The accuracy of the fusion model reached 85.372%, suggested the improvement of performance of the fusion approach over the individual models. The Wilcoxon-Test yielded a W-Statistic of 19410.0 and a p-value of 〖2.96×10〗^(-14), which is below the threshold of 0.05. Conclusion A fusion model combining deep learning and radiomics methods was developed and showed improved accuracy in thyroid tumor diagnosis in a small dataset. The results showed a statistically significant difference between the fusion model and the individual models.

Cone-beam computed tomography (CBCT) is an X-ray-based imaging modality widely used in medical practices. Due to the ionizing imaging dose induced by CBCT, many studies were conducted to reduce the number of projections (sparse sampling) to lower the imaging dose while maintaining good image quality and fast reconstruction speed. Conventionally, a CBCT volume is reconstructed analytically with the Feldkamp Davis Kress (FDK) algorithm that backprojects filtered projections according to projection angles. However, the FDK algorithm requires a dense angular sampling that satisfies the Shannon-Nyquist theorem. The FDK algorithm reconstructs CBCT with a high speed but requires relatively high patient imaging dose. The iterative methods like algebraic reconstruction technique (ART) and compressed sensing (CS) methods are investigated to reduce patient imaging dose. These iterative methods update estimated images iteratively and the CS methods apply penalty terms to award desired features. Yet these methods are limited by the iterative design with substantially increased computation time and consumption of computation power. Scholars have also conducted research on bypassing the limit of Shannon-Nyquist theorem by interpolating densely sampled CBCT projections from sparsely sampled projections. However, blurred structures in reconstructed images remain to be a concern for analytical interpolation methods. As such, previous research indicates that it is hard to achieve the three goals of lowered patient imaging dose, good image quality, and fast reconstruction speed all at once.

As deep learning (DL) gained popularity in fields like computer vision and data science, scholars also applied DL techniques in medical image processing. Studies on DL-based CT image reconstruction have yielded encouraging results, but GPU memory limitation made it challenging to apply DL techniques on CBCT reconstruction.

In this dissertation, we hypothesize that the image quality of CBCT reconstructed from under-sampled projections (low-dose) using deep learning techniques can be comparable to that of CBCT reconstructed from fully sampled projections for treatment verification in radiation therapy. This dissertation proposes that by applying DL techniques in pre-processing, reconstruction, and post-processing stages, the challenge of improving CBCT image quality with low imaging dose and fast reconstruction speed can be mitigated.

The dissertation proposed a geometry-guided deep learning (GDL) technique, which is as the first technique to perform end-to-end CBCT reconstruction from sparsely sampled projections and demonstrated its feasibility for CBCT reconstruction from real patient projection data. In this study, we have found that incorporating geometry information into the DL technique can effectively reduce the model size, mitigating the memory limitation in CBCT reconstruction. The novel GDL technique is composed of a GDL reconstruction module and a post-processing module. The GDL reconstruction module learns and performs projection-to-image domain transformation by replacing the traditional single fully connected layer with an array of small fully connected layers in the network architecture based on the projection geometry. The additional deep learning post-processing module further improves image quality after reconstruction.

This dissertation further optimizes the number of beamlets used in the GDL technique through a geometry-guided multi-beamlet deep learning (GMDL) technique. In addition to connecting each pixel in the projection domain to beamlet points along the central beamlet in the image domain as GDL does, these smaller fully connected layers in GMDL connect each pixel to beamlets peripheral to the central beamlet based on the CT projection geometry. Due to the limitation of GPU VRAM, the proposed technique is demonstrated through low-dose CT image reconstruction and is compared with the GDL technique and a large fully connected layer-based reconstruction method.

In addition, the dissertation also investigates deep learning-based CBCT projection interpolation and proposes a patient-independent deep learning projection interpolation technique for CBCT reconstruction. Different from previous studies that interpolate phantom or simulated data, the proposed technique is demonstrated to work on real patient projection data with unevenly distributed projection angles. The proposed technique re-slices the stack of interpolated projections axially, and each acquired slice is processed by a deep residual U-Net (DRU) model to augment the slice’s image quality. The resulting slices are reassembled into a stack of densely-sampled projections to be reconstructed into a CBCT volume. A second DRU model further post-processes the reconstructed CBCT volume to improve the image quality.

In summary, a geometry-guided deep learning (GDL) technique was proposed as the first deep learning technique for end-to-end CBCT reconstruction from sparsely sampled real patient projection data. The geometry-guided multi-beamlet deep learning (GMDL) technique further optimizes the number of beamlets based on the GDL technique. A patient-independent deep learning projection interpolation technique was also proposed for the pre-processing and post-processing stage of CBCT reconstruction.

In conclusion, the work presented in this dissertation demonstrates the feasibility of improving CBCT image quality with low imaging dose and fast reconstruction speed. The techniques developed in this dissertation also have great potential for clinical applications to enhance CBCT imaging for radiation therapy.

Virtual imaging trials (VITs) are a growing part of medical imaging research. VITs are a powerful alternative to the current gold-standard for determining or verifying the efficacy of new technology in healthcare: the clinical trial. Prohibitively high expenses, multi-site standardization of protocols, and risks to the health of the trial’s patient population are all challenges associated with the clinical trial; conversely, these challenges highlight the strengths of virtualization, particularly with regard to evaluating medical imaging technologies.Virtual imaging requires a combination of virtual subjects, physics-based imaging simulation platforms, and virtual pathologies. Currently, most computational phantom organs and pathologies are segmented or generated from clinical CT images. With this approach, most computational organs and pathologies are necessarily static, comprising only a single instantaneous representation. Further, this static-anatomy–static-pathology approach does not address the underlying physiological constraints acting on the organs or their pathologies—making some imaging exams (e.g., perfusion, coronary angiography) difficult to simulate robustly. It also does not provide a clear path toward including anatomical and physiological (functional) detail at sub-CT resolution. This project aims to integrate high-resolution, dynamic features into computational human models. The focus is primarily an advanced model known as XCAT. These additions include healthy and progressive-disease anatomy and physiology, micron-level–resolution coronary artery lesions, and an array of pathologies. In particular, we focus on the physiology needed for CT perfusion studies, dynamic lesions, or coronary artery disease (CAD), and means to integrate each of these features into XCAT via custom software. The outcome is further to demonstrate the utility of each of these advances with representative simulated imaging. Chapter 1 presents a method using clinical information and physiological theory to develop a mathematical model that produces the liver vasculature within a given XCAT. The model can be used to simulate contrast perfusion by taking into account contrast position and concentration at an initial time t and the spatial extent of the contrast in the liver vasculature at subsequent times. The mathematical method enables the simulation of hepatic contrast perfusion in the presence or absence of abnormalities (e.g., focal or diffuse disease) for arbitrary imaging protocols, contrast concentrations, and virtual patient body habitus. The vessel growing method further generalizes to vascular models of other organs as it is based on a parameterized approach, allowing for flexible repurposing of the developed tool. Chapter 2 presents a method for using cardiac plaque histology and morphology data acquired at micron-level resolution to generate new, novel plaques informed by a large, original patient cohort. A methodology for curating and validating the anatomical and physiological realism was further applied to the synthesized plaques to ensure realism. This method was integrated with the XCAT heart and coronary artery models to allow simulated imaging of a wide variety of coronary artery plaques in varied orientations and with unique material distribution and composition. Generation of 200 unique plaques has been optimized to take as little as 5 seconds with GPU acceleration. This work enables future studies to optimize current and emerging CT imaging methods used to detect, diagnose, and treat coronary artery disease. Chapter 3 focuses on small-scale modeling of the internal structure of the bones of the chest. The internal structure of the bones appears as a diffuse but recognizable texture under medical imaging and corresponds to a complex physical structure tuned to meet the physical purpose of the bone (e.g., weight-bearing, protective structure, etc.). The project aimed to address the limitations of prior texture-based modelling by creating mathematically based fine bone structures. The method was used to generate realistic bone structures, defined as polygon meshes, with accurate morphological and topological detail for 45 chest bones for each XCAT phantom. This new method defines the spatial extent of the complementary bone–marrow structures that are the root cause of the characteristic image texture 1-4 and provides a transition from using image-informed characteristic power law textures to a ground-truth model with exact morphology—which we additionally paired with the DukeSim CT simulator5 and XCAT phantoms6 to produce radiography and CT images with physics-based bone textures. This work enables CT acquisition parameter optimization studies that can inform clinical image assessment of osteoporosis and bone fractures. Chapter 4 proposes a new model of lesion morphology and insertion and was created with the intent to be informed and validated by—rather than constrained by—imaging data. It additionally includes the new incorporation of biological data, intended to provide dynamic computational lung lesion models for use in CT simulation applications. Each chapter includes a section presenting an example application of the respective tools in virtual medical imaging. Chapter 5 concludes this work with a brief summary of the content and is followed by Appendices A–D. The appendices are organized by topic and contain a visual demonstration of the work in a series of high-resolution, full-page images.

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.

Computational phantoms are set to play an important role in imaging research. As medicine moves increasingly towards providing individualized, patient-specific care, it is imperative that simulations be completed on patient-specific anatomy, rather than a reference standard. To that end, there is need for a variety of realistic phantoms for clinical studies.

This work adds to the existing extended cardiac and torso (XCAT) adult phantom series (two phantoms based on visual human data) by building new models based on adult patient computed tomography (CT) image data. These CT datasets were obtained from Duke University's patient CT database.

Each image-set was segmented using in-house segmentation software, defining bony structures and large organs within the field of view. 3D non-uniform rational b-spline (NURBS) surfaces were fitted to the segmented data. Using the multi-channel large diffeomorphic deformation metric mapping (MC-LDDMM) network, a transform was calculated to morph an existing XCAT model to the segmented patient geometry. Fifty-eight adult XCAT models were added to the phantom library.

In addition to the expanding the XCAT library, the feasibility of incorporating texture was investigated. Currently, the XCAT phantom structures are assumed to be homogeneous. This can lead to unrealistic appearance when the phantoms are combined with imaging simulations, particularly in CT. The purpose of this project was to capture anatomical texture and test it in a simulated phantom. Image data from the aforementioned patient CT database served as the source of anatomical texture.

The images were de-noised using anisotropic diffusion. Next, several regions of interest (ROIs) were taken from the liver and lungs of CT images. Using the ROIs as a source of texture, a larger stochastic texture image-set was created using the Image Quilting algorithm.

The visual human adult male XCAT phantom was voxelized at the same resolution as the texture image. The voxels inside the liver were directly replaced by the corresponding voxels of texture. Similarly for the lung, the voxels between the existing lung bronchi/blood vessels and the lung wall were replaced by texture voxels. This procedure was performed using ten different patient CT image-sets as sources of texture.

To validate the similarity of the artificial textures to the source textures, reconstructions of the adult male XCAT phantom with added textures were compared to the clinical images via receiver operator characteristic (ROC) analysis, a two-sample t-test, equivalence test, and through comparing absolute differences between scores.

It was concluded that this framework provides a valuable tool in which anatomical texture can be incorporated into computational phantoms. It is anticipated that this step towards making many anatomically variable virtual models indicative of a patient populace and making these models more realistic will be useful in medical imaging research, especially for studies relating to image quality.

The goal of this research was to investigate the feasibility of using a weighted dual-energy subtraction method in dedicated breast CT using quasi-monochromatic x-ray beams, to better distinguish soft tissues in the breast. This research used computer simulations and measurements, includes development of protocols and phantoms, and yields quantitative results and physical components. Quasi-monochromatic beams were created using different specially processed, hermetically sealed ultra-thick K-edge filters yielding a maximum mean energy difference of 7 keV; also, different kVp potentials on the x-ray tube were used with higher atomic number filters, yielding a maximum mean energy difference of 11 keV. A cylindrical phantom containing muscle tissue equivalent material, glandular tissue equivalent material, polyethylene, and acrylic was filled with methanol (adipose tissue equivalent) was developed and scanned to investigate dual-energy contrast enhancement of the different materials. The CT scans were acquired using a dual-modality SPECT-CT system for dedicated breast imaging. The weighted dual-energy subtraction method is adapted from dual-energy x-ray absorptiometry, where a high and low energy CT scan is acquired using two different ultra-thick k-edge filters (or tube potentials) and the data is reconstructed. Region of interest values are obtained from each image, which is then multiplied by a weighted value (k), and the higher energy image is subtracted from the lower energy image to achieve contrast enhancement. The k-values were calculated from the ratio of measured attenuation coefficients for the material to be subtracted. In a projection-noise normalized acquisition scenario, it was found that weighted dual-energy subtraction with quasi-monochromatic beams might not be feasible under the current circumstances due to the especially high noise (poor contrast to noise ratios) and poor contrast. While this was not an optimized scenario, the approach does have promise, indicating that more investigation is needed.

The work presented in this dissertation focuses on evaluation of absolute quantification accuracy in dedicated breast SPECT-CT. The overall goal was to investigate through simulations and measurements the impact and utilization of various correction methods for scattered and attenuated photons, characterization of incomplete charge collection in Cadmium Zinc Telluride detectors as a surrogate means of improving scatter correction, and resolution recovery methods for modeling collimator blur during image reconstruction. The quantification accuracy of attenuation coefficients in CT reconstructions was evaluated in geometric phantoms, and a slice-by-slice breast segmentation algorithm was developed to separate adipose and glandular tissue. All correction and segmentation methods were then applied to a pilot study imaging parathyroid patients to determine the average uptake of Tc-99m Sestamibi in healthy breast tissue, including tissue specific uptake in adipose and glandular tissue.

Monte Carlo methods were utilized to examine the changes in incident scatter energy distribution on the SPECT detector as a function of 3D detector position about a pendant breast geometry. A simulated prone breast geometry with torso, heart, and liver was designed. An ideal detector was positioned at various azimuthal and tilted positions to mimic the capabilities of the breast SPECT subsystem. The limited near-photopeak scatter energy range in simulated spectra was linearly fit and the slope used to characterize changes in scatter distribution as a function of detector position. Results show that the detected scatter distribution changes with detector tilt, with increasing incidence of high energy scattered photons at larger detector tilts. However, reconstructions of various simulated trajectories show minimal impact on quantification (<5%) compared to a primary-only reconstruction.

Two scatter compensation methods were investigated and compared to a narrow photopeak-only windowing for quantification accuracy in large uniform regions and small, regional uptake areas: 1) a narrow ±4% photopeak energy window to minimize scatter in the photopeak window, 2) the previously calibrated dual-energy window scatter correction method, and 3) a modified dual-energy window correction method that attempts to account for the effects of incomplete charge collection in Cadmium Zinc Telluride detectors. Various cylindrical phantoms, including those with imbedded hot and cold regions, were evaluated. Results show that the Photopeak-only and DEW methods yield reasonable quantification accuracy (within 10%) for a wide range of activity concentrations and phantom configurations. The mDEW demonstrated highly accurate quantification measurements in large, uniform regions with improved uniformity compared to the DEW method. However, the mDEW method is susceptible to the calibration parameters and the activity concentration of the scanned phantom. The sensitivity of the mDEW to these factors makes it a poor choice for robust quantification applications. Thus, the DEW method using a high-performance CZT gamma camera is still a better choice for quantification purposes

Phantoms studies were performed to investigate the application of SPECT vs CT attenuation correction. Minor differences were observed between SPECT and CT maps when assuming a uniformly filled phantom with the attenuation coefficient of water, except when the SPECT attenuation map volume was significantly larger than the CT volume. Material specific attenuation coefficients reduce the corresponding measured activity concentrations compared to a water-only correction, but the results do not appear more accurate than a water-only attenuation map. Investigations on the impact of image registration show that accurate registration is necessary for absolute quantification, with errors up to 14% observed for 1.5cm shifts.

A method of modeling collimator resolution within the SPECT reconstruction algorithm was investigated for its impact on contrast and quantification accuracy. Three levels of resolution modeling, each with increasing ray-sampling, were investigated. The resolution model was applied to both cylindrical and anthropomorphic breast phantoms with hot and cold regions. Large volume quantification results (background measurements) are unaffected by the application of resolution modeling. For smaller chambers and simulated lesions, contrast generally increases with resolution modeling. Edges of lesions also appear sharper with resolution modeling. No significant differences were seen between the various levels of resolution modeling. However, Gibbs artifacts are amplified at the boundaries of high contrast regions, which can significantly affect absolute quantification measurements. Convergence with resolution modeling is also notably slower, requiring more iterations with OSEM to reach a stable mean activity concentration. Additionally, reconstructions require far more computing time with resolution modeling due to the increase in number of sampling rays. Thus while the edge enhancement and contrast improvements may benefit lesion detection, the artifacts, slower convergence, and increased reconstruction time limit the utility of resolution modeling for both absolute quantification and clinical imaging studies.

Finally, a clinical pilot study was initiated to measure the average uptake of Tc-99m Sestamibi in healthy breast tissue. Subjects were consented from those undergoing diagnostic parathyroid studies at Duke. Each subject was injected with 25mCi of Sestamibi as part of their pre-surgical parathyroid SPECT imaging studies and scanned with the dedicated breast SPECT-CT system before their diagnostic parathyroid SPECT scan. Based on phantom studies of CT reconstructed attenuation coefficient accuracy, a slice-by-slice segmentation algorithm was developed to separate breast CT data into adipose and glandular tissue. SPECT data were scatter, attenuation, and decay corrected to the time of injection. Segmented CT images were used to measure average radiotracer concentration in the whole breast, as well as adipose and glandular tissue. With 8 subjects scanned, the average measured whole breast activity concentration was found to be 0.10µCi/mL. No significant differences were seen between adipose and glandular tissue uptake.

In conclusion, the application of various characterization and correct methods for quantitative SPECT imaging were investigated. Changes in detected scatter distribution appear to have minimal impact on quantification, and characterization of low-energy tailing for a modified scatter subtraction method yields inferior overall quantification results. Comparable quantification accuracy is seen with SPECT and CT-based attenuation maps, assuming the SPECT-based volume is fairly accurate. In general, resolution recovery within OSEM yields higher contrast, but quantification accuracy appears more susceptible to measurement location. Finally, scatter, attenuation, and resolution recovery methods, along with a breast segmentation algorithm, were implemented in a clinical imaging study for quantifying Tc-99m Sestamibi uptake. While the average whole breast uptake was measured to be 0. 10µCi/mL, no significant differences were seen between adipose and glandular tissue or when implementing resolution recovery. Thus, for future clinical imaging, it's recommended that the application of the investigated correction methods should be limited to the traditional DEW method and CT-based attenuation maps for quantification studies.

Purpose: Radiation doses from computed tomography (CT) examinations have come under public and governmental scrutiny because of several recent misadministrations of radiation across the country. Current CT dosimetry methods in the clinic use standardized cylindrical water phantoms to measure radiation dose across various scanning protocols and different scanner manufacturers. These methods and equipment are too generalized to provide accurate risk assessment for patients of varying ages, genders, and anatomies. The advent of computer models based on real CT imaged anatomy has made patient specific and organ specific dosimetry achievable.

With a population of both pediatric and adult patient models comprised of a wide range of anatomies, Monte Carlo based dose calculations can be cataloged. A patient can receive a prospective dose estimation from a phantom within our population that best exhibits the patient's age and anatomical characteristics. Knowledge of organ size and location is essential to finding a proper match between the patient and the computer model. To this end, very little information is currently available regarding organ size and location across a diverse human population. The purpose of this study was to develop a predictive model to ascertain organ locations and volumes for pediatric and adult patients.

Methods: This study included 51 adults and 40 pediatrics from which Extended NURBS-based Cardiac-Torso (XCAT) phantoms were generated. Large organs were manually segmented from clinical CT data. The remaining organs and other anatomical structures were created by transforming an existing human model template to fit the framework of the segmented structures. The maximum and minimum points of the organs were recorded with respect to the axial distance from the tip of the sacrum. The axial width and midpoint for each organ were then determined. The organ volumes were also calculated. All three organ parameters were plotted as functions of patient age and weight for adults and patient age for pediatrics.

Results: The adult patients showed no statistically significant correlation between organ parameters and age and BMI. There were slight, positive linear trends with organ midpoint (max r2=0.365, mean r2=0.185) and organ volume (max r2=0.510, mean r2=0.183) versus adult patient weight. The height correlations were also positive for midpoint (r2=0.485, mean r2=0.271). Gaussian fits performed on probability density functions of adult organs resulted in r2-values ranging from 0.945 to 0.996. Pediatric patients demonstrated strong cube root relationships with organ midpoints (max r2=0.857, mean r2=0.790) and organ widths (max r2=0.905 , mean r2=0.564) versus age. Pediatric organ volumes showed positive linear relationships versus age (max r2=0.983, mean r2=0.701).

Conclusions: Adult patients exhibited small variations in organ volume and location with respect to weight, but no meaningful correlation existed between these parameters and age. Once adulthood is reached, organ morphology and positioning seems to remain static; however, clear trends are evident between pediatric age and organ volumes and locations. Such information can aid in the selection of an appropriate computer model that has the highest probability of mirroring the anatomy of a patient undergoing a clinical exam. Applications could also extend into comparing PET versus CT determination of organ volume and location.

X-ray computed tomography (CT) is a non-destructive medical imaging technique for assessing the cross-sectional images of an object in terms of attenuation. As it is designed based on the physical processes involved in the x-ray and matter interactions, faithfully modeling the physics in the reconstruction procedure can yield accurate attenuation distribution of the scanned object. Otherwise, unrealistic physical assumptions can result in unwanted artifacts in reconstructed images. For example, the current reconstruction algorithms assume the photons emitted by the x-ray source are mono-energetic. This oversimplified physical model neglects the poly-energetic properties of the x-ray source and the nonlinear attenuations of the scanned materials, and results in the well-known beam-hardening artifacts (BHAs). The purpose of this work was to incorporate the poly-energetic nature of the x-ray spectrum and then to eliminate BHAs. By accomplishing this, I can improve the image quality, enable the quantitative reconstruction ability of the single-spectrum CT scanner, and potentially reduce unnecessary radiation dose to patients.

In this thesis, in order to obtain accurate spectrum for poly-energetic reconstruction, I first presented a novel spectral estimation technique, with which spectra across a large range of angular trajectories of the imaging field of view can be estimated with a single phantom and a single axial acquisition. The experimental results with a 16 cm diameter cylindrical phantom (composition: ultra-high-molecular-weight polyethylene [UHMWPE]) on a clinical scanner showed that the averaged absolute mean energy differences and the normalized root mean square differences with respect to the actual spectra across kVp settings (i.e., 80, 100, 120, 140) and angular trajectories were less than 0.61 keV and 3.41%, respectively

With the previous estimation of the x-ray spectra, three poly-energetic reconstruction algorithms are proposed for different clinical applications. The first algorithm (i.e., poly-energetic iterative FBP [piFBP]) can be applied to routine clinical CT exams, as the spectra of the x-ray source and the nonlinear attenuations of diverse body tissues and metal implant materials are incorporated to eliminate BHAs and to reduce metal artifacts. The simulation results showed that the variation range of the relative errors of various tissues across different phantom sizes (i.e., 16, 24, 32, and 40 cm in diameter) and kVp settings (80, 100, 120, 140) were reduced from [-7.5%, 17.5%] for conventional FBP to [-0.1%, 0.1%] for piFBP, while the noise was maintained at the same low level (about [0.3%, 1.7%]).

When iodinated contrast agents are involved and patient motions are not readily correctable (e.g., in myocardial perfusion exam), a second algorithm (i.e., poly-energetic simultaneous algebraic reconstruction technique [pSART]) can be applied to eliminate BHAs and to quantitatively determine the iodine concentrations of blood-iodine mixtures with our new technique. The phantom experiment on a clinical CT scanner indicated that the maximum absolute relative error across material inserts was reduced from 4.1% for conventional simultaneous algebraic reconstruction technique [SART] to 0.4% for pSART.

Extending the work beyond minimizing BHAs, if patient motions are correctable or negligible, a third algorithm (i.e., poly-energetic dynamic perfusion algorithm [pDP]) is developed to retrieve iodine maps of any iodine-tissue mixtures in any perfusion exams, such as breast, lung, or brain perfusion exams. The quantitative results of the simulations with a dynamic anthropomorphic thorax phantom indicated that the maximum error of iodine concentrations can be reduced from 1.1 mg/cc for conventional FBP to less than 0.1 mg/cc for pDP.

Two invention disclosure forms based on the work presented in this thesis have been submitted to Office of Licensing & Ventures of Duke University.

Since its inception in the 1970's, computed tomography (CT) has revolutionized the practice of medicine and evolved into an essential tool for diagnosing numerous diseases not only in adults but also in children. The clinical utility of CT examinations has led to a rapid expansion in CT use and a corresponding increase in the radiation burden to patients. CT radiation is of particular concern to children, whose rapidly growing tissues are more susceptible to radiation-induced cancer and who have longer life spans during which cancerous changes might occur. In recent years, the increasing awareness of CT radiation risk to children has brought about growing efforts to reduce CT dose to the pediatric population. The key element of all dose reduction efforts is to reduce radiation dose while maintaining diagnostic accuracy. Substantiating the tradeoff between the two is the motivation behind this dissertation work.

The first part of this dissertation involved the development of an accurate method for estimating patient-specific radiation dose and potential cancer risk from CT examinations. A Monte Carlo program was developed and validated for dose simulation in a state-of-the-art CT system. Combined with realistic computer models of patients created from clinical CT data, the program was applied to estimate patient-specific dose from pediatric chest and abdomen-pelvic CT examinations and to investigate the dose variation across patients due to the variability of patient anatomy and body habitus. The Monte Carlo method was further employed to investigate the effects of patient size and scan parameters on dose and risk for the entire pediatric population.

The second part of this dissertation involved the development of tools needed to study the diagnostic accuracy of small lung nodules on pediatric CT images. A prior method for modeling two-dimensional symmetric liver/lung lesions was extended to create three-dimensional nodules with asymmetric shapes and diffused margins. A method was also developed to estimate quantum noise in the lung region of a CT image based on patient size.

The last part of this dissertation involved assessment of diagnostic accuracy using receiver operating characteristic (ROC) observer experiments. A pilot study of 13 pediatric patients (1-7 years old) was first conducted to evaluate the effect of tube current on diagnostic accuracy, as measured by the area under the ROC curve (Az). A study of 30 pediatric patients (0-15 years old) was then conducted to assess protocol- and scanner-independent relationships between image quality (nodule detectability and noise) and diagnostic accuracy. The relationships between diagnostic accuracy and nodule detectability, between noise and scan parameters, and between dose/risk and scan parameters were lastly combined to yield the relationship between diagnostic accuracy and dose/risk.

For pediatric patients in the same weight/protocol group, organ dose variation across patients was found to be generally small for large organs in the scan coverage (< 10%), larger for small organs in the scan coverage (1-18%), and the largest for organs partially or completely outside the scan coverage (6-77%). Across the entire pediatric population, dose and risk associated with a chest scan protocol decreased exponentially with increasing patient size. The average chest diameter was found to be a stronger predictor of dose and risk than weight and total scan length.

The effects of bowtie filter and beam collimation on dose and risk were small compared to the effects of helical pitch and tube potential. The effects of any scan parameter were patient size-dependent, which could not be reflected by the difference in volume-weighted CT dose index (CTDIvol).

Over a nodule detectability (product of nodule peak contrast and display diameter to noise ratio) range of approximately 52-374 mm with an average of 143 mm, tube current or dose had a weak effect on the diagnostic accuracy of lung nodules. The effect of 75% dose reduction was comparable to inter-observer variability, suggesting a potential for dose reduction.

Diagnostic accuracy increased with increasing nodule detectability over the range of 25-374 mm, but reached a plateau beyond a threshold of ~ 99 mm. The trend was analogous to the relationship between Az and signal-to-noise ratio and suggested that the performance of the radiologists saturates (or increases slowly) beyond a threshold nodule detectability level; further reducing noise or increasing contrast to improve nodule detectability beyond the threshold yields little gain in diagnostic accuracy.

For a typical product of nodule contrast and physical diameter (1400 HU·mm) and a set of most commonly used scan parameters (tube potential of 120 kVp, helical pitch of 1.375, slice thickness of 5 mm, gantry rotation period of 0.4 second, image pixel size of 0.48 mm), diagnostic accuracy increased with effective dose and effective risk for a given patient size, but reached a plateau beyond a threshold dose/risk level. At a given effective dose, Az increased with decreasing patient size, i.e., the dose needed to achieve the same noise and hence diagnostic accuracy increased with patient size. To achieve an Az of 0.90, the dose needed for a 22-cm diameter (male) patient was about quadruple of that for a 10-cm diameter patient. While the effective risk associated with achieving the same diagnostic accuracy also increased with patient size, the risk associated with an Az of 0.90 was only twice as high for a 22-cm diameter (male) patient than for a 10-cm diameter patient due to the older age of the larger patient.

The research in this dissertation has two important clinical implications. First, the quantitative relationships between patient dose/risk and patient size, between patient dose/risk and scan parameters, between diagnostic accuracy and image quality, and between diagnostic accuracy and radiation dose can guide the design of pediatric CT protocols to achieve the desired diagnostic accuracy at the minimum radiation dose. Second, patient-specific dose and risk information, when included in a patient's dosimetry and medical records, can inform healthcare providers of prior radiation exposure and aid in decisions for image utilization, including the situation where multiple examinations are being considered.

X-ray computed tomography (CT) forms the basis for attenuation corrected positron emission tomography (PET) using combined PET/CT scanners. With concerns of high radiation exposure to patients through widespread use of CT, the lowest photon flux that will provide uniform attenuation correction for PET to within 5% over a range of body sizes was investigated. Additionally, clinical uniformity measurements are performed on a uniform phantom, but their results may not be applicable as an estimate of error of hot lesions. PET simulations of variability and localized error were performed with and without hot lesions using a tapering phantom. Images were reconstructed using a variety of fixed and modulated tube-current CT scans and various levels of scatter correction. A physical phantom was designed and scanned to augment the simulation results. Attenuation correction of uniform images was within 5% error when using 120 kVp using a noise index of 50 and 140 kVp using a noise index of 50 for all phantom sizes. Variability with hot lesions was within 5% for scans using 120 kVp and greater than 24 mAs for 21.9 cm and 31.7 cm effective diameters and greater than 48 mAs for 38.5 cm effective diameter. Variability was worse in the background than on hot lesions for poor attenuation correction and poor scatter correction cases. Background error overestimates the error in hot lesions when attenuation correction is biased. Variability was within 5% when estimation of scatter magnitude was within 20% of its true value both with and without hot lesions. Errors in background due to under and overcorrected scatter lead to an over and underestimate of hot lesion errors, respectively. Physical phantom uniformity was within 5% when using 120 kVp and 10 mAs, albeit with a much smaller phantom size. The background error and its underestimation of lesion error was also measured in the physical phantom.