Statistical iterative reconstruction (SIR) algorithms for x-ray computed tomography (CT) have the potential to reconstruct images with less noise and systematic error than the conventional filtered backprojection (FBP) algorithm. More accurate reconstruction algorithms are important for reducing imaging dose and for a wide range of quantitative CT applications. The work presented herein investigates some potential advantages of one such statistically motivated algorithm called Alternating Minimization (AM). A simulation study is used to compare the tradeoff between noise and resolution in images reconstructed with the AM and FBP algorithms. The AM algorithm is employed with an edge-preserving penalty function, which is shown to result in images with contrast-dependent resolution. The AM algorithm always reconstructed images with less image noise than the FBP algorithm. Compared to previous studies in the literature, this is the first work to clearly illustrate that the reported noise advantage when using edge-preserving penalty functions can be highly dependent on the contrast of the object used for quantifying resolution. A polyenergetic version of the AM algorithm, which incorporates knowledge of the scanner’s x-ray spectrum, is then commissioned from data acquired on a commercially available CT scanner. Homogeneous cylinders are used to assess the absolute accuracy of the polyenergetic AM algorithm and to compare systematic errors to conventional FBP reconstruction. Methods to estimate the x-ray spectrum, model the bowtie filter and measure scattered radiation are outlined which support AM reconstruction to within 0.5% of the expected ground truth. The polyenergetic AM algorithm reconstructs the cylinders with less systematic error than FBP, in terms of better image uniformity and less object-size dependence. Finally, the accuracy of a post-processing dual-energy CT (pDECT) method to non-invasively measure a material’s photon cross-section information is investigated. Data is acquired on a commercial scanner for materials of known composition. Since the pDECT method has been shown to be highly sensitive to reconstructed image errors, both FBP and polyenergetic AM reconstruction are employed. Linear attenuation coefficients are estimated with residual errors of around 1% for energies of 30 keV to 1 MeV with errors rising to 3%-6% at lower energies down to 10 keV. In the ideal phantom geometry used here, the main advantage of AM reconstruction is less random cross-section uncertainty due to the improved noise performance.

Accelerated tumor cell repopulation during radiation therapy is one of the leading causes for low survival rates of head-and-neck cancer patients. The therapeutic effectiveness of radiotherapy could be improved by selectively targeting proliferating tumor subvolumes with higher doses of radiation. Positron emission tomography (PET) imaging with 3´-deoxy-3´-[18F]-fluorothymidine (FLT) has shown great potential as a non-invasive approach to characterizing the proliferation status of tumors. This thesis focuses on histopathological validation of FLT PET imaging specifically for image-guidance applications in biologically adaptive radiotherapy. The lack of experimental data supporting the use of FLT PET imaging for radiotherapy guidance is addressed by developing a novel methodology for histopathological validation of PET imaging. Using this new approach, the spatial concordance between the intratumoral pattern of FLT uptake and the spatial distribution of cell proliferation is demonstrated in animal tumors. First, a two-dimensional analysis is conducted comparing the microscopic FLT uptake as imaged with autoradiography and the distribution of active cell proliferation markers imaged with immunofluorescent microscopy. It was observed that when tumors present a pattern of cell proliferation that is highly dispersed throughout the tumor, even high-resolution imaging modalities such as autoradiography could not accurately determine the extent and spatial distribution of proliferative tumor subvolumes. While microscopic spatial coincidence between high FLT uptake regions and actively proliferative subvolumes was demonstrated in tumors with highly compartmentalized/aggregated features of cell proliferation, there were no conclusive results across the entire set of utilized tumor specimens. This emphasized the need for addressing the limited resolution of FLT PET when imaging microscopic patterns of cell proliferation. This issue was emphasized in the second part of the thesis where the spatial concordance between volumes segmented on FLT simulated FLT PET images and the three dimensional spatial distribution of cell proliferation markers was analyzed.

Dose verification is a critical component of adaptive radiotherapy, as it provides a measurement of treatment delivery success. Based on the measured outcome, the plan may be adapted to account for differences between the planned dose and the delivered dose. Although placement of an EPID behind the patient during treatment allows for exit dosimetry which may be used to reconstruct the delivered patient dose via backprojection of the fluence, there have not been any studies examining the basic assumption of backprojection-based dose verification: that deviations between the expected and delivered exit fluences are totally caused by errors in the delivered fluence, and not caused by patient geometry changes. In this dissertation, the validity of this assumption is tested. Exit fluence deviations caused by machine fluence delivery errors are measured as well as those caused by interfractional changes in the patient anatomy. Dose reconstruction errors resulting from the backprojection assumption are assessed. Correlations are examined between exit fluence deviations and patient dose reconstruction deviations. Based on these correlations, a decision tree is proposed detailing when caution should be taken in performing dose reconstruction to achieve delivery verification. Finally, a semi-automated dose verification tool is constructed for both clinical and research purposes.

The objective of this dissertation is to develop treatment planning techniques that have the potential to improve radiation therapy of time-dependent (4D) anatomy. Specifically, this study examines dose estimation, dose evaluation, and decision making in the context of optimizing lung cancer radiation therapy. Two methods of dose estimation are compared in patients with locally advanced and early stage lung cancer: dose computed on a single image (3D-dose) and deformably registered, accumulated dose (or 4D-dose). The results indicate that differences between 3D- and 4D- dose are not significant in organs at risk (OARs), however, 4D-dose to a moving lung cancer target can deviate from 3D-dose. These differences imply that optimization of the 4D-dose through multiple-anatomy optimization (MAO) can improve radiation therapy in 4D-anatomy. MAO incorporates time-dependent target and OAR geometry while enabling a simple, clinically realizable delivery. MAO has the potential to enhance the therapeutic ratio in terms of target coverage and OAR sparing in 4D-anatomy. In dose evaluation within 4D-anatomy; dose-to-mass is a more intuitive and precise metric in estimating the effects of radiation in tissues. Assuming physical density is proportional to functional tissue density, dose-to-mass has a 1-1 correspondence with radiation damage. Dose-to-mass optimization boosts dose in massive regions of lung cancer targets and can reduce integral dose to lung by preferentially treating through regions of low-density lung tissue. Finally, multi-criteria optimization (MCO) is implemented in order to clarify decision making during plan design for lung cancer treatment. An MCO basis set establishes a patient-specific decision space which reveals trade-offs in OAR-dose at a fixed, constrained target dose. By interpolating the MCO basis set and evaluating the plan on 4D-anatomy, patient- and organ- specific conservatism in plan design can be expressed in real time. Through improved methods of dose estimation, dose evaluation, and decision making, this dissertation will positively impact radiation therapy of time-dependent anatomy.

The development of hybrid PET/MRI imaging systems needs to be paralleled with the development of a hybrid intrinsic PET/MRI probes. The aim of this work was to develop and validate a novel radio-superparamagnetic nanoparticle (r-SPNP) for hybrid PET/MRI imaging. This was achieved with the synthesis of superparamagnetic iron oxide nanoparticles (SPIONs) that intrinsically incorporated 59Fe and manganese iron oxide nanoparticles (MIONs) that intrinsically incorporated 52Mn. Both [59Fe]-SPIONs and [52Mn]-MIONs were produced through thermal decomposition synthesis. The physiochemical characteristics of the r-SPNPs were assessed with TEM, DLS, and zeta-potential measurements, as well as in imaging phantom studies. The [59Fe]-SPIONs were evaluated in vivo with biodistribution and MR imaging studies. The biodistrubution studies of [59Fe]-SPIONs showed uptake in the liver. This corresponded with major MR signal contrast measured in the liver. 52Mn was produced on natural chromium through the 52Cr(p,n)52Mn reaction. The manganese radionuclides were separated from the target material through a liquid-liquid extraction. The αVβ3 integrin binding of [52Mn]-MION-cRGDs was evaluated with αVβ3 integrin solid phase assays, and the expression of αVβ3 integrin in U87MG xenograft tumors was characterized with fluorescence flow cytometry. [52Mn]-MION-cRGDs were used for in vivo PET and MR imaging of U87MG xenograft tumor bearing mice. PET data showed increased [52Mn]-MION-cRGD uptake compared with untargeted [52Mn]-MIONs. ROI analysis of PET and MRI data showed that MR contrasted corresponded with PET signal. Future work will utilize [52Mn]-MION-cRGDs in other tumor models and with hybrid PET/MRI imaging systems.

Platinum based chemotherapy is amongst the mainstream DNA-damaging agents used in clinical cancer therapy today. Agents such as cisplatin, carboplatin are clinically prescribed for the treatment of solid tumors either as single agents, in combination, or as part of multi-modality treatment strategy. Despite the potent anti-tumor activity of these drugs, overall effectiveness is still hampered by inadequate delivery and retention of drug in tumor and unwanted normal tissue toxicity, induced by non-selective accumulation of drug in normal cells and tissues. Utilizing molecular imaging and nanoparticle technologies, this thesis aims to contribute to better understanding of how to improve the profile of platinum based therapy. By developing a novel fluorinated derivative of carboplatin, incorporating a Flourine-18 (18F) moiety as an inherent part of the molecule, quantitative measures of drug concentration in tumors and normal tissues can be directly determined in vivo and within the intact individual environment. A potential impact of this knowledge will be helpful in predicting the overall response of individual patients to the treatment. Specifically, the aim of this project, therefore, is the development of a fluorinated carboplatin drug derivative with an inherent positron emission tomography (PET) imaging capability, so that the accumulation of the drug in the tumor and normal organs can be studied during the course of therapy . A secondary objective of this research is to develop a proof of concept for simultaneous imaging of a PET radiolabeled drug with a SPECT radiolabeled liposomal formulation, enabling thereby bi-modal imaging of drug and delivery vehicle in vivo. The approach is challenging because it involves development in PET radiochemistry, PET and SPECT imaging, drug liposomal encapsulation, and a dual-modal imaging of radiolabeled drug and radiolabeled vehicle. The principal development is the synthesis of fluorinated carboplatin 19F-FCP using 2-(5-fluoro-pentyl)-2-methyl malonic acid as the labeling agent to coordinate with the cisplatin aqua complex. It was then used to treat various cell lines and compared with cisplatin and carboplatin at different concentrations ranging from 0.001 µM to 100 µM for 72 hrs and 96 hrs. IC50 values calculated from cell viability indicated that 19F-FCP is a more potent drug than Carboplatin. Manual radiosynthesis and characterization of [18F]-FCP was performed using [18F]-2-(5-fluoro-pentyl)-2-methyl malonic acid with coordination with cisplatin aqua complex. Automated radiosynthesis of [18F]-FCP was optimized using the manual synthetic procedures and using them as macros for the radiosynthesizer. [18F]-FCP was evaluated in vivo with detailed biodistribution studies and PET imaging in normal and KB 3-1 and KB 8-5 tumor xenograft bearing nude mice. The biodistribution studies and PET imaging of [18F]-FCP showed major uptake in kidneys which attributes to the renal clearance of radiotracer. In vivo plasma and urine stability demonstrated intact [18F]-FCP. [111In]-Labeled Liposomes was synthesized and physiochemical properties were assessed with DLS. [111In]-Labeled Liposome was evaluated in vivo with detailed pharmacokinetic studies and SPECT imaging. The biodistribution and ROI analysis from SPECT imaging showed the spleen and liver uptake of [111In]-Labeled Liposome and subsequent clearance of activity with time. [18F]-FCP encapsulated [111In]-Labeled Liposome was developed and physiochemical properties were characterized with DLS. [18F]-FCP encapsulated [111In]-Labeled Liposome was used for in vivo dual tracer PET and SPECT imaging from the same nanoconstruct in KB 3-1 (sensitive) and COLO 205 (resistant) tumor xenograft bearing nude mice. PET imaging of [18F]-FCP in KB 3-1 (sensitive) and COLO 205 (resistant) tumor xenograft bearing nude mice was performed. Naked [18F]-FCP and [18F]-FCP encapsulated [111In]-Labeled Liposome showed different pharmacokinetic profiles. PET imaging of [18F]-FCP showed major uptake in kidneys and bladder. However, [18F]-FCP encapsulated [111In]-Labeled Liposome showed major uptake in RES in both PET and SPECT images. ROI analysis of SPECT image enabled by 111In corresponded with PET image enabled by 18F demonstrating the feasibility of dual tracer imaging from the single nanoconstruct. Future work involves the intensive in vitro characterization of [18F]-FCP encapsulated [111In]-Labeled Liposome and detailed in vivo evaluation of [18F]-FCP encapsulated [111In]-Labeled Liposome in various tumor models.

In radiation therapy, interfraction organ motion introduces a level of geometric uncertainty into the planning process. Plans, which are typically based upon a single instance of anatomy, must be robust against daily anatomical variations. For this problem, a model of the magnitude, direction, and likelihood of deformation is useful. In this thesis, principal component analysis (PCA) is used to statistically model the 3D organ motion for 19 prostate cancer patients, each with 8-13 fractional computed tomography (CT) images. Deformable image registration and the resultant displacement vector fields (DVFs) are used to quantify the interfraction systematic and random motion. By applying the PCA technique to the random DVFs, principal modes of random tissue deformation were determined for each patient, and a method for sampling synthetic random DVFs was developed. The PCA model was then extended to describe the principal modes of systematic and random organ motion for the population of patients. A leave-one-out study tested both the systematic and random motion model’s ability to represent PCA training set DVFs. The random and systematic DVF PCA models allowed the reconstruction of these data with absolute mean errors between 0.5-0.9 mm and 1-2 mm, respectively. To the best of the author’s knowledge, this study is the first successful effort to build a fully 3D statistical PCA model of systematic tissue deformation in a population of patients. By sampling synthetic systematic and random errors, organ occupancy maps were created for bony and prostate-centroid patient setup processes. By thresholding these maps, PCA-based planning target volume (PTV) was created and tested against conventional margin recipes (van Herk for bony alignment and 5 mm fixed [3 mm posterior] margin for centroid alignment) in a virtual clinical trial for low-risk prostate cancer. Deformably accumulated delivered dose served as a surrogate for clinical outcome. For the bony landmark setup subtrial, the PCA PTV significantly (p30, D20, and D5 to bladder and D50 to rectum, while increasing rectal D20 and D5. For the centroid-aligned setup, the PCA PTV significantly reduced all bladder DVH metrics and trended to lower rectal toxicity metrics. All PTVs covered the prostate with the prescription dose.

External-beam radiotherapy is one of the primary methods for treating cancer. Typically a radiotherapy treatment course consists of radiation delivered to the patient in multiple daily treatment fractions over 6-8 weeks. Each fraction requires the patient to be aligned with the image acquired before the treatment course used in treatment planning. Unfortunately, patient alignment is not perfect and results in residual errors in patient setup. The standard technique for dealing with errors in patient setup is to expand the volume of the target by some margin to ensure the target receives the planned dose in the presence of setup errors. This work develops an alternative to margins for accommodating setup errors in the treatment planning process by directly including patient setup uncertainty in IMRT plan optimization. This probabilistic treatment planning (PTP) operates directly on the planning structure and develops a dose distribution robust to variations in the patient position. Two methods are presented. The first method includes only random setup uncertainty in the planning process by convolving the fluence of each beam with a Gaussian model of the distribution of random setup errors. The second method builds upon this by adding systematic uncertainty to optimization by way of a joint optimization over multiple probable patient positions. To assess the benefit of PTP methods, a PTP plan and a margin-based plan are developed for each of the 28 patients used in this study. Comparisons of plans show that PTP plans generally reduce the dose to normal tissues while maintaining a similar dose to the target structure when compared to margin-based plans. Physician assessment indicates that PTP plans are generally preferred over margin-based plans. PTP methods shows potential for improving patient outcome due to reduced complications associated with treatment.

In this thesis the utility of a bivariate gamma distribution is explored. In the field of nutritional epidemiology a nutrition density transformation is used to reduce collinearity. This phenomenon will be shown to result due to the independent variables following a bivariate gamma model. In the field of radiation oncology paired comparison of variances is often performed. The bivariate gamma model is also appropriate for fitting correlated variances. A method for simulating bivariate gamma random variables is presented. This method is used to generate data from several bivariate gamma models and the asymptotic properties of a test statistic, suggested for the radiation oncology application, is studied.

Usefulness of a particular clinical assay is directly correlated with its ability to extract highest possible signal from available data. This is particularly relevant for personalized radiation therapy since early plan modifications confer greater benefits to treatment outcome. Recent studies have demonstrated capability of single-cell Raman microscopy to detect cellular radiation response at clinical (below 10Gy) doses, but only in certain strongly responding cell lines and after at least two day incubation. One possible cause is rather unoptimized signal processing used. This work investigates application of several advanced multivariate methods - weighted principal component analysis (WPCA), robust PCA, probabilistic PCA, and nonlinear PCA to increase radiation response signal. Representative datasets from strongly (H460 - human lung) and weakly (LNCaP - human prostate) responding cell lines were analysed in 0-50Gy and 0-10Gy dose ranges and results quantified to determine relative and absolute algorithm performance. It was found that with careful tuning, significant improvements in sensitivity and better signal separation could be achieved as compared to conventional PCA. / Graduate