Chronic diseases such as cancer and cardiovascular disease are among the leading causes of death in industrialized nations such as the United States. In addition, many other chronic diseases such as osteoarthritis result in disability and reduced quality of life for even more people, especially as they age. Imaging modalities, such as single-photon emission computed tomography (SPECT) magnetic resonance imaging (MRI), and x-ray computed tomography (CT), have a unique advantage in the area of medical research relative to their classical counterparts such as histology and gross pathology. Noninvasive imaging methodologies are an attractive alternative for obtaining physiologic assessments without disruption of the phenomenon of interest.

In response to the importance placed on noninvasive small animal studies at the biological and medical research fronts, the focus of this work is on the advancement of imaging technologies which will augment and extend the current capabilities of noninvasive imaging systems. In the initial portion of this thesis, we will focus on the advancement of SPECT and CT techniques used to measure perfusion and those biological factors to which it is correlated, including improvements in physiological monitoring, dual-isotope imaging, and dual-modality imaging. In addition, we will complement our design focus with an illustration of the effectiveness of the small animal imaging approach to the study of osteoarthritis using MRI. We will illustrate the utility of imaging techniques via an ex vivo MRI study of human osteochondral specimens, and then we will apply them to study the development of osteoarthritis in vivo in a rabbit model.

Musculoskeletal disease is the leading cause of disability worldwide, with the 2019 Global Burden of Disease study reporting global disease prevalence of approximately 1.714 billion1. X- ray and magnetic resonance imaging (MRI) are routinely used for clinical diagnosis and monitoring of musculoskeletal disease, however, due to an increasing volume of acquired images and limited time, image assessments are mainly qualitative. This thesis aims to elevate the role of imaging in the assessment of musculoskeletal disease by developing fully automatic image analysis tools to improve image analysis sensitivity, speed, and/or precision. We target the two conditions with the highest prevalence and healthcare expenditure in the United States: knee osteoarthritis (OA) and back pain. We use deep learning to develop fully automatic tools for image analysis and demonstrate their utility in the assessment and analysis of research and clinical datasets. I will be presenting four main projects:(1) A deep learning segmentation method for quantitative analysis of knee cartilage from structural MRI to conduct longitudinal analysis on cartilage thickness over 8 years (2) A point cloud algorithm for feature learning from structural and compositional knee MRI to assess the importance of shape and composition features in predicting OA onset (3) A registration pipeline for voxel-based analysis of MR imaging of the lumbar spine to examine local associations between T1ρ, T2, and patient reported outcomes (4) A curve extraction algorithm for analysis of global spine shape from x-ray imaging to build a shape model that examines 3D spine shape variations in the UCSF patient population

With the aging of the general population, musculoskeletal (MSK) diseases have moved to the forefront of healthcare concerns and are the leading causes of disability globally. Noninvasive imaging is routinely utilized in the clinic to diagnose and monitor onset and progression of MSK conditions. However, due to the qualitative nature of imaging assessments and increasing labor costs of evaluating advanced imaging modalities, there is a crucial need for automatic quantitative approaches. In this dissertation, we explore the development of computer vision techniques for extracting morphological features associated with low back pain and knee osteoarthritis, two of the most prevalent and debilitating MSK conditions.

We begin by addressing the costs of image annotation via automation with deep learning. More specifically, we developed convolutional neural networks for two purposes: (1) to semantically segment various tissues, allowing for geometric tissue characterization, and (2) to detect and localize lesions and abnormalities. Then, leveraging these models for feature extraction, we harmonized tissue geometries in 3D Euclidean space using atlas-based registration to identify tissue shapes predisposed to disease onset. These techniques were applied to both large-scale and small, limited datasets, demonstrating the utility of computer vision techniques for morphological evaluation in a data-driven, exploratory manner.

This thesis developed high signal-to-noise ratio (SNR) yielding magnetic resonance imaging (MRI) methods that require relatively short measurement times. The methods were primarily aimed at addressing existing technical limitations of low SNR and long scan times (~ 20 minutes) in the field of in vivo high resolution (HR) imaging of trabecular bone micro-architecture. HR-MRI of trabecular bone provides a non-invasive way of monitoring trabecular bone structural integrity for assessment of the disease condition, osteoporosis. Osteoporosis is a debilitating skeletal disorder affecting 1 in 3 women and 1 in 12 men over the age of 50 worldwide and is characterized by loss of bone mass and structure leading to atraumatic fractures at the vertebrae, hip, wrists and other sites. SNR limitations in trabecular bone MRI were overcome by incorporating a high magnetization yielding pulse sequence and translating the imaging protocol to the higher field strength of 3 Tesla (T) from the clinical standard of 1.5 T. A simulation model was developed to estimate SNR in bone tissue. The feasibility of imaging trabecular bone structure in vivo at the proximal femur (hip), until then an SNR impeded application, was demonstrated. Next, an autocalibrating parallel imaging (PI) method was implemented to accelerate data acquisition. Using an eight channel coil array, scan time was reduced 2-4 fold without any significant impact on image resolution or edge sharpness; although bone structural measures derived from the accelerated images showed overestimation. Consequently, parallel reconstruction and image processing algorithms were developed to address the causes of overestimation. With the availability of whole body 7 T magnets and the motivation for further improvement in SNR, the PI methods were adapted to the even higher field strength and its performance was found to improve relative to 3 T. To summarize, this thesis developed SNR efficient MR methods that facilitate quantitative HR-MRI of trabecular bone within 2-10 minutes, thus increasing the clinical feasibility of this application. Additionally, the PI methods were applied to diverse applications such as cartilage imaging for osteoarthritis; and susceptibility weighted imaging of brain vasculature and 3D spectroscopic imaging of brain metabolites in brain tumor patients.

HR-pQCT enables noninvasive in vivo characterization of trabecular bone structure at peripheral sites. It can be a powerful tool for assessment of skeletal status and monitoring effect of therapy. The overall objective of this dissertation is two-fold: (i) to characterize errors in the indices describing trabecular structure introduced by partial volume averaging (PVA) and motion artifacts, and (ii) to develop a quantitative technique for estimating error introduced by motion image degradation and a regional analysis technique to account for the inherit spatial heterogeneity of bone structure at the distal radius and tibia.

The first study quantified the degree of error introduced to the topological indices of trabecular bone structure by PVA at in vivo resolution of 82 µm. The error in structure model index increased with resampled spatial resolution in a structure-type dependent manner. Connectivity density and degree of anisotropy were constantly underestimated at 82 µm.

The second study proposed a non-subjective technique for measuring subject motion, in which the parallelized projection images at 0° and 180° were compared using sum of squared intensity difference (SSD). SSD correlated strongly to motion-induced error in density, cortical geometric, and trabecular structural indices at the distal radius and tibia.

The third study characterized spatial variability in trabecular structure at the distal radius and tibia by subdividing the cross-section to eight subregions. In addition to substantial regional variations, the subregions with pronounced gender and age effects were identified.

The fourth study applied subregional analysis to a placebo-controlled randomized pilot study. Significant treatment effects on cortical geometry in the lateral quadrant of the distal radius and the posterior quadrant of the distal tibia were detected in subjects treated with an anti-fracture therapy (alendronate) for 2 years, which was not detected by conventional global analysis.

The results of this work provide a basis for understanding errors in trabecular indices derived from HR-pQCT images as well as a frame-work for quantitative motion detection and regional analysis. The results also serve as a guideline for standardizing protocols across studies, scanners, and study centers, to promote HR-pQCT as an effective tool for assessing skeletal status in research and clinical studies.

Diffusion Imaging has been primarily focused on brain application with limited applications in the knee. One limitation of diffusion imaging in the knee is the long TE (40-60 ms) in most of the sequences that have been used. While this is not a detriment

in brain, it can be a problem in the knee where several tissues have short T2 relaxation times including the cartilage (32 ms) and meniscus (11 ms). Thus imaging of the knee with a short TE diffusion sequence would substantially increase signal to noise, which would in turn be applied to improve diffusion measurements in meniscus and cartilage. Research has shown that diffusion weighted imaging in knee has a strong potential as a biomarker and can act as a new and potent investigation tool for tissue integrity of meniscus and for early diagnosis of cartilage degeneration.

A new sequence for diffusion weighted imaging of knee at 3T has been proposed and evaluated. The proposed stimulated echo with MAPSS acquisition sequence is more signal efficient than the conventional spin echo sequence and can image the whole

knee volume in half the acquisition time compared to the most commonly used line scan sequence. The sequence was tested in phantoms, ex-vivo specimens and in-vivo knees with encouraging results. Further optimization and validation of the sequence is

proposed for successful acquisition of diffusion values in knee cartilage and meniscus in healthy volunteers and osteoarthritis patient cohorts.

The extracellular matrix (ECM) is a network of extracellular macromolecules that is present in both cartilage and intervertebral disc tissues. Osteoarthritis (OA) is a debilitating disease causing pain and stiffness in approximately 14% of adults. Intervertebral disc degeneration (IVDD) is the leading cause of pain and disability in adults in the United States. Despite the high prevalence of OA and IVDD worldwide, diagnosis in the early stages of symptomatic disease is elusive in clinical practice. Certain quantitative MRI methods including T1rho and T2 relaxation time mapping are sensitive to biochemical changes in the ECM, and thus may be valuable for early diagnosis of OA and IVDD. The purpose of this dissertation is to develop a methodology for the quantification of T1rho and T2 magnetic resonance relaxation times in the intervertebral disc and knee cartilage, and characterize the spatial distribution of T1rho and T2 in the extracellular matrix in IVDD and OA.

This study evaluated the impact of signal to noise ratio and T2 fitting algorithms on the T2 quantification. Simulations were performed determine the minimum signal to noise ratio that can be used to distinguish healthy cartilage from degenerative cartilage. In addition, the mean and spatial distribution of cartilage T2 in subjects with and without OA was evaluated. The mean T2 values and their entropy were greater in OA patients than in controls, indicating that the T2 values in osteoarthritic cartilage are not only elevated, but also more heterogeneous than those in healthy cartilage. In addition, the entropy of cartilage T2 at baseline was associated with an increase in pain scores over two years.

MRI T1rho relaxation time was investigated in subjects with IVDD. The results demonstrated positive correlations between T1rho and pain scores and negative correlations between T1rho and degenerative grade. This study suggests that T1rho may be sensitive to early degenerative changes and clinical symptoms in IVDD.

The results of this project suggest that quantifying the spatial distribution of MRI relaxation times improves the clinical assessment of OA and IVDD, by providing a non-invasive evaluation of biochemical composition in cartilage and intervertebral disc tissues.

Knee osteoarthritis (OA) is a multifaceted diseases affecting many older adults. A main physiological characteristic of OA is the progressive loss of articular hyaline cartilage. Recently, there have been many proposed techniques that attempt to quantify early hyaline cartilage breakdown associated with OA. Preliminary stages of OA are marked by a reduction in PG and GAG followed by the degradation of the collagen network. Many techniques utilizing magnetic resonance imaging (MRI) have been proposed to noninvasively probe biochemical changes in human osteoarthritic cartilage. T1ρ and T2 relaxation time mapping may be useful as a noninvasive quantification of cartilage degradation. Moreover, osteoarthritic cartilage has displayed elevated voxel spatial heterogeneity of cartilage T1ρ and T2 relaxation time, as calculated by the grey-level co-occurrence (GLCM).

In this study we monitored cross-sectional and longitudinal evaluation of T1ρ and T2 relaxation times, GLCM spatial heterogeneity (texture analysis) of T1ρ and T2 relaxation, and cartilage volume and thickness in OA subjects versus healthy controls. Our results display significant increases over time in GLCM entropy in the medial tibia for both T1ρ and T2 relaxation and in the medial femoral condyle for T1ρ relaxation. A significant decrease in volume and thickness was observed in the medial tibia was also noted. These results indicate medial knee OA cartilage experiences a heightened level of spatial heterogeneity, possibly suggesting a predicative utility of T1ρ and T2 relaxation time texture analysis in the development of medial knee OA.

Low back pain is an expensive, widespread healthcare concern. The mechanisms of its progression and association with intervertebral disc degeneration are not fully understood, but recent studies suggest that lactate accumulation and a subsequent drop in pH may be initiating events. Chemical exchange saturation transfer (CEST) of glycosaminoglycans (gagCEST) has emerged as way to quantify glycosaminoglycan (GAG) concentrations in the intervertebral disc, but no studies have examined its dependency on pH. This study seeks to assess the pH-dependence of gagCEST and use iopromide, a common contrast agent used in CT imaging, as a pH-sensitive CEST probe to explore these agents' potential to measure pH of the intervertebral discs. We first create chondroitin sulfate and Ultravist® phantoms over a range of pH values to explore the pH-dependency of the CEST imaging of these probes and apply these findings to porcine intervertebral disc specimens. Our results demonstrate the non-linear dependence of gagCEST on pH and a linear regression of the Iopromide CEST effect with pH (R2 = 0.95). Iopromide was then infused into the disc and the calibration created by the phantom studies was used to determine pH in the disc.

These findings provide what is to our knowledge the first description of the pH dependence of gagCEST imaging and the first use of the iopromide contrast agent in the CEST MR imaging of the intervertebral disc specimen. Because iopromide CEST imaging is independent of the local concentration of macromolecules, it particularly shows great potential in reporting pH in intervertebral disc specimen studies. The ability to report pH in a tissue non-invasively by one of these methods could be valuable in better understanding disease progression.