The distinct advantages of magnetic resonance imaging (MRI), including exquisite soft tissue contrast, diverse contrast mechanisms, and tomographic imaging with adjustable plane orientation, have made it a promising technology for guiding and directing interventions and surgeries. Nonetheless, MRI-guided percutaneous interventions encounter challenges in obtaining accurate, real-time 3D needle localization due to the intricate structure of biological tissue and the variability of needle features on in vivo MR images. This thesis aims to develop and assess a deep learning-based automatic pipeline for rapid 3D needle localization using intra-procedural 3D MRI.

Firstly, the pipeline incorporated Shifted-window UNEt TRansformers (Swin UNETR) for 3D needle feature segmentation on intra-procedural 3D MRI. Next, a post-processing method was developed to determine and extract the 2D reformatted image plane that passes through the main axis of the segmented 3D needle feature. Lastly, a 2D Swin Transformer network was adapted and trained for fine segmentation of needle features on reformatted 2D image planes. The 3D needle location was calculated based on the 2D coordinates of the needle feature tip and entry point on the 2D reformatted images.

The pipeline was evaluated using in vivo 3D MR images acquired during MRI-guided interventional experiments in pre-clinical pig models. The automatic pipeline achieved real-time and accurate 3D needle localization, with a needle tip and axis localization accuracy comparable to human intra-reader variations and a processing time of about 6 seconds from start to end. This pipeline can offer real-time visual assistance for physicians during percutaneous procedures, expediting the interventional workflow and holding the potential for enabling robotic-assisted MRI-guided interventions.

Image-guided minimally invasive percutaneous interventions, including needle-based targeted biopsy and focal therapy, play key roles in cancer diagnosis and treatment. With excellent soft tissue contrast and the absence of ionizing radiation, magnetic resonance imaging (MRI) has become a promising imaging modality for intra-procedural and real-time guidance of percutaneous interventions. However, MRI-guided percutaneous interventions still face two main challenges: limited access to patients inside the scanner and tissue displacement due to physiological motion. MRI-compatible remotely controlled systems for instrument manipulation inside the scanner are being developed to address the first challenge. To address the second challenge, computer-assisted navigation methods using intra-procedural and real-time MRI are being investigated to provide essential information regarding tissue and instrument positions to guide percutaneous interventions. Accurate tissue motion tracking and automatic needle localization techniques are the core components for computer-assisted navigation. Respiratory motion remains the main challenge for procedures in abdominal organs. The tissue target tracking accuracy is negatively impacted by the system latency of the MRI acquisition, reconstruction, and processing pipeline. In addition, passive needle tracking using MR images is challenged by variations of the needle-induced signal void feature in different situations. Discrepancies between the needle feature position on MRI and the underlying physical needle position could increase needle localization errors during procedures. Therefore, this dissertation aims to address these challenges by establishing new computer-assisted navigation techniques for MRI-guided interventions, including prediction of tissue motion due to respiration using fusion-based multi-rate Kalman filtering and deep learning-based needle localization methods.

Firstly, this work investigated image-based and surrogate-based motion tracking methods using real-time golden-angle radial MRI to achieve real-time MRI guidance for interventions in organs affected by respiration (e.g., liver). Images with different temporal footprints were reconstructed from the same golden-angle radial MRI data stream to simultaneously enable image-based and surrogate-based tracking at 10 Hz. Phantom experiments confirmed that the median online tracking error of image-based tracking was lower than surrogate-based methods, however, with higher median system latency. This work proposed a new fusion-based respiratory motion prediction framework to combine the lower tracking error of image-based tracking with the lower latency of surrogate-based tracking. The fusion-based method was evaluated in retrospective studies using in vivo real-time free-breathing liver MRI. The motion prediction accuracy of the proposed framework achieved low-latency feedback with improved accuracy compared with image-based and surrogate-based methods.

Next, to develop an automatic needle tracking algorithm for MRI-guided interventions, this work adapted the Mask Region Proposal-Based Convolutional Neural Network (R-CNN) to localize the passive needle features on MRI. Mask R-CNN was adapted and trained to segment the needle feature using intra-procedural images from MRI-guided prostate biopsy cases and real-time images from MRI-guided needle insertion in ex vivo tissue. The segmentation masks were passed into a needle feature localization algorithm to extract the needle feature tip location and axis orientation. The proposed algorithm was evaluated on MR images from in vivo intra-procedural prostate biopsy cases and ex vivo real-time MRI experiments with a range of different conditions. It achieved pixel-level tracking accuracy in real time and has the potential to assist MRI-guided percutaneous interventions.

Lastly, to overcome in-plane and through-plane discrepancies between the needle feature position on MRI and the underlying physical needle position, this work developed a deep learning-based framework to automatically localize the physical needle position using single-slice and 3- slice MRI. The proposed framework consists of two Mask R-CNN stages. Physics-based simulations were performed to generate single-slice and 3-slice images with needle features from a range of underlying needle positions and MRI parameters to form datasets for training the single-slice and 3-slice physical needle Mask R-CNN models. Using the single-slice model, the proposed physics-driven Mask R-CNN framework achieved sub-millimeter physical needle localization accuracy on single-slice images aligned with the needle. The 3-slice model further reduced the through-plane physical needle localization error in situations where the imaging plane may be misaligned with the needle. Both frameworks can achieve physical needle localization in real time for interventional MRI.

Purpose

Magnetic resonance imaging (MRI) can non-invasively quantify fat and the transverse relaxation rate (R2*) in the human body. This quantitative information can provide further insight about diseases such as non-alcoholic fatty liver disease (NAFLD), obesity, and ischemic placental disease (IPD). Conventional MRI methods for quantifying fat and R2* require breath-holding, which limits the spatial resolution, volumetric coverage, and signal-to-noise ratio that may be achieved. Moreover, several subject populations, including sick, elderly, and mentally impaired patients, as well as children, infants, and pregnant women, may have difficulty performing a breath-hold or are unable to breath-hold. The purpose of this work is to develop and evaluate a new free-breathing 3D stack-of-radial MRI technique (FB radial) for fat and R2* quantification at 3 Tesla (T) that overcomes the aforementioned limitations of conventional breath-holding MRI.

Methods

To enable free-breathing MRI, a multiecho golden-angle ordered 3D stack-of-radial radiofrequency-spoiled gradient echo sequence with gradient calibration and correction (FB radial) was developed. First, to evaluate FB radial without motion, fat quantification accuracy using FB radial was compared to conventional Cartesian and reference single-voxel magnetic resonance spectroscopy (SVS) sequences using a fat fraction phantom and in the pelvis of five healthy subjects at 3 T. To evaluate FB radial fat quantification accuracy in subjects capable of breath-holding, a population consisting of eleven healthy adults were recruited and imaged at 3 T. The fat quantification accuracy of FB radial was compared to conventional breath-held Cartesian (BH Cartesian) MRI and reference breath-held SVS (BH SVS).

The feasibility and repeatability of FB radial for hepatic fat quantification was evaluated in children, which represents a population that may have limited breath-hold ability or may have difficulty complying with operator instructions. Ten healthy children and nine overweight children with NAFLD, 7-17 years of age, were imaged at 3 T using FB radial, BH Cartesian and BH SVS. Acquisitions were performed twice to assess repeatability. Images and proton-density fat fraction (PDFF) maps were scored for image quality. Liver coverage was measured.

Ten healthy infants aged 2-7 months were recruited to evaluate the feasibility of FB radial for quantifying hepatic fat and body composition in a population incapable of breath-holding. The preparation time and scan time (median ± interquartile range) for each non-sedated MRI exam was recorded. Abdominal and head and chest FB radial scans and abdominal Cartesian scans were performed. Abdominal scans were scored for motion artifacts by a radiologist, masked to the trajectory. Visceral adipose tissue (VAT), subcutaneous adipose tissue (SAT), and brown adipose tissue (BAT) (volume and PDFF) and hepatic PDFF were measured using FB radial. Repeatability of FB radial hepatic PDFF was assessed.

To evaluate the quantitative accuracy of FB radial for R2* mapping without motion, FB radial was compared to a conventional Cartesian sequence using a R2* phantom. To evaluate FB radial R2* mapping in the presence of motion, thirty subjects with normal pregnancies and three subjects with ischemic placental disease (IPD) were scanned twice: between 14-18 and 19-23 weeks gestational age (GA). Feasibility and repeatability of FB radial placental R2* mapping was assessed. The mean and spatial coefficient of variation (CV) of placental R2* was determined for all subjects, and separately for anterior and posterior placentas, at each GA range.

For all analyses, quantitative accuracy of fat or R2* quantification was evaluated using linear correlation (Pearson’s correlation coefficient, r; Lin’s concordance correlation coefficient, ρc) and Bland-Altman analyses (mean difference, MD; limits of agreement, LoA = MD ± 1.96 ± standard deviation). The repeatability of FB radial between back-to-back scans for fat or R2* quantification was assessed by calculating the within-technique mean difference (MDwithin) and the coefficient of repeatability (CR). To compare image quality between FB radial and BH Cartesian, differences in the distribution of scores between FB radial and Cartesian were determined using McNemar-Bowker tests. For all statistical analyses, a p-value (P) < 0.05 was considered significant.

Results

In a fat fraction phantom, FB radial demonstrated accuracy with r and ρc > 0.995 (P < 0.001), absolute MD < 2.2 ± 4.9% compared to SVS and absolute MD < 0.6 ± 3.3% compared to Cartesian. In the pelvis of healthy adults, FB radial demonstrated fat quantification accuracy with absolute MD < 1.2 ± 3.2% in low fat fraction regions (< 5% PDFF) and absolute MD < 4.6 ± 5.6% in high fat fraction regions (> 80% PDFF). In the liver and abdomen, PDFF showed significant correlation (ρ > 0.986, ρc > 0.985), and absolute MD < 1.0 ± 10.6% between FB radial and BH SVS, and significant correlation (r > 0.996, ρc > 0.995), and absolute MD < 0.9 ± 5.7% between FB radial and BH Cartesian.

In children with NAFLD, FB radial demonstrated significantly less motion artifacts compared to BH Cartesian (P < 0.05). FB radial PDFF demonstrated a linear relationship (P < 0.001) versus BH SVS PDFF and BH Cartesian PDFF with r = 0.996 and ρc = 0.994, and r = 0.997 and ρc = 0.995, respectively. The absolute MD in PDFF between FB radial, BH Cartesian, and BH SVS was less than 0.7 ± 2.6% Repeated FB radial had MDwithin = 0.25% and CR = 1.53% for PDFF.

Ten infants were enrolled, and nine infants completed the study (90% completion). The preparation time and scan time were 32 ± 7 min and 24 ± 11 min, respectively. FB radial MRI demonstrated significantly higher image quality scores compared to Cartesian MRI (P = 0.01). The measurements using FB radial were VAT = 48.2 ± 16.5 cm3; VAT-PDFF = 41.6 ± 3.8%; SAT = 281.6 ± 104.5 cm3; SAT-PDFF = 86.1 ± 4.8%; BAT = 1.5 ± 0.6 cm3; and BAT-PDFF = 28.9 ± 8.8%. Hepatic PDFF measured using FB radial was 3.5 ± 1.1% and had repeatability of MDwithin < 0.25% and CR < 2.0%.

For R2* mapping, FB radial demonstrated accurate (ρc ≥ 0.996; P < 0.001; absolute MD < 0.2 ± 4 s-1) and repeatable (MDwithin = 0.23 s-1; CR = 3.9 s-1) quantification in a phantom, and repeatable (MDwithin < 0.5 s-1; CR ≤ 4.6 s-1) quantification in thirty subjects with normal pregnancies. At 3T, placental R2* mean ± standard deviation was 12.9 ± 2.7 s-1 for 14-18 and 13.2 ± 1.9 s-1 for 19-23 weeks GA. The CV was significantly greater (P = 0.043) at 14-18 (0.63 ± 0.12) than 19-23 (0.58 ± 0.13) weeks GA. At 19-23 weeks, the CV was significantly lower (P < 0.001) for anterior (0.49 ± 0.08) than posterior (0.67 ± 0.11) placentas. One IPD subject had a lower mean R2* than normal subjects at both GA ranges (Z ̂ < −2).

Conclusion

FB radial demonstrated quantitative accuracy compared to BH techniques in a fat fraction phantom and in the pelvis and liver of healthy volunteers. In healthy children and children with NAFLD, non-sedated free-breathing radial MRI provided accurate and repeatable hepatic PDFF measurements and improved image quality, compared to standard breath-held MR techniques. Finally, in a R2* phantom FB radial provided accurate and repeatable three-dimensional R2* mapping and in pregnant subjects FB radial provided repeatable R2* mapping in the entire placenta at 3T during early GA.

Magnetic resonance imaging (MRI) is a powerful imaging technique for visualizing soft tissues and characterizing tissue properties. It can be used for the assessment of meniscal, ligamentous, and cartilaginous lesions in the knee. When the extracellular matrix of articular cartilage is compromised, water moves more freely within the cartilage, which leads to prolonged T2 relaxation times. Consequently, quantitative T2 maps have been used to understand the pathophysiology of osteoarthritis or follow-up monitoring of knee cartilage after surgery. Standard T2 quantification techniques usually use spin echo-based sequences which require long acquisition times. Three-dimensional (3D) dual echo steady state (DESS) MRI has shown its potential for time-efficient T2 mapping in the knee. DESS acquires images with distinct contrasts (“FID” and “Echo”). This feature of DESS allows for both morphological T2-weighted (T2w) imaging and quantitative T2 mapping in a single scan. Our research group at UCLA has developed a protocol to acquire 3D DESS scans with isotropic high-resolution. However, the scan time still needs to be reduced to facilitate eventual translation. In order to reduce acquisition time in T2 mapping sequences, previous works developed methods to reconstruct images using undersampled data. Previous proposals such as compressed sensing (CS) reconstruction have been used to reconstruct images from undersampled data while mitigating undersampling artifacts. However, CS reconstruction methods are time-consuming and the results are sometimes overregularized. Deep learning (DL) based image enhancement or reconstruction methods provide a solution to the shortcomings of CS reconstruction by learning the mapping between undersampled input data and high-quality output images from large reference datasets, and providing rapid inference times. Since the isotropic high-resolution 3D DESS acquisition designed by our group at UCLA is relatively recent and has a limited number of datasets, strategies such as transfer learning, i.e., pretraining on a larger available reference dataset, may be needed to obtain a DL network that can produce high-quality output images from our undersampled high-resolution 3D DESS MRI data. Another main challenge of DL-based image enhancement or reconstruction methods is that it is difficult to understand why/how they work or to predict its performance. In medical image enhancement, image fidelity must be prioritized. Obstruction or elimination of image details may confound diagnostic decisions. To overcome this problem, recently there are some works on developing DL networks with uncertainty estimation to predict error in enhancement tasks. Thus, the aim of this work is to shorten the scan time of 3D DESS MRI by reconstructing high quality images from undersampled data using DL networks that incorporate uncertainty estimation and transfer learning.

Metabolic dysfunction-associated steatotic liver disease (MASLD), formerly known as non-alcoholic fatty liver disease (NAFLD), is the most common chronic liver disease with a current global prevalence of 25% to 40%. MASLD is associated with the metabolic syndrome and cardiovascular morbidity, and can progress to fibrosis and cirrhosis. Chronic liver diseases such as viral hepatitis and MASLD can also lead to hepatic iron overload. Magnetic resonance imaging (MRI) provides non-invasive evaluation of hepatic steatosis and iron overload by quantifying proton-density fat fraction (PDFF) and R2*. Conventional MRI techniques for liver PDFF and R2* quantification require breath-holding, which can be challenging for children and elderly patients. 3D stack-of-radial MRI techniques have been proposed for self-gated free-breathing liver PDFF and R2* quantification. However, several challenges remain, including residual streaking artifacts from system imperfections, long scan acquisition times, computationally expensive reconstructions, and insufficient modelling of non-rigid liver motion during free-breathing. Techniques to overcome these challenges are important for a wide clinical adoption of free-breathing MRI techniques for liver PDFF and R2* quantification. Additionally, in recent years, there has been an increased interest in lower-field MRI systems. A less expensive lower-field MRI system with a larger bore diameter may improve accessibility and comfort for populations with obesity and at risk for fatty liver diseases. However, the low signal-to-noise ratio problem can impact image quality and quantification accuracy. Therefore, noise reduction techniques are important to improve liver PDFF and R2* quantification in lower-field MRI systems.

This work focuses on developing MRI reconstruction techniques to improve liver PDFF and R2* quantification. First, this work developed a phase-preserving beamforming-based technique to effectively reduce radial streaking artifacts from system imperfections. This technique can be further integrated with motion-resolved reconstruction to improve self-gated free-breathing liver PDFF and R2* quantification. Second, this work developed an uncertainty-aware physics-driven deep learning network for rapid reconstruction of PDFF and R2* maps from self-gated free-breathing MRI. The uncertainty maps generated from the network can be used to predict quantification errors and improve reliability of deep learning reconstruction results. Third, this work developed a compressed sensing reconstruction model with non-rigid motion compensation to improve and accelerate self-gated free-breathing liver PDFF and R2* quantification. Last, this work developed and evaluated image and k-space denoising techniques that can improve quantification accuracy and precision of Cartesian-based liver PDFF and R2* quantification at 0.55T. These technical advancements can provide accurate and motion-robust liver fat and R2* quantification.

Prostate Cancer (PCa) remains the second most common cause of cancer-related death in men in the U.S. Multi-parametric (mp) MRI is playing an increasingly important role for the localization, detection, and risk stratification of PCa. However, prostate mp-MRI still misses PCa in up to 45% of men and faces challenges in distinguishing clinically significant PCa from indolent PCa. Therefore, MRI technology must be improved to enhance diagnostic performance for PCa. This thesis aimed to improve prostate MRI by addressing two challenges. First, the diffusion-weighted imaging (DWI) component of mp-MRI often suffers from artifacts such as distortion and low signal-to-noise ratio (SNR), which can lead to low diagnostic image quality. Second, prostate microstructure features are key determinants for histopathological assessment of cancer aggressiveness; however, current MRI techniques have limitations in capturing this information. To address the first challenge, in Aim 1, we translated and evaluated an eddy current-nulled convex optimized diffusion encoding (ENCODE) based prostate DWI technique that achieves short echo time (TE) to maintain SNR while reducing prostate geometric distortion from eddy currents and susceptibility effects. Further, in Aim 2, we developed a combined TE-minimized ENCODE diffusion encoding acquisition with a random matrix theory-based denoising reconstruction technique to improve the SNR and robustness of high-resolution (in-plane: 1.0x1.0 mm2) prostate DWI and apparent diffusion coefficient mapping. To address the second challenge, in Aim 3, we performed a first proof-of-concept ex vivo evaluation and validation of the diffusion-relaxation correlation spectrum imaging (DR-CSI) technique at 3T for quantifying microscopic tissue compartments (epithelium, stroma, and lumen) in PCa using whole-mount digital histopathology as the reference standard. Further, in Aim 4, we explored and evaluated sequential backward selection analysis for the acceleration of DR-CSI through subsampling of the diffusion-relaxation contrast encoding space while maintaining the accuracy of prostate microstructure mapping in PCa.

This dissertation aims to evaluate therapies that have the potential to improve the recovery of the left ventricle (LV) following myocardial infarction (MI) burdened with intramyocardial hemorrhage (IMH). Prior studies have shown that hemorrhagic MI results in the deposition of chronic iron in the myocardium, which induces cytotoxic effects as well as impacts the normal function of the microvasculature. In order to characterize the content of intramyocardial iron, T2* magnetic resonance imaging (MRI) was used as the gold standard in non-invasive detection of cardiac iron, and therefore acts as a diagnostic guidance tool informing the efficacy of the therapies. The dissertation may be broken down into three phases, all performed with MRI guidance to elucidate: 1) mechanistic understanding of the microvascular environment in chronic MI subjects; 2) effects of iron chelation therapy via Deferiprone (DFP) administration; and 3) effects of therapeutic hypothermia (TH) induced post reperfusion. The first phase investigated the long-term changes in myocardial perfusion assessed via MRI, in patients with reperfused myocardial infarction and an animal model of ischemia reperfusion (I/R) injury. From animal studies, histology, immunohistochemistry (IHC), and western blotting analysis were performed to elucidate the mechanistic underpinnings of MRI observations. The outcomes of this study led to the discovery that hemorrhagic MI results in reduced myocardial perfusion within hemorrhagic, but not non-hemorrhagic, MI territories. Further, the protein expression investigations enabled the proposal of a mechanistic pathway to examine the role chronic iron deposition plays in the perfusion defects observed in hemorrhagic MI. The subsequent study in a canine model of hemorrhagic MI to remove iron from within chronic infarction regions using the small-molecule iron chelator deferiprone (DFP), is the first to show the potential to abrogate resting perfusion defects observed in the hemorrhagic MI setting. Furthermore, the study showed that the recovery of rest perfusion did not persist following termination of the DFP therapy, indicating the potential need for continuous, or extended, iron chelation in this population to maintain persistent benefits. Lastly, the study showed the potential beneficial impact of DFP therapy on LV remodeling by resulting in reduced end-diastolic mass, suggesting a possible role of iron in LV hypertrophy and diastolic dysfunction. Finally, a study of post-reperfusion localized therapeutic hypothermia was conducted in a pig model to evaluate the potential impact of hypothermia therapy on IMH and chronic iron deposition. This is the first study to show the capability of therapeutic hypothermia to reduce chronic iron deposition in hemorrhagic MI. The results of this study showed that post-reperfusion hypothermia did not impact acute infarct size and did not affect hemorrhage in the acute phase (day 3 post-MI). However, by 1 month hypothermia-treated animals showed significantly reduced T2*-derived iron deposition volume, which held when normalized by infarct size. By 2 months post-MI, absolute T2* values were also indicative of decreased myocardial iron content, with significantly increased T2* values (lower iron content) in hypothermia-treated animals. Furthermore, LV ejection fraction (LVEF) was significantly elevated at 2 months in the hypothermia group, suggesting a positive effect of therapeutic hypothermia on chronic LV function. In summary, this dissertation used animal models of hemorrhagic MI to investigate two promising therapeutic methods for alleviating the adverse remodeling in hemorrhagic MI subjects, showing promising results that will aid the future development of adjunctive clinical therapies for advancing treatment in MI and ischemia reperfusion injury.

The current gold-standard approach for detection and quantification of intramyocardial hemorrhage (IMH) is T2* cardiovascular magnetic resonance imaging (CMR). T2*-based imaging techniques have been demonstrated to have high sensitivity for detecting hemorrhage and residual iron. The conventional T2*-based imaging employed for IMH imaging is based on a 2D breath-held, ECG-triggered, segmented, multi-gradient-echo sequence. More recently, a dark-blood cardiac T2* MRI technique has emerged for imaging of global iron overload such as thalassemia. It has been interchangeably used with bright-blood T2* MRI for imaging of local iron overload such as intramyocardial hemorrhage. To date however, dark-blood T2* techniques for intramyocardial hemorrhage characterization has not been validated. In Chapters 2 and 3, we investigated the diagnostic capacity of dark-blood T2* MRI against bright-blood T2* MRI for intramyocardial hemorrhage characterization in both clinical and preclinical settings. We found that double-inversion-recovery prepared dark-blood T2* images provide lower signal-to-noise ratio and lower contrast-to-noise ratio between hemorrhage and remote myocardium, consequently underestimating the hemorrhage extent. Dark-blood T2* MRI also demonstrated weaker sensitivity, specificity, accuracy, and inter-observer variability compared to bright-blood T2*-weighted MRI. Our studies also showed that the loss in SNR and CNR in dark-blood T2* imaging emerges from the signal loss following double-inversion-recovery preparation and insufficient recovery time between double-inversion-recovery preparation and readout. Hence, we conclude that dark-blood T2* MRI does not have the same diagnostic capacity for assessment of intramyocardial hemorrhage and bright-blood T2* MRI should be the preferred choice for clinical use. Studies have shown that fat infiltration is a common phenomenon in chronic myocardial infarction. However, signal from fat protons can confound the T2* assessment of intramyocardial hemorrhage. To address this issue, in Chapter 4, we studied the influence of fat infiltration on iron quantification in T2* mapping using a widely accepted water-fat separation algorithm. Specifically, we evaluated the temporal dependence of fat infiltration in hemorrhagic myocardial infarctions. We found that fat infiltration was observed in early and late chronic phases of myocardial infarctions, which if not corrected for, can underestimate the extent of iron content within the infarct zone. Notably, we also found that the amount of fat infiltration in chronic phase of MI was closely correlated with the amount of iron. Another major confounder in conventional 2D breath-held ECG-gated T2* imaging is motion artifacts. In clinical settings, patients with acute myocardial infarctions often find it difficult to hold their breath during cardiac MRI exams. Some patients may even suffer from arrhythmia (irregular heartbeat). Both situations can lead to unsuccessful gating during data acquisition leading to motion artifacts on T2* images especially with long echo times. To address this issue, in Chapter 5, we developed a motion-resolved fully ungated free-breathing 3D cardiac T2* imaging technique using a low-rank tensor framework to accommodate clinical needs and to mitigate motion artifacts due to unsuccessful breath-holds or ECG gating. We tested our 3D LRT technique in healthy volunteers and animal models for image quality, SNR and T2*. We found that the proposed 3D LRT technique can provide superior image quality compared to conventional T2* techniques at the same level of signal-to-noise ratio. T2* measured from proposed 3D LRT data showed excellent agreement with T2* from conventional 2D approach. We also found that a key benefit of 3D acquisition is that it permits the reconstruction of high-resolution T2* images using the proposed 3D LRT T2* approach. High-resolution T2* images from proposed 3D LRT approach showed superior image quality and diagnostic capacity for assessment of intramyocardial hemorrhage. In Chapter 6, the proposed 3D LRT T2* imaging approach was validated on an animal model for feasibility and capability for characterization of intramyocardial hemorrhage. We found that our 3D LRT approach had excellent image quality and diagnostic accuracy in the assessment of intramyocardial hemorrhage compared to the 2D breath-held and gated acquisitions. Broadly, this dissertation identified and corrected a number of critical confounders affecting the accuracy of T2* MRI in assessment of intramyocardial hemorrhage. By identifying and solving these confounders in T2* imaging, we aim to improve the diagnostic capability of MRI for prognosis and therapeutic care of patients with hemorrhagic myocardial infarctions. In future work, feasibility of the newly developed fully ungated free-breathing 3D LRT T2* imaging technique will be investigated on patients for imaging of intramyocardial hemorrhage. And the potential of high-resolution T2* imaging which greatly improved intravoxel dephasing due to off-resonance will be explored.

Obesity affects around 2.1 billion people all over the world and is associated with type 2 diabetes, cardiovascular diseases, and metabolic dysfunction-associated steatotic liver disease (MASLD), formerly known as non-alcoholic fatty-liver disease (NAFLD). MASLD is defined as abnormal fat accumulation in the liver and can range from simple steatosis to steatohepatitis and fibrosis. Magnetic resonance imaging (MRI) is a non-invasive imaging modality that can quantify subcutaneous (SAT) and visceral adipose tissue (VAT). Increased SAT and VAT are highly associated with cardiometabolic diseases, and particularly MASLD. Conventional techniques to quantify SAT and VAT on MRI require manual annotations, which are time-consuming, not suitable for large-scale studies, and may suffer from intra/inter-rater variability. Although several deep-learning methods were proposed to automatically segment SAT/VAT, serious challenges remain including inability to fully capture detailed and disconnected VAT features, no studies performed in longitudinal MRI, not focusing on children with wide age ranges, or not using a more child-appropriate acquisition scheme such as free-breathing (FB) abdominal MRI. The first part of this thesis focuses on advanced deep-learning techniques to accurately and automatically segment SAT and VAT in longitudinal MRI in adults with obesity. Then SAT and VAT are also automatically segmented in children using the images from FB-MRI based on a motion-robust stack-of-radial sampling trajectory. These technical developments could enable detection and monitoring of risk for MASLD before the disease progresses into more severe forms. Additionally, MR elastography (MRE) can quantify liver stiffness, which is a biomarker to detect a more severe effect of MASLD: liver fibrosis. However conventional MRE requires breath-holding to mitigate motion artifacts, which could be particularly challenging in people with decreased lung capacity, the elderly, and children. In the second part of this thesis, we developed and evaluated a FB radial MRE tool that employed rapid encoding to reduce the scan time, fractional encoding to have sufficient SNR in the liver, and self-gating to compensate for breathing motion. The proposed radial FB-MRE could accurately quantify liver stiffness with acceptable repeatability when compared to conventional breath-holding (BH) MRE. Radial FB-MRE, which generates liver stiffness in excellent agreement with the conventional BH-MRE, is a promising tool that quantifies liver stiffness and can help sensitive populations such as children, where BH is challenging or not possible/feasible.

Cardiovascular disease is the leading cause of death in patients with Duchenne muscular dystrophy (DMD) – a fatal X-linked genetic disorder characterized by progressive muscle weakness and pediatric onset cardiomyopathy. The genetic mutations associated with DMD are variable since the very large dystrophin gene is comprised of 79 exons encoded within 2.4 million nucleotides, resulting in a wide range of phenotypes observed in affected patients. Symptom recognition is difficult in non-ambulatory patients, therefore clinical evidence of cardiac dysfunction is frequently limited to imaging findings when severe or end-stage cardiomyopathic change has occurred. Conventional late gadolinium enhancement (LGE) imaging is the clinical current gold standard for detecting myocardial tissue remodeling (i.e., replacement fibrosis), but it is often a late finding (mean onset observed at 15.2 � 5.1 years) in the DMD disease process. Reduced left ventricular ejection fraction (LVEF), a marker of cardiac dysfunction, is also a late outcome in DMD. Sensitive imaging methods are needed to identify early cardiac involvement in this high-risk population. Importantly, a biomarker capable of detecting early myocardial remodeling prior to LGE, a decline in EF, and without the need for an exogenous contrast agent could significantly improve the care of boys with DMD. A brief introduction to DMD and its natural progression to the heart is provided in Chapter 1. In Chapter 2, an introduction to Nuclear Magnetic Resonance (NMR) and MRI fundamentals is offered. This leads to the description of cardiovascular MRI and common techniques that provide a quantitative assessment of cardiac function and microstructure in Chapter 3. In Chapter 4, we present a descriptive overview of current and emerging biomarkers of microstructure in DMD. In Chapter 5 and Chapter 6, the use T1 mapping MRI, an alternative to conventional LGE imaging, is extended for the quantification of myocardial fibrosis in the left and right heart of boys with DMD. In Chapter 5 specifically, we use 3T MRI in the left heart for the following: (1) characterize global and regional myocardial pre-contrast T1 differences between healthy controls and LGE + and LGE− boys with DMD; and (2) to report global and regional myocardial post- contrast T1 values and myocardial ECV estimates in boys with DMD; and (3) to identify left ventricular (LV) T1-mapping biomarkers capable of distinguishing between healthy controls and boys with DMD and detecting LGE status in DMD. In Chapter 6, using the same technique, we evaluate several analysis methods and identify the most reliable one to measure RV pre- and post-contrast T1 (RV-T1) and to characterize myocardial remodeling in the RV of boys with DMD. In Chapter 7 we explore the onset of intramyocardial and excessive pericardial fat in boys with DMD. Specifically, we present the use of a chemical-shift-based multi-echo MRI (CS-MRI) technique, another alternative to LGE imaging, for the detection of myocardial remodeling in boys with DMD. We report and compare the LV intramyocardial and pericardial fat content in boys with DMD and healthy controls and we aim to determine if fatty infiltration precedes the appearance of LGE in boys with DMD. In Chapter 8, we evaluate T1 mapping biomarkers for repeatability and prognosis. We first define the intra-exam repeatability of T1 mapping biomarkers and calculate the smallest detectable change. Regional myocardial T1 differences between the septal and lateral mid- ventricular LV wall suggest myocardial heterogeneity in the DMD heart which that may be indicative of progressive myocardial remodeling. Thus, we characterize regional pre- and post- contrast T1 and ECV differences at the basal, mid-ventricular, and apical wall in boys with DMD and healthy controls. In Chapter 9, we provide concluding thoughts on quantitative MRI biomarkers of myocardial remodeling in DMD. Additionally, we share a future perspective for the work presented in this dissertation, and its implications on the clinical care for boys with DMD. The work described in this dissertation provides a framework for evaluating CMR biomarkers of tissue-composition in boys with DMD. This work will enable us to detect early myocardial remodeling and to quantify an individual patient’s degree of disease severity and/or progression, which in turn, establishes a step towards providing patient-specific diagnostic and prognostic information and informing treatment for cardiomyopathy.