MSBI Thesis & Abstracts 2023-2024
Radhika Bhalerao
Thesis Title: Predicting Meningioma Genetics from multi-sequence Magnetic Resonance Imaging using Machine Learning
Advisor: Andreas Rauschecker, MD, PhD
Abstract: Meningiomas, the most prevalent primary central nervous system tumors, present a significant challenge in neuro-oncology due to their variable clinical behaviors and recurrence rates (1). While magnetic resonance imaging (MRI) remains the primary diagnostic tool, recent advancements in our understanding of meningioma genetics have highlighted the critical role of molecular profiles in determining tumor behavior and treatment outcomes (2). This thesis presents a comprehensive exploration of the intersection between imaging features, genetic biomarkers, and artificial intelligence in meningioma management, with the overarching goal of enhancing diagnostic accuracy, treatment planning, and prognostication.The work is structured in four interconnected chapters, each addressing a crucial aspect of this multifaceted challenge: Chapter 1 introduces a novel, large-scale dataset comprising 3,101 pre-processed, multi- sequence MR images along with corresponding genetic and demographic data from patients with histopathological confirmed intracranial meningiomas. This dataset serves as the foundation for subsequent analyses and model development, offering researchers an unprecedented resource to investigate imaging-genetic correlations in meningiomas. Building upon this dataset, Chapter 2 presents the development of a machine learning model designed to predict genetic mutation status in meningiomas using preoperative multi-sequence MRI. By combining radiomics features, convolutional neural network (CNN) outputs, and clinically informed features, this approach demonstrates the potential for non-invasive assessment of genetic biomarkers, which could significantly impact clinical decision-making, especially in settings where extensive genetic testing is not readily available. Chapter 3 addresses a fundamental challenge in medical imaging AI: accurate identification of MRI sequences. Recognizing the limitations of existing methods, this chapter proposes an innovative approach using large language models (LLMs) to parse MRI metadata for sequence identification. This method improved robustness to human errors in metadata entry and better generalization across institutions, potentially streamlining the preparation of large, multi-center datasets for AI model training. Finally, Chapter 4 provides a comprehensive discussion of the findings, their implications for clinical practice and research, and future directions for advancing the field of meningioma management through integrated imaging and genetic approaches. Throughout this thesis, we demonstrate the potential of combining advanced imaging techniques, genetic profiling, and artificial intelligence to enhance our understanding and management of meningiomas. By bridging the gap between radiological features and underlying genetic alterations, we aim to pave the way for more personalized and effective treatment strategies, ultimately improving outcomes for patients with these complex tumors.
Duc Huy Doan
Thesis Title: Application of Convolutional Neural Networks in Multiparametric MR Imaging to Predict Prostate Cancer Progression
Advisor: Susan Noworolski, PhD
Abstract: Prostate cancer progression after radical prostatectomy poses a significant risk to patient health. The ability to predict which patients are at a higher risk of progression is crucial for determining appropriate adjuvant therapies. This study investigates the application of convolutional neural networks (CNNs) to pre-surgical multiparametric MRI (mpMRI) for predicting post-surgical prostate cancer progression. The study utilizes a retrospective patient cohort and explores the performance of different CNN architectures (ResNet and DenseNet), normalization methods, and slice selection techniques. The results demonstrate the potential of CNNs in predicting prostate cancer progression, with the best-performing model achieving an accuracy of 0.712. The study highlights the importance of appropriate image normalization and slice selection methods for optimal performance. The findings suggest that CNNs could serve as a valuable tool for aiding clinical decision-making in prostate cancer management.
Tangran Dong
Thesis Title: Optimizing Diffusion Weighted Imaging for Breast Cancer Treatment Evaluation: A Study on Correcting Echo Planar Imaging Distortions and Gradient Nonlinearity
Advisor: Nola Hylton, PhD
Abstract: Diffusion Weighted Imaging (DWI) offers promising enhancements to breast cancer imaging by providing detailed examinations of tissue microstructure without the need for contrast agents, potentially improving specificity compared to the current standard, Dynamic Contrast-Enhanced MRI (DCE-MRI). Despite its advantages, DWI faces challenges such as Echo Planar Imaging (EPI)-related distortions, gradient nonlinearity, and eddy currents, which compromise its diagnostic sensitivity. This thesis tackles these issues by implementing reverse polarity gradient (RPG) sequences to correct EPI distortions and applying Gradient Nonlinearity Correction (GNC) to improve the accuracy of quantitative tissue measurements calculated from DWI.Utilizing data from the I-SPY2 TRIAL, a multi-center clinical trial testing novel agents for breast cancer, this study evaluates the effectiveness of these correction techniques on a subset of 6 patients and 16 exams scanned with Diffusion Tensor Imaging (DTI) and RPG sequences. The application of RPG and GNC has shown significant improvements in the Apparent Diffusion Coefficient (ADC) and Fractional Anisotropy(FA) measurements in phantom studies, affirming the potential of these corrections to standardize clinical DWI protocols and enhance its reliability alongside DCE-MRI in breast cancer treatment evaluation. Furthermore, this thesis explores the lack of correlation between changes in ADC values and Functional Tumor Volume (FTV) measured from DCE-MRI, underscoring DWI's potential role in providing independent and complementary information for assessing biological changes and aiding treatment evaluation.
Chase Fitch
Thesis Title: Dorsal Root Ganglia Diffusion Metrics in Patients with Lumbar Radiculopathy Undergoing Injection
Advisor: Cynthia Chin, MD
Abstract: We compared injected, assumed to be symptomatic, and non-injected, assumed to be asymptomatic, dorsal root ganglia (DRG) in patients receiving lumbar facet and/or nerve injections using DTI MRI data and foraminal stenosis (FS) and canal stenosis (CS) grades. Materials and Methods: Healthy volunteers (HVs) and patients receiving lumbar facet and/or nerve block injections for pain underwent lumbar MRIs including axial T2-weighted fat-water separated FLEX 3D FSE and axial DTI. Patients were imaged up to a month prior to injection and up to six months after injection. Processing included: DRG segmentation (MD.ai), 3D volume (MorphACE), DTI (spherical harmonic and Constant Solid Angle). ADC, FA and volume were compared between HVs, asymptomatic and symptomatic DRG and correlated with stenosis grades using paired t-tests. Results: 25 patients and 5 HVs DTI scans were analyzed (34 patients and 10 HVs for volume). There was a sequential increase in DRG volume from cranial-caudal L1-S1 in the HVs (p<0.001). Symptomatic DRG had higher FA than asymptomatic DRG before injection (p<0.01) and symptomatic DRG FA decreased after injection (p<0.05) while asymptomatic DRG FA slightly increased. Severe CS was associated with lower ADC than no or mild CS (p<0.001). Conclusion: Cranial-caudal sequential increase in DRG volume from L1-S1, consistent with cadaver data, may reflect degree of cutaneous and muscle area/volume innervation. Symptomatic DRG have higher FA than asymptomatic DRG that return to asymptomatic levels after injection. This could be due to phospholipase A2 (PLA2) inhibition by corticoid-steroid suppressing nerve cell growth and thus organized diffusion alternatively receptor inhibition. Severe CS could cause arterial flow constriction leading to ischemic DRG and reduced ADC.
Ellis Mayne
Thesis Title: Evaluating the Efficacy of Combination CD46-Targeted 225Ac Radioimmunotherapy and Antibody-Drug Conjugate Therapy in Multiple Myeloma
Advisor: Robert Flavell, MD, PhD
Abstract: Multiple myeloma is one of the most common blood cancers. Many drugs have been developed to treat multiple myeloma, yet refractory and relapsed disease remain prevalent. The recent identification of CD46, an antigen overexpressed on multiple myeloma cells, has led to new development of antibody-based immunotherapies, including the antibody-drug conjugate CD46 ADC and the radioimmunotherapy [225Ac]Ac-Macropa-PEG4-YS5. The structures of both drugs include the CD46-binding monoclonal antibody YS5, and both have shown high tumor binding and antitumor efficacy in pre-clinical models for multiple myeloma. A Phase 1 clinical trial is also underway to evaluate CD46 ADC in multiple myeloma. While initial results are promising, both drugs have noted drawbacks, with blood cancers like multiple myeloma known to develop resistance to ADCs over time and [225Ac]Ac-Macropa-PEG4-YS5 studies displaying nephrotoxicity at higher doses. Hypothesizing that the use of the two drugs in tandem could improve therapeutic efficacy, this study investigated the performance of RIT & ADC combined treatment across preclinical multiple myeloma models in-vitro and in-vivo. Results of initial cell-based assays demonstrated a high cell killing ability of combination treatment, with evidence of a synergistic interaction between drugs at select combination concentrations. A pilot xenograft mouse model comparing combination treatment to RIT & ADC monotherapies showed a widespread reduction in tumor burden with animal body weight remaining stable, potentially indicating reduced off-target toxicity. Future work will monitor the antitumor efficacy of RIT & ADC combination therapy over a longer time scale and measure organ damage ex-vivo to further compare against prior monotherapy results. This encouraging initial data provides guidance on combination dosing regimens for future preclinical research and suggests that RIT & ADC combination therapy may be a valuable clinical option for treating multiple myeloma.
Siddharthasiva Anbu Rajan
Thesis Title: MRI - Based Measures of Metabolic Health in the Assessment of Patients with Chronic Inflammatory States
Advisor: Susan Noworolski, PhD
Abstract: Disease conditions like obstructive sleep apnea (OSA) and human immunodeficiency virus (HIV) infections are characterized by chronic low-grade inflammation, leading to poor metabolic health. This work is focused on comparing MR-based fat measures to indicators of worse metabolic health, namely OSA severity and hydroxyproline, a biomarker of subcutaneous fat (SAT) fibrosis. The study also explored the utility of novel MRI methods of diffusion-weighted imaging of fat and T1 mapping to detect fibrosis in the SAT. There were 33 participants with OSA and 58 participants with or without HIV infection who had hydroxyproline measured via SAT biopsy, 13 of whom had the novel MRI measures of fibrosis. The liver, visceral, and SAT volumes were segmented using artificial intelligence-based methods, and the pancreas was manually drawn on proton density fat fraction (PDFF) images. Apparent diffusion coefficient (ADC) and T1 were measured in regions of interest drawn in the SAT. Liver fat fraction, liver fat content, and pancreatic fat fraction were higher with increased severities of sleep apnea, p£0.01. Furthermore, liver fat fraction, pancreatic fat fraction, and visceral fat volume were higher in subjects with high hydroxyproline levels on biopsy, p£0.03. The ADC of SAT at b-values of 3×10-3 s/mm2 were negatively correlated with hydroxyproline levels. It can be concluded that more severe OSA and higher hydroxyproline were associated with worse metabolic health. The role of diffusion-weighted imaging of fat with high b-values to detect SAT fibrosis is also encouraging and opens up avenues for future research.
Nadezhda Urmanov
Thesis Title: Improved chemoenzymatic radiosynthesis of fluorine-18 labeled sakebiose for microPET-CT imaging of Staphylococcus aureus
Advisor: David Wilson, MD, PhD
Abstract: Staphylococcus aureus (S. aureus) is a gram-positive bacterium that can cause severe infections such as pneumonia, osteomyelitis, and endocarditis when it breaches the skin. This study aimed to enhance the chemoenzymatic radiosynthesis of 2-deoxy-2-[18F]-fluoro-sakebiose ([18F]FSK), a radiotracer potentially useful for imaging S. aureus infections. By optimizing the synthesis of 2-deoxy-2-[19F]-fluoro-sakebiose ([19F]FSK), we identified key factors—such as increased enzyme concentration and decreased precursor levels—that significantly improved the yield. Applying these optimized conditions to the synthesis of [18F]FSK resulted in a 30% increase in the radiochemical yield (RCY%) from the control experiment. In vitro evaluation showed that [18F]FSK was successfully incorporated into two strains of S. aureus, suggesting its potential utility for imaging bacterial infections in vivo. This work lays the groundwork for using [18F]FSK in PET/CT imaging to diagnose and monitor S. aureus infections.
Jo Veres
Thesis Title: The Impact of Proteoglycan Degradation and Fragmentation on T1rho Relaxation Times
Advisor: Aaron Fields, PhD
Abstract: 266 million individuals are diagnosed with intervertebral disc degeneration annually. Most of these individuals experience pain and reduced mobility, but some are asymptomatic. In contrast to age-related physiologic disc degeneration, pathologic disc degeneration is hypothesized to entail changes to the proteoglycan matrix within the disc. T1rho-weighted MR imaging has the ability to detect subtle changes in disc biochemistry, and may therefore discriminate between pathologic vs. physiologic disc degeneration. However, the contribution of individual biochemical changes to T1rho signals is unclear. Therefore, the goal of this study is to elucidate the individual roles of proteoglycan concentration and molecular weight on T1rho relaxation time. To do this, MRI phantoms with prescribed differences in the concentration and molecular weight of proteoglycan-mimicking dextrans were fabricated and imaged with T1rho-weighted MRI. Results showed that changing the dextran concentration changed T1rho relaxation times by up to 8 percent, while varying dextran size changed T1rho relaxation times by up to 47 percent. Proteoglycan fragmentation is an early marker of disc degeneration that precedes proteoglycan loss from the disc. Prior studies reported that T1rho relaxation times are correlated with disc proteoglycan content. Thus, our findings are important because they are the first to show that T1rho relaxation times are also sensitive to polysaccharide fragment size, and therefore T1rho relaxation times in the disc may reflect proteoglycan degradation in the very early stages of disc degeneration.
Tianrun Xiao
Thesis Title: Optimization of the MR-guided Focused Ultrasound Induced Blood-Brain Barrier Opening in a rat model
Advisor: Eugene Ozhinsky, PhD
Abstract: IntroductionThe blood-brain barrier (BBB) restricts of therapeutic agents delivery for brain conditions. Low-intensity pulsed focused ultrasound (FUS) with microbubble injection offers a non-invasive method for localized, reversible BBB disruption. This study explores the efficacy and safety of various FUS parameters for BBB opening in a rat model. Methods Experiments were performed on ten rats using a stereotactic preclinical focused ultrasound system. Microbubble clearance was assessed in four brain areas for microbubble delivery time. Acoustic pressures ranging from 0.3 MPa to 0.65 MPa with burst lengths of 5 ms and 10 ms were used to perform BBB opening. MRI scans were conducted pre- and post-treatment to assess tissue damage and BBB integrity. The signal enhancement rate was calculated to evaluate BBB opening. The FUS system monitored the treatment using a hydrophone, and the recorded spectra were analyzed to measure subharmonic and ultra-harmonic peaks. Results An acoustic pressure of 0.4 MPa with a 10 ms burst length resulted in BBB opening without tissue damage. MRI signal hyperintensity and sonication feedback indicated safe BBB opening at pressures of 0.4 MPa to 0.5 MPa. Both 5 ms and 10 ms burst lengths are effective. The microbubble clearance experiment revealed that microbubble concentrations were sufficient for BBB opening up to 8 minutes post-injection. Conclusion An optimal acoustic pressure of 0.4 MPa and 10 ms burst length is safe and effective for BBB opening in rats. Higher acoustic pressures were less effective and potentially led to brain edema.