MSBI Thesis & Abstracts 2018-2019
Ravi Chachad
Advisor: Dr. Susan Noworolski, PhD
Thesis Title: Temporal Impact of Sugar Metabolism on the Liver
Abstract:
Purpose: Increased sugar consumption is associated with metabolic conditions that can result in poor health outcomes. To investigate the impact of sugars such as glucose and fructose in the body, this study aims to determine the timing of effects and assess the impact of these effects. This study aims to demonstrate acute effects in metabolism as a result of sugar metabolism.
Methods: Six male participants were imaged on a 3T MRI scanner at a fasted state then subsequently every hour for up to eight measurements. Between scanning sessions, they consume a 13C labeled glucose or fructose shake, have breath collected, and have blood drawn; this is repeated on a separate day to satisfy the other experimental condition. The MRI exam consists of Proton Density Fat Fraction (PDFF), proton Magnetic Resonance Spectroscopy (1H MRS), and 13C MRS. The images are processed to analyze liver volume and the spectra from the MR Spectroscopy are normalized and the peaks are quantified.
Results: Liver volume is significantly different from baseline measurements at 2-, 3-, and 4-hours post-feeding with p=0.032, p=0.003, and p=0.009 respectively. Fat content is significantly different from baseline measurements at 3- and 4-hours post-feeding with p=0.026 and p=0.048 respectively. Different MR measures of fat fraction in the body produce a significant positive correlation (p=0.003). The median change of lipids (CH2) positively correlates with the median change in glycerol (p=0.011). Fat fraction does not significantly correlate with the blood measures taken, but when high choline and low choline groups are separated, new formed lipids in the blood and long-term storage of fat differ significantly, p=0.021 and p=0.03 respectively.
Conclusions: Choline classification of participants resulted in a difference in new lipids in the blood and long-term storage in the liver. Low choline individuals tended to export less lipids in the blood and store more in the liver. High choline individuals exported more lipids and retained less in the liver. MR exams of the liver evaluate the health of the liver and burden to abdominal organs, whereas blood collection provided a glimpse into cardiovascular burden.
Qing Dai
Advisor: Dr. Peder Larson, PhD
Thesis Title: Clear Cell Renal Cell Carcinoma: Deep Learning-Based Prediction of Tumor Grade from Contrast-Enhanced CT
Abstract:
Tumor grading is an important prognostic parameter for renal cell carcinoma (RCC). However, current grading schemes require an invasive surgical procedure, putting patients at risks including increased risk of hemorrhage, infection, renal failure, or death. Furthermore, low grade RCC is indolent with low mortality risk and may not require treatment. Therefore, a pre-operative and non-invasive assessment of malignancy grade may be beneficial and facilitate optimal timing of treatment. In recent years, deep learning-based image analysis has gained wide popularity in cancer prognosis and prediction. The goal of this study is to investigate the feasibility and performance of a deep-learning-based model for clear cell RCC grading prediction from contrast-enhance computed tomography (CECT). After institutional review board approval, an institutional pathology database was queried for all renal biopsies between December 2002 and October 2018. All included patients received a CT with a non-contrast and at least one post-contrast series. All patients have Fuhrman grade confirmation from surgical pathology, with CT scan prior to the procedure. Tumors were manually annotated by a radiologist on either the corticomedullary or nephrographic phase CECT. Rectangular regions of interest (ROI) were drawn on each slice throughout the tumor and used as inputs to the Deep CNN ResNet50. A binary label of low grade or high grade was assigned to each patient. Sensitivity, specificity, accuracy, and AUC were calculated based on a five-fold cross-validation. Preliminary results from a small subset of datasuggests that a deep learning model can be used to predict clear cell RCC grading based on CT imaging prior to surgical procedures.
Emilie Decavel-Bueff
Advisor: Dr. Steven Hetts, MD
Thesis Title: Drug capture efficacy using polystyrenesulfonate-coated chemofilter device
Abstract:
Endovascular chemotherapy is an effective treatment option for cancer, however, the therapeutic agents used in this procedure often travel to non-target tissues and cause severe toxicity. Side-effects of chemotherapy range from nausea to life-threatening conditions. A strategy to reduce exposure of healthy tissues and organs to the toxicity of chemotherapeutic agents, such as doxorubicin (DOX), is to remove these drugs from systemic circulation after they have passed through the tumor site. With this goal in mind, different types of ChemoFilter devices have shown promise in alleviating these detrimental side effects. When placed downstream from the targeted tumor during intra-arterial chemotherapy, excess therapeutic agents bind to the device, preventing them from entering systemic circulation. In this study, we evaluated the doxorubicin-binding efficacy of a 3D printed porous cylindrical ChemoFilter device coated with sulfonated pentablock copolymers. Closed-circuit flow models experiments integrating 11 devices (uncoated or coated) at two concentrations of DOX (0.01 mg/mL and 0.05 mg/mL) were conducted. Samples collected from these flow models were used to treat H9c2 cell cultures, a rat embryonic cardiac cell line selected due to DOX cardiotoxicity. After a 24-hour treatment period, cell viability was calculated using the Trypan blue exclusion method. At 0.01 mg/mL DOX and 0.05 mg/mL DOX, the 3D printed polystyrenesulfonate-coated absorbers effectively filtered and eliminated DOX toxicity, increasing the H9c2 cell viability by 12.97% and 23.11%, respectively. These results confirm the ChemoFilter’s ability to successfully absorb DOX in vitro, showing promise for its possible future use in clinical trials.
Virginia Hinostroza
Advisor: Dr. David Saloner, PhD
Thesis Title: 4D Flow with Compressed Sensing for the Evaluation of Intracranial Aneurysmal Flow
Abstract:
ackground: 4D flow (4DF) magnetic resonance imaging (MRI) offers a promising way to evaluate blood flow patterns in intracranial aneurysms, although long scan times present a major limitation to broad clinical implementation. Compressed sensing (CS), an accelerated imaging technique using strategically undersampled data for data reconstruction, offers a possible solution to reduce scan times. The aim of this study was to understand the effects and limitations of varying compressed sensing acceleration factors, R, at different resolutions in in vitro 4D flow acquisitions.
Methods: This study employed a phantom depicting a saccular aneurysm. Experiment 1 evaluated the reliability of 4D flow with varying levels of compressed sensing acceleration factors (R=7.6, 12.8, and 16.6). Experiment 2 assessed the effects of varying resolutions (0.5, 1.0, 1.5, and 2.0 mm) with a compressed sensing R=12.8. Qualitative analysis included a visual assessment of velocity vectors and streamlines. Quantitative analysis compared the velocity components, peak velocity, flow rate, and wall shear stress in each experiment. All studies were post-processed using a clinically-geared software as well as with an In-House engineering pipeline, with the purpose of understanding the advantages and disadvantages of each approach and validating any results.
Results: The addition of compressed sensing reduced scan times to approximately 4-7 minutes. The In-House processing pipeline is superior to the clinical software in visualizing of velocity; visual analysis showed velocity overestimations in the R=16.8 streamlines, indicating the limits of compressed sensing to be 7.6 < R < 12.8. As expected, comparison of velocity components reflects a decrease in linear regression slopes and correlations as acceleration factors increased (m > 0.90 for all acceleration factors except R=16.6, and all r > 0.9). Experiment 2 highlights partial voluming effect at R=12.8: as resolutions decrease, velocities along the wall are less reliable than velocities furthest from the wall, which retain high slope and correlation values (both above 0.9 at 1.0 mm and 1.5 mm resolutions). High variability peak velocity, flow rate, and wall shear stress in both pipelines point to the need for a reliable way to post-process 4D flow.
Conclusion: This study showed reliable velocity data can be obtained from 4D flow studies acquired with compressed sensing lower to moderate acceleration factors at higher resolutions. With clinically-acceptable scan times, the focus now shifts towards establishing a robust and validated workflow for 4D flow studies before clinical implementation can truly be feasible.
Alison Myoraku
Advisor: Dr. Duygu Tosun, PhD
Thesis Title: Lifespan Changes of the Human Insula in Major Depression
Abstract:
Using cross-sectional structural magnetic resonance imaging (MRI) data from six cohorts originating from three sites, this study investigated the cortical morphometric trajectories of six insular subregions of individuals with major depressive disorder (MDD) compared to healthy individuals across the lifespan to better understand the neurodevelopmental and neurodegeneration aspects of MDD. The insula is a centrally located region of the brain responsible for emotional regulation and awareness and has been implicated in many psychiatric disorders including MDD. Participants across all sites included in this study totaled 203 individuals with current MDD (F=137, M=66) and 215 healthy controls (F=110, M=105). T1-weighted magnetic resonance (MR) images from each cohort were registered and segmented using Advanced Normalization Tools (ANTs) and a 3D probabilistic atlas of the human brain, including the following insular regions: posterior long gyrus, anterior long gyrus, anterior short gyrus, middle short gyrus, posterior short gyrus, and anterior inferior cortex. In addition, we examined the amygdala, anterior cingulate gyrus, lateral occipital cortex, cuneus, subgenual cortex, and lateral orbitofrontal cortex. The outputs were then standardized and harmonized to adjust site effects in morphometric measurements while preserving biological variation due to age, sex, relative intracranial volume, and diagnosis. We hypothesized that the relationships among morphometric measures and age are dynamic across the lifespan and influenced by MDD. For each region of interest considered in this study, linear, quadratic, and cubic models were tested to model morphometry and age association first within each group separately and then tested for group-age interaction. Our statistical analyses indicate that the volumes of all insular subregions in the left hemisphere, as well as the right anterior short gyrus, middle short gyrus and anterior long gyrus, exhibited a significant age-associated difference between the control and MDD groups. Furthermore, the group-age interaction revealed that these deviations were particularly significant for 60-79 years of age, indicating co-morbidity of depression with neurodegeneration. Most non-insular regions showed significant differences between groups, but association with age was less robust. Results for the surface area analysis were less conclusive. Further analysis of other metrics related to neurodevelopment (i.e. cortical thickness) will better inform this facet of the disease.
Samuel Shu
Advisor: Dr. Benjamin Yeh, PhD
Thesis Title: Understanding differences in quantification of conventional versus experimental contrast materials with different clinical dual-energy computed tomography systems, and development of universal contrast quantification “handshake”
Abstract:
Dual energy computed tomography (DECT) has unique imaging capabilities with the potential to improve clinical diagnosis compared to conventional single energy CT. The impending development of novel contrast agents that can take advantage of the unique material differentiation capability of DECT could dramatically expand the diagnostic value of this technology. Unfortunately, clinical DECT systems show intersystem variations in contrast quantification. These inter-scanner differences have been recognized to a limited extent in the current literature. A polyurethane abdominal phantom containing various conventional and experimental contrast materials was constructed to quantify the variance in Hounsfield units (HU) across six clinically available DECT systems. The attenuation profiles of conventional and novel contrast materials presented in this study may serve as a means to correct for intersystem differences across the DECT systems examined.
Benjamin Sipes
Advisor: Dr. Srikantan Nagarajan, PhD
Thesis Title: EEG microstates in Neurofeedback Attention Training
Abstract:
Attention has come under acute focus within the neuropsychological world in past decades, and the rise of brain-computer interfaces (BCI) during EEG offers a means to personalize attention training therapies. Semi-stable EEG topographies, called “microstates,” have been found to be functionally relevant to attention-oriented tasks and shown to influence awareness in the time period directly before a stimulus. In a BCI designed to train attention, we may expect to see a group difference in microstates. Specifically, it could be that microstate D—functionally relevant to attention and task-switching—increases while microstate C—functionally relevant to task-negative and saliency networks—decreases within the group that successfully learns via neurofeedback. The reversed pattern may be true in groups that either fails to learn through neurofeedback or received sham neurofeedback. We may also expect microstates D and C to relate to a behavioral outcome measure that indexes training performance. Accordingly, we used EEGLAB to process BCI attention-training data, derive microstate topographies for individual participants, cluster grand mean topographies for the entire study group, and extract temporal statistics to measure microstate temporal presence during pre-stimulus training. Overall, microstate D had greater temporal presence in those who successfully self-regulated neural cognition during the BCI task compared to those who could not achieve this; microstate C had greater temporal presence in those who could not self-regulate neural cognition during the BCI task compared to those who did so successfully. This analysis highlights differences in BCI performance but failed to find meaningful changes over training.
Torie Tsuei
Advisor: Dr. Victor Valcour, PhD
Thesis Title: Longitudinal Changes in Brain Structure and Integrity During Acute HIV
Abstract:
Cognitive impairment persists in the form of HIV-associated neurocognitive disorder (HAND) among chronically infected individuals despite successful viral suppression. Widespread access to combination antiretroviral therapy (cART) has allowed infected individuals to initiate treatment at an earlier time point and effectively reduce the risk of HIV-related mortality and morbidity. The current study examines whether cART, when initiated within days to weeks following infection, can longitudinally preserve brain health. Quantitative magnetic resonance image (MRI) methodologies were used to analyze T1-weighted structural images and diffusion tensor imaging (DTI) metrics. Specifically, region of interest and voxel-wise volume and tensor-based spatial statistics (TBSS) approaches were performed to evaluate differences in brain volumes and white matter microstructure. We examined 31 acute HIV (AHI) participants who had paired month 0 (baseline) and month 24 (two-year follow-up) scans. Participants were comparatively analyzed in both a longitudinal manner to themselves, and a cross-sectional manner against 25 healthy control (CO) participants. As an indication of inflammation, CD8 t-lymphocyte counts was examined as a clinical covariate. The 31 AHI participants had a median (IQR) age of 26 (23-30) years at the time of enrollment and a median (IQR) baseline CD4 count of 576 (370-868) cells/μL. All immediately initiated cART. The 25 healthy controls had a median (IQR) age of 31 (26-37) years. Differences of brain integrity in the AHI group followed longitudinally were observed. Baseline CD8 count was significantly associated with increased mean diffusivity (MD) in the longitudinally infected, present in the genu and the splenium of the corpus callosum, the corona radiata, and the superior longitudinal fasciculus (all p<0.05). Structural analysis revealed enlarged corpus callosum (p<0.01) volumes, as well as enlarged caudate and thalamus subcortical gray matter volumes in the longitudinally infected AHI participants (both p<0.05). Differences of brain integrity in the AHI group after 24 months of treatment were observed compared to healthy controls. Specifically, fractional anisotropy (FA) was reduced in AHI at 24 months compared to controls in models adjusting for age in the corpus callosum, the corona radiata, and the left superior longitudinal fasciculus (all p<0.05). Structural analyses revealed enlarged putamen and the caudate volumes (both p<0.05). We conclude that differences in both brain integrity and structural volumes can be seen in AHI with successful viral suppression when compared to healthy controls. Future work will include longitudinal imaging data from healthy controls who are followed over two-year follow-up to ensure that observed changes are disease specific. We will also comparatively investigate longitudinally treated chronic HIV-infected participants (‘positive controls’) to examine if the differences are similar to those seen when therapy is initiated in the chronic stage of infection. We will examine inflammatory plasma and CSF biomarkers to inform potential mechanisms and, separately, cognitive testing to inform clinical significance.
Jack Williams
Advisor: Dr. Steven Hetts, MD
Thesis Title: Evaluating EPflex MRline Guidewire for Endovascular Interventions Guided by MRI at 3T vs. X-ray Fluoroscopy
Abstract:
This project sought to evaluate the efficacy of using the EPflex MRline guidewire for endovascular treatment guided by real-time Magnetic Resonance Imaging (MRI). MRI theoretically has numerous advantages over x-ray, the current clinical standard imaging modality for endovascular procedures. The most profound advantage of MRI is the ability to acquire physiological functional information, via perfusion and diffusion measures, for better intervention planning before and during the procedure. The EPflex guidewire was selected because of its ability to perform in an MRI and x-ray fluoroscopy environment, allowing for relevant and useful comparisons of the guiding imaging modality. An abdominal aorta phantom was used to assess the ability of experienced and inexperienced operators to successfully navigate the wire under each imaging modality. It was found that x-ray guidance provided statistically faster and more successful navigation attempts than MRI guidance; however, more clinical tests need to be performed in order to assess the clinical significance of these results.
This study represents an important step in the direction of developing safer and more effective imaging systems for guiding endovascular procedures.
Fei Xiong
Advisor: Dr. David Saloner, PhD
Thesis Title: Prediction of Abdominal Aortic Aneurysms (AAAs) Progression by Automatic Segmentation and Radiomics Feature Quantification
Abstract:
An accurate assessment of abdominal aortic aneurysm (AAA) progression is essential to its clinical management. Currently, the maximum diameter of AAA at diagnosis is considered as the primary indicator of rupture risk. However, it is not optimal as rupture can happen at any size. Several patient-specific factors may also influence AAA rupture risk. Given the clinical variability in aneurysm progression, additional prognostic markers are desirable to enhance patient-specific risk stratification. Radiomics is an image processing technique that extracts quantitative and high-dimensional features from medical images. While it has emerged as a novel approach for solving diagnosis in oncology, its application in cardiovascular diseases is still limited.
This study set out with an aim to determine the feasibility of radiomics in identifying AAA with a fast growth rate (>0.3cm/year) using CT images. An automatic AAA segmentation algorithm was developed in our pipeline. Based on the radiomics features of an 84 CT dataset, supervised classification models were implemented with two feature selection algorithms and two classifiers in a machine-learning framework. An AUC of 0.80 was achieved and the predictive power was proved through comparisons to the maximum diameter and conventional risk factors. Further multivariate analysis suggested that a radiomics-based classification model could be used as an independent, yet strong predictor for fast AAA growth rate.
Jiamin Zhou
Advisor: Dr. Ashish Raj, PhD
Thesis Title: Spectrome-AI: a Neural Network Framework for Inferring MEG Spectra
Abstract:
Computational modeling is a tool that allows for biological systems involving large networks to be studied, such as in studying the correlations between structural connectivity and functional connectivity in the human brain. Raj et al. proposed the spectral graph model in 2019 as a linear, low-dimensional alternative to conventional neural field and mass models that are more computationally expensive, especially when optimizing parameters, which is necessary in order to obtain quantitative and qualitative information about functional neural activity. The initial method used for inferring the spectral graph model parameters was Markov chain Monte Carlo (MCMC) sampling, which provided a robust way to estimate what the target parameter distributions were most likely to be. However, MCMC methods are still slow and computationally expensive. In this study, we trained a fully connected neural network on MCMC-simulated magnetoencephalography (MEG) data to perform parameter estimation for the spectral graph model in an accelerated manner. We found that the neural network was able to predict most parameters of interest without much loss in precision while generating the parameters in less than a second. This approach puts us closer to obtaining real time neurophysiological information from functional neuroimaging data for applications in diagnosis, prognosis, and characterization of various neurological diseases.