22689 research outputs found
Sort by
Engineering a Large Matched MIMIC-III ICU Waveform Dataset with Clinical Covariates to Determine the Association Between Hemodynamic Variables and Acute Kidney Injury
Get PDFAcute Kidney Injury (AKI) is the dysfunction of the kidney that happens suddenly. Besides, this clinical syndrome worsens the status of a significant number of hospitalized patients [1]. According to iData Research, there are more than 900,000 cardiac surgeries each year in the United States, and this number is still increasing every year [2]. Up to 30% to 40% of people who undergo cardiac surgery develop AKI [3], [4]. Therefore, patients need to be diagnosed at an early stage in order to be treated successfully.
Physiological experiments in a dog model of AKI demonstrate a causal relation between low blood pressure and kidney dysfunction [5]. Studies in humans demonstrate an association between hypotension and the development of AKI in acutely ill patients in the intensive care unit (ICU) [6]. However, the mechanisms for AKI in humans are complex, and a causal relation between hypotension and other hemodynamic variables, such as cardiac output (CO) and central venous pressure (CVP) is less clear. It is essential for researchers to understand the complicated relationship between parameters such as cardiac output (CO) and blood pressure (BP) regarding AKI in order to develop a more effective clinical treatment.
Currently, few large physiologic waveform datasets exist to evaluate these relationships between hemodynamics to guide treatment. This lack of data limits the development and application of novel analytic approaches to elucidate these relationships and improve patient care.
One main reason that physiologic databases are not available is due to the immense complexity (size, frequency, time stamping) that makes data management difficult to combine multiple different data sources with the needed accuracy to conduct studies of causal inference.
This thesis aims to unravel the complex challenge of managing the publicly available MIMIC-III waveform database in order to conduct observational studies between AKI, CO, and BP in physiologic waveform data that will have a direct impact on clinical practice
SMARTPHONE HEAD AND EYE TRACKING FOR HIGH-PRECISION MEASUREMENTS
No full textEye movement assessment is the best evidence-based approach for diagnosing the underlying cause of dizziness in the acute setting. The HINTS (Head Impulse, Nystagmus, Test of Skew) eye examination has proven to be a sensitive and specific test for differentiating central from peripheral causes of dizziness in patients with acute vestibular syndrome. However, the eye movement abnormalities can be subtle and, therefore, challenging to recognize. A way to overcome this barrier is to quantify eye movements using Video-Oculography (VOG) goggles; however, the cost, lack of availability, and the need for a technician have limited their use. In response, our team developed a smartphone application to quantify eye movements using standard mobile phones.
This dissertation attempts to establish the accuracy and use of our smartphone application (Eyephone) in quantifying eye movements. The first contribution of this dissertation is to compare the smartphone app with the FDA-approved VOG goggles and discuss their current state and future potential.
The second contribution of this dissertation is to establish the markers of accuracy and precision of the app's head and eye movement recordings. We demonstrate how accurate the application recordings are in a controlled lab setting. The third contribution is the evaluation of measured induced nystagmus in healthy participants and comparison with the VOG goggles. We answer important questions about various calibration methods and whether we can use an “average” calibration approach instead of individual calibration before each recording.
The fourth contribution of this dissertation is measuring nystagmus in real-world patients in the neuro-vestibular clinic and comparing the results with those of a neuro-vestibular expert.
Finally, we discuss the application’s current state, usability, the required infrastructure, and future plans, including regulatory (FDA) application, stakeholder engagement, payor approval, and the path forward
Exploring and establishing rodent fibrosis models for evaluating inflammatory bowel disease treatment
Get PDFInflammatory bowel disease is a rapidly growing concern in modern medicine. Its prevalence only increases as more countries become developed and it has jeopardized the quality of life of many patients afflicted with the disorder. Despite its prevalence, the mechanism of action for IBD remains unclear, making treatment development quite an uphill battle. We believe that to increase the efficiency of treatment development for IBD, a reliable and accessible model needs to be established so that more laboratories can gain substantial amounts of data in a relatively swift matter. For this paper, we tried to focus on murine models – the most common type of animal model that almost every lab has access to, and we wanted to utilize the foreign body response as an easy artificial way to generate fibrosis that emulates a chronically inflamed bowel. On top of that, we also wanted to establish a model to specifically address fistulae, a particularly complicated symptom that creates painful tunnels between different parts of the digestive tract and is particularly difficult to treat without remission. We believe that this study will benefit all future endeavors toward finding a cure for inflammatory bowel disease
MECHANISMS OF DECISION-MAKING IN A DYNAMIC FORAGING TASK
Get PDFFlexible decision-making in a dynamic world, crucial for survival, involves selecting actions with higher probabilities of favorable outcomes. When these outcomes have varying attributes, the decision-maker faces the challenge of considering both simultaneously. Alternatively, decisions might prioritize one attribute over the other, focusing on the most salient
aspect to the subject. We test these questions using a dynamic foraging task where mice engage in a probabilistic foraging with variable reward sizes. The mice integrate both reward size and probability into their decision-making, as evidenced by the impact of reward size history on their choices. This integration does not occur abruptly; instead, choices are gradually updated based on the reward statistics of the environment, consistent with action value reinforcement learning models. We show that the activity of single neurons in the frontal cortex of mice supports this observation. The extracellular activity of these neurons is modulated by the integrated value of the outcomes’ attributes, rather than individual attributes alone. Our results indicate that the activity of neurons in the frontal cortex correlates with the reinforcement learning model that utilizes reward volume, suggesting that the brain integrates the values of individual outcome attributes to inform decision-making
Investigation Of Nascent Protein Folding And Trigger Factor Action On Nascent Proteins Using Single-Molecule Spectroscopy
No full textProteins begin to fold during their synthesis by the ribosome. When a protein folds during translation is not well understood. I have studied the folding pathway of G-domain of elongation factor G to show that the domain remains unstructured until the entire domain is synthesized (Chapter 2). Specialized ribosome-binding molecular chaperones stabilize unfolded proteins that accumulate co-translationally, preventing premature folding into non-native states. Chaperones also rescue proteins from misfolded structures and protect folded domains against destabilizing interactions. However, the molecular mechanisms underpinning these functions are not well understood. Trigger factor is the first molecular chaperone encountered by nascent proteins emerging from the ribosome. I have developed a single-molecule approach for directly visualizing the interaction of trigger factor with nascent chains, using E. coli elongation factor G as a model protein (Chapter 3). I found that trigger factor dynamically engages with unfolded polypeptides on the timescale of seconds. Binding parameters change with nascent chain length and are dependent on recruitment of trigger factor to the ribosome, suggesting a multivalent mode of interaction. Experiments combining fluorescence detection with force spectroscopy indicate that trigger factor binds to compact nascent chains more strongly than to extended polypeptides. These results suggest that the chaperone compacts its nascent chains clients and keeps elongating proteins poised for productive structure formation. This experimental system sets the stage for exploring general molecular mechanisms of chaperone action on nascent proteins
MEMBRANE-BASED DIRECT AIR CAPTURE
Get PDFClimate change, driven mainly by human activities such as the combustion of fossil fuels, presents a pressing global challenge with far-reaching consequences. This phenomenon is characterized by long-term alterations in temperatures and weather patterns, primarily attributed to the elevated concentrations of greenhouse gases in the atmosphere. Mitigating carbon emissions emerges as a critical strategy in curbing the pace of climate change.
One promising technology for carbon capture is Membrane-Based Direct Air Capture (DAC), which directly captures carbon dioxide (CO2) from the atmosphere. Membrane materials, particularly Polymeric membranes like PDMS, have seen significant advancements, achieving record-breaking CO2 permeance. These membranes exhibit unparalleled potential for various applications, particularly in carbon capture and separation processes. This study investigates the properties of market-available membranes due to the lack of information on them. Also, membrane material and DAC are combined to capture CO2 from ambient air with two different methods to discover the applicability of using DAC to capture CO2. It is found that the thinnest membrane with a low air flow rate and low relative humidity level has better performance compared to other conditions
Artificial Life in the Transwar Japanese Imagination
No full textThis dissertation uses the lens of the jinzō ningen (“artificial” or, more literally, “human-constructed” human) to offer a novel account of transwar (1920s to 1960s) Japanese modernism and media history and their contemporary afterlives. From artificial wombs and test-tube babies to electrical men and pneumatically-powered sages, the late-1920s Japanese mediascape abounded with images of artificial humanity. Prompted in part by the translation of Karel Čapek’s R.U.R. into Japanese in 1923, fictional depictions of jinzō ningen became a regular fixture in newspaper articles, popular science magazines, literary and genre fiction, manga, illustrations, and other visual media, reaching a peak with the 1929 Japanese release of Fritz Lang’s Metropolis (1927). Scholars have often referred to this period as an early “robot boom”; ironically, however, in the earliest Japanese translations of R.U.R., one can already observe a project of developing conceptual alternatives to Čapek’s “robot” to have been underway.
Departing from the popular image of contemporary Japan as a “robot nation,” this dissertation offers a new interpretation of Japanese media history during the transwar period by exploring the role of deliberately non-robotic modes of artificial human production therein. Viewed as microcosms of the media ecologies structuring modern life—which threatened to become what some called “mechanical civilization” (kikai bunmei)—artificial humans afforded the construction of alternative forms of imagined community. Such construction, though enabled by the uneven cultural flows and infrastructure of imperial and post-imperial Japan, was not wholly reducible to nation and empire; rather, as I argue through close examination of artificial human production across literary translation and transculturation, cinematic animation, and design, it consisted in novel forms of aesthetic community emerging through intermedial practices. What I term “artificial human media ecologies” afforded novel forms of community beyond nation and empire by assembling diverse materials and forms, interweaving disparate and conflicting temporalities, and mobilizing old and new media ranging from the handwritten Sinograph to cutting-edge photographic, cinematic, and electronic media
FROM FIRESIDE CHATS TO TWITTER FEUDS: THE INTERSECTION OF PRESIDENTIAL PERSONALITY & FORMATS OF MEDIA COMMUNICATION
Get PDFThe President of the United States is one of the most powerful people in the world. When they speak, the world listens. The purpose of this paper is to examine and explain the importance of American Presidential communication throughout recent American history in a way that blueprints the future. In a case study, seven United States Presidents are historically examined from Franklin D. Roosevelt to Joe Biden with respect to their preferred forms of media communication. While other researchers have explained the importance of understanding Presidential communication, none have examined it alongside the co-developing American media technology. In each instance, there is a link and intersection between the personal needs and personality of the President as an individual and their preferred media platform. Roosevelt hid his physical ailments via radio broadcasts. Kennedy pushed the image of the all-American family in visual media. Reagan used rhetorical skills he honed as an actor to share his policy aims with the public. Clinton channeled his charisma in such a way to avoid one of the biggest Presidential scandals of all time. Obama shared his life on social media like a modern-day influencer, giving Americans unprecedented access into life at the White House. Biden has allowed other politicians to speak for him to avoid his personal struggles with public speaking. And Trump serves as a cautionary tale to future leaders as someone who did not correctly identify this intersection between his personality and his Twitter account. Not only does understanding the media communication strategies help us better understand the Presidency as a whole, but it also helps us better understand our leaders and become more informed voters
Characterizing and Suppressing the Effects of Correlated Noise in Quantum Computing Systems
Get PDFQuantum computing systems are highly susceptible to noise that corrupts quantum information and reduces computational fidelity. This dissertation develops methods for modeling, characterizing, and mitigating spatio-temporally correlated noise. The work begins with a review of quantum computing fundamentals, noise, and the theory of open quantum systems using the filter function formalism.
We introduce a comprehensive noise characterization protocol that integrates Markovian and non-Markovian noise models to learn the parameters governing noise in single- and two-qubit systems. Experimental results from cloud-based quantum devices show significant non-Markovian effects such as colored noise and crosstalk, underscoring the need for specific noise-suppression strategies.
Novel control waveforms are presented, designed to enhance sensitivity to control noise spectra while reducing susceptibility to dephasing noise. Simulated noise spectroscopy validates these waveforms, demonstrating improved fidelity in noise reconstruction.
Further chapters detail the development of noise-robust quantum control protocols. Analytical conditions for crosstalk-robust single-qubit control in multi-qubit systems are derived, with experimental validation on processors with up to 27 qubits showing a tripling in coherence decay times. This technique also enables the first 7-qubit reconstruction of local dephasing spectra through noise injection.
To counteract correlated dephasing noise, we introduce the Filter Gradient Ascent in Function Space (F-GRAFS) framework. F-GRAFS optimizes control settings to avoid spectral regions with high noise power, demonstrating efficacy in two-qubit systems for simultaneously mitigating crosstalk and dephasing.
The dissertation concludes by examining the impact of spatio-temporally correlated errors on the Quantum Approximate Optimization Algorithm, applied to Grover search. Using perturbative filter function analysis, the susceptibilities to various error mechanisms are elucidated.
This suite of techniques, tested on superconducting transmon qubits via cloud platforms like IBM, provides a comprehensive toolbox for enhancing the robustness and scalability of quantum devices. These advancements are anticipated to benefit a broad spectrum of quantum computing architectures
INTRODUCING FUNCTIONAL MONTE CARLO TREE SEARCH METHOD FOR REINFORCEMENT LEARNING
Get PDFMonte Carlo Tree Search (MCTS) has become very influential in reinforcement learning (RL) and consequently artificial intelligence. However, these algorithms require large computational resources to generate good results and they are inflexible in terms of making dynamic corrections as decision trees are developed. Motivated by these limitations, this work introduces the concept of Functional MCTS (FMCTS) that fuses functional graph theory with MCTS to characterize its decision tree paths and corrective alternatives incrementally and explicitly. Next, a hybrid FMCTS and RL algorithm schema is proposed in which the agent has the "freedom" to make internal anticipatory moves (via MCTS); compare them to corrective moves (via functional graph theory); and finalize each move all while incrementally considering the winning likelihood. The empirical results from playing the games of Tic-tac-toe and Connect Four against various trained MCTS agents further illustrate the promise of this approach. Overall, the results of this work demonstrate the direct correlation between current and future decisions with respect to their path deviations as well as can be applied to the development of dynamic and adaptable RL agents