18525 research outputs found
Sort by
Overcoming Data Scarcity Challenges in Medical Deep Learning: Innovations and Strategies
No full textMedical data's scarcity, multimodal nature, and high dimensionality complicate training robust machine learning (ML) models. This work tackles these challenges with novel techniques for efficient learning processes, dataset development with automatic annotation, and multimodal data fusion. First, it addresses the inefficiency of learning systems when handling small and imbalanced datasets, proposing a new method that enhances data efficiency and reduces sample complexity. Second, it introduces a new dataset and automatic segmentation model for diagnosing musculoskeletal soft tissue tumors (MSTTs), facilitating machine-assisted data annotation and supporting the creation of more accurate ML models. Third, it proposes diagnostic models for MSTTs using single modalities and multimodal fusion strategies, demonstrating that integrating various data types (e.g., magnetic resonance images and clinical data) results in more sensitive diagnostic tools. In summary, this thesis provides comprehensive solutions to critical issues in applying ML to limited medical data, aiming to significantly improve ML models' performance and applicability in clinical practice through efficient learning techniques, novel datasets, and advanced multimodal strategies.
First, we address the inefficiency of learning systems in dealing with small and imbalanced datasets prevalent in medical fields. Our novel learning method enhances data efficiency and reduces sample complexity, mitigating the impact of data imbalance on model training. Second, we develop a new dataset and an automatic segmentation model for diagnosing musculoskeletal soft tissue tumors (MSTTs), facilitating machine-assisted data annotation. This dataset includes multi-modal data and aims to support the creation of accurate and efficient ML models. Third, we propose models for MSTT diagnosis using both single modalities and multimodal fusion strategies. We discover that the integration of various data types (i.e., magnetic resonance (MR) images with clinical data) can yield more sensitive diagnostic tools. Our methods provide notable advancements in both the performance and applicability of ML models in clinical practice.
In summary, this thesis presents comprehensive solutions to several critical issues in applying ML to limited medical data. By proposing efficient learning techniques, creating a novel dataset for automatic annotation, and developing advanced multimodal fusion strategies, this work aims to significantly improve the use of ML models in medical diagnostics and patient care
The Coronavirus Helicase and Exoribonuclease in Viral Replication and Fitness
No full textCoronavirus replication is directed by viral proteins that function as critical regulators of infection. The significant public health and agricultural burden of coronavirus diseases emphasizes the need to identify conserved determinants of replication. These studies are necessary in order to support surveillance efforts of circulating and emerging coronaviruses and develop targets for antiviral drugs. In this dissertation research, I investigate the role of the coronavirus helicase and exoribonuclease in virus replication and fitness. The enzymatic activities of both proteins have been established in vitro and have been shown to be important for efficient viral replication, but the contribution of discrete amino acid residues and protein domains was not well-studied. This work used structure-guided mutagenesis, passaging experiments, and amino acid alignments to generate mutant viruses for study in combination with in vitro experiments testing mutant protein function. This parallel approach leveraged viral genetics and biochemistry to define key protein residues across multiple systems. This work identifies a novel determinant of resistance to the approved antiviral remdesivir in the helicase, demonstrates that the helicase zinc binding domain is sensitive to mutagenesis, characterizes a novel regulator of replication kinetics in the exoribonuclease, and presents evidence toward a new model for the spatial-temporal regulation of the coronavirus multi-protein replication transcription complex during early and late infection
Looking to and Processing of Audiovisual Speech and Links with Language in Autism: Effects from Infancy Through Adolescence
No full textThis dissertation encompassed a comprehensive characterization of audiovisual speech processing and its potential influence on language development in autism. First, we reviewed differences in audiovisual speech processing and perception in autism and potential influences on language development. In Chapter 2, we evaluated the stability of variables of audiovisual speech processing and perception derived from measures commonly used in the literature in autistic school-age children. We then focused on one of these variables – the P2 amplitude of the event-related potential (ERP) as measured by EEG in response to audiovisual versus auditory-only speech – to evaluate whether neural processing of audiovisual speech differs in autistic versus non-autistic school-age children, and whether individual differences in P2 amplitude responses covary with vocabulary. Finally, in Chapter 4, we extended our ERP findings to earlier in life, investigating whether ERP amplitudes in response to audiovisual versus auditory-only speech are present in 12-18-month-old infants, differ between infant siblings of autistic and non-autistic children, and covary with looking to audiovisual speech and emerging language ability. Through these approaches, we identified looking to audiovisual speech as measured by eye tracking and P2 amplitudes as measured by EEG as highly stable in autistic children. Additionally, we found ERP amplitude differences in response to audiovisual compared to auditory-only speech across infants and school-age children, but we did not find evidence of group differences in these ERP amplitude effects between autistic and non-autistic children or infant siblings of autistic children and infant siblings of non-autistic children. However, we found substantial individual variability in the magnitude of ERP amplitude differences in response to audiovisual versus auditory-only speech that covary with both looking to audiovisual speech and language ability. Importantly, we identified several participant characteristics that may influence these associations. Altogether, these findings provide ideas for mechanisms of the cascading effects of audiovisual integration onto language that may provide clinical utility for maximizing language outcomes for autistic individuals
Cancer Cell Small Molecule Secretome Induces the Immune Checkpoint NKG2A and Dysfunction of Human CD8+ T Cells
No full textPD-1 blockade has been approved for head and neck squamous cell carcinoma (HNSCC) patients. However, many HNSCC patients do not respond to this treatment, and other tumor microenvironmental (TME) factors may promote resistance to PD-1 blockade. We previously identified increased expression of the inhibitory receptor NKG2A on CD8+ T cells in HNSCC tumors compared to T cells in matching PBMC samples. Mechanisms that promote NKG2A expression and the role of NKG2A on human T cells in the TME, however, are uncertain. We hypothesize that cancer cells can induce the expression of NKG2A and promote a dysfunctional phenotype on CD8+ T cells. Here we show that tumor-conditioned media (TCM) of HNSCC cancer cell lines or malignant ascites is sufficient to induce the expression of NKG2A and other inhibitory receptors on activated CD8+ T cells isolated from PBMCs of healthy donors. Boiling or small molecular weight cut-off filtering did not eliminate the effect of TCM, suggesting a small molecule promotes NKG2A. T cell activation in TCM decreased mitochondrial respiration to metabolically restrain CD8+ T cells. Functionally, T cell activation in TCM reduced CD8+ T cell cytotoxicity as shown by lower production of cytokines, granzyme B, and perforin. Furthermore, TCM prevented CD8+ T cells from killing cancer cells in response to a bispecific T cell engager. Thus, a secreted small molecule from HNSCC cancer cells can induce NKG2A expression and promote T cell dysfunction. Our findings may lead to targets for novel cancer therapies or biomarkers for NKG2A blockade response and provide a model to study T cell dysfunction and impaired metabolism
Addressing Statistical Challenges in Implementing Real-World Evidence-Based Risk Prediction Models Into EHR Systems
No full textClinical prediction models have been widely acknowledged as informative tools providing evidence-based support for clinical decision making. However, prediction models are often underused in clinical practice due to various reasons. One major challenge is to handle missing risk factors in real-time risk score calculation. In this dissertation, we proposed a novel submodel approach for prediction models developed using the model approximation approach for model selection, and later we extended this approach for prediction models developed using logistic regression. The proposed submodel approaches have the advantage of borrowing information from the target population. We conducted comprehensive simulations studies to assess the model performance of our proposed approaches and compared them with the existing “one-step-sweep”-based submodel approach as well as the imputation approach. The simulation results show the proposed submodel approaches are robust to various heterogeneity scenarios and are comparable to the imputation-based approach, while the "one-step-sweep" approach is less robust under certain heterogeneity scenarios. The proposed submodel approaches were applied to address missing risk factor issues in the real-time implementation of the STRATIFY prediction model to safely discharge low-risk acute heart failure patients. Another common challenge is to adapt the prediction model into local setting with uncollected risk factors. Here, we assessed and compared multiple approaches to revise a risk prediction model with proxy risk factors of those uncollected risk factors. The proposed approaches were compared under various simulation scenarios and were later applied to revise the STRATIFY prediction model using EHR data from Henry Ford Health
Cross-Abstraction Artifacts to Detect Adverse Manipulation
No full textAs cyber-physical systems (CPS) such as aircraft and autonomous vehicles become increasingly intertwined with critical infrastructure and software, ensuring the security posture of their underlying hardware components and software programs is of paramount importance. Cyber-physical systems integrate computational processes with physical entities, presenting unique challenges because of the dual exposure to both cyber threats, such as vulnerability exploitation, and physical threats, such as counterfeit components. This dissertation introduces a multi-faceted approach to enhance the security and assurance of CPS by leveraging artifacts at three abstraction levels: physical components, software binaries, and software source code.
First, we introduce a method for unique component identification using electromechanical impedance, providing a mechanism to authenticate and verify the integrity of physical components within a CPS and mitigate the use of undetected counterfeit components in the construction of safety-critical systems.
Second, we introduce a technique to analyze evasive malware binaries by strategically configuring virtual machines that do not expose artifacts detected by a malicious program. We develop a dynamic analysis pipeline that uses optimal virtual machine configurations to faithfully execute evasive malware samples, enabling the collection of execution traces to inform the development of defensive measures.
Third, we introduce a novel approach to vulnerability prediction in software source code by employing augmented Abstract Syntax Trees (AST) and Large Language Model (LLM) text embeddings to capture syntactic and semantic nuances of engineered software. Taken together, these features comprise the input to a lightweight Transformer model that predicts vulnerabilities in individual functions of software source code to minimize the frequency of vulnerabilities found in software
Electrochemically actuated metasurfaces for low power nanophotonic and energy storage devices
No full textIn an age of ubiquitous access to information, metamaterials offer a solution to a wide range of information display needs. Notably, metamaterial-based structural color promises vivid coloration across a broad color gamut, promising fade-resistance and a reduced environmental footprint. However, once structural parameters are set using traditional fabrication methods wavefront operations cannot be modified. In recent years, active elements have been incorporated into metasurface geometries to enable broader functionality controlled by an applied stimulus. In particular, electrochemical ion insertion allows for reversible carrier doping, volume expansion, and phase transition within the same system, and is an attractive actuation mechanism for a meta-optic platform. The research presented here aims to elucidate this platform by studying the use of electrochemical ion insertion and associated lattice modification to manipulate dielectric structural color within the visible spectrum. This study begins with amorphous silicon as a lithium-ion host material, due to its outstanding specific capacity and volumetric expansion during alloying. By designing a silicon structural color metasurface and monitoring its optical response in real time during ion insertion, we demonstrated continuous color bleaching and restoration while simultaneously establishing color resilience after irreversible capacity loss and energy storage ability. In contrast to silicon, anatase TiO2 offers ion hosting ability without associated volume expansion effects and reduced losses at visible frequencies, offering another attractive platform for electrochemically actuated structural color. Here, switching times were improved by minimizing diffusion distances and selecting H+ as a dopant for increased diffusivity in the material. The resulting device improved coloration time by an order of magnitude over actuation times for Li-ion insertion based active structural color. This dissertation demonstrates that electrochemical ion insertion is a viable method for dynamic structural color metasurface operation, and shows that improvement in color shifting, switching timescales, and energy storage ability are possible with careful host material, ion, and electromagnetic unit cell design
Leveraging the Gel-to-Sol Transition of Physically Crosslinked Thermoresponsive Polymer Hydrogels to Enable Reactions Induced by Lowering Temperature
No full textMuch work has been done on the use of heating to trigger reactions via the temperature-dependent removal of a barrier or constraint separating reagents. Far less work, however, has been done on the use of cooling to achieve a similar goal. Numerous applications, such as those involving components or materials susceptible to persistent low temperatures and cases in which energy for heating is not available would benefit from this inverse approach. Hence, in this study we explore whether physically crosslinked hydrogels can be reliably used as thermoresponsive constraints that allow reagents to react only upon cooling. We achieve this by loading reagents into adjacent blocks of thermoresponsive hydrogel and showing that these reagents can only react with each other after the temperature of the hydrogel falls below its lower critical solution temperature (LCST). Above the LCST, the reagents remain sequestered in separate gels and no reaction occurs; this “OFF” state is stable for extended periods. When the system is allowed to cool, the hydrogels liquify and flow into each other, allowing mixing of the embedded reagents (“ON” state). We tune the hydrogels’ LCSTs using NaCl, quantify the NaCl’s tuning effect using rheometry, and determine that reactions are triggered reproducibly at temperatures similar to the tuned LCSTs. We also demonstrate generalizability of the concept by exploring situations involving radically different reaction types. This concept therefore constitutes a new approach to autonomous material behavior based on cooling
EV Fingerprinting: Resolving Extracellular Vesicle Heterogeneity using Multiparametric Flow Cytometry
No full textCancer metastasis remains a major unsolved clinical challenge as over 90% or cancer-related deaths are attributed to metastasis. Dynamic regulation of cell adhesion is crucial for successful metastatic dissemination. Cell adhesion can be fine-tuned through direct modulation of cell adhesion molecules or the functional contributions of extracellular vesicle (EVs). We have previously identified Activated Leukocyte Cell Adhesion Molecule (ALCAM) as a dynamic regulator of cell adhesion by expression of alternatively spliced isoform with differential proteolytic susceptibility that impacts metastasis. EVs are lipid bilayer enclosed particles secreted by cells that mediate of cell-cell communication in normal physiology as well as disease pathology. In cancer, EVs transfer bioactive cargoes (proteins, lipids, and nucleic acids) to facilitate directional migration, cell motility, and speed. Characterization of cargoes and biogenesis is therefore needed to identify populations of EVs with a distinctive function. ALCAM-mediated cell adhesion and EV biogenesis are linked through a common molecular factor, Syntenin-1. ALCAM is connected to the actin cytoskeleton through Syntenin-1 to form stable adhesion complexes. Syntenin-1 also participates in a distinctive class of EV biogenesis where it supports intraluminal vesicle budding through associations with Syndecan and ALIX on maturing endosomal compartments. This overlap in molecular machinery suggests that Syntenin-1 availability through ALCAM sequestration could affect EV biogenesis. Using a novel single-EV flow cytometry method called “EV Fingerprinting”, we showed that ALCAM shedding alters EV biogenesis by enhancing production of larger EVs. Additionally, cargo loading into EVs was affected by ACLAM expression. Examination of EV Fingerprints from bladder cancer cell lines cultured in a 3D organotypic model also showed similar profiles according to ALCAM shedding. In order to better understand the results of EV Fingerprinting, further validation of the method was required. Technical validation of EV Fingerprinting demonstrated that the lipophilic dye, di-8-ANEPPS, enables characterization of EVs based on relative fluorescence through size and liquid order by generating multiparametric data. Dimensional reduction and clustering of data revealed resolved populations of EVs. We used EV Fingerprinting to identify population-specific changes upon molecular perturbation of EV secretion through Rab27a knockdown and CD63 overexpression according to liquid order. Additionally, we determined that cargo partitioning is a non-random process by using multiplex analysis of CD63 and CD81 EV cargoes. In summary, this dissertation identifies a relationship between cell adhesion and EV biogenesis as well as establishes a tool to resolve heterogeneous EV populations
LGBTQ Christian Soldiers: A Deeper Understanding of Their Journey and How Army Chaplains Can Provide More Equitable Spiritual Care
No full textDivinity School Doctor of Ministry in Integrative Chaplaincy Final ProjectsLGBTQ Christian Soldiers who embrace an identity contrary to their religious tradition of origin may experience certain types of spiritual distress, sacramental shame, and moral injury. While current Army policy allows for chaplains to deny specific rites, rituals, and sacraments to LGBTQ personnel, further consideration of the other forms of spiritual care that must be available to all Army personnel from all Army Chaplains is warranted. First, is a description of potential spiritual needs facing some LGBTQ Christian Soldiers and obstacles they may face. Second is a theological reflection and discussion on possibilities for spiritual care. Lastly, several suggestions for strategic practice are introduced including the importance of inter-faith dialogue, clarification on chaplaincy practice in a pluralistic environment, the need for established standards of practice, and the potential for a more integrated approach to spiritual care through the utilization of Acceptance and Commitment Therapy (ACT)