22689 research outputs found
Sort by
Diffuse2Adapt: Controlled Diffusion for Synthetic-to-Real Domain Adaptation
No full textSynthetic data generated from graphics engines has been shown to be effective for learning, while also being a cost-effective alternative to annotating data. However, models trained on synthetic data often face a drop in performance when evaluated on real data due to the synthetic-to-real domain gap. Unsupervised domain adaptation (UDA) techniques attempt to leverage a set of unlabeled target data in various ways for bridging this domain gap. Recently, conditional image generation models such as Stable Diffusion have shown impressive results in generating realistic images from text and image inputs. In this study, we investigate the utility of Stable Diffusion for translating the synthetic images to the target domain for synthetic-to-real UDA. The translated images must accurately represent the class semantics of source domain data while also exhibiting properties of the target domain. We investigate various
strategies to leverage the unlabeled target domain data with Stable Diffusion to guide the generation towards the target distribution
Minors 15 And Older Seeking An Abortion Should Not Have To Obtain Parental Consent
Get PDFAbortion is a deeply politicized subject that has taken over discussions within society. What is not talked about within these discussions is the subject of unsafe abortions. There are several different kinds of abortion restrictions that would push a pregnant person into pursuing an unsafe abortion, one of which is the requirement of parental consent for adolescents under a certain age (which varies by state). Social and familial pressures lead to adolescents seeking unsafe abortions, in order to terminate their pregnancy confidentially.
Unsafe abortions can lead to severe injuries, disability, and even death. If it is impossible to legally get rid of unsafe abortion procedures, then attention must be turned to how to drive people with unwanted pregnancies away from them, and towards safe procedures.
To address this issue, a harm reduction approach is necessary. The precedent of harm reduction, as seen in the context of America’s opioid epidemic, can be translated over to harm reduction in abortion. Abortions exist in the world, but the harms that come with unsafe abortions can be minimized by allowing greater access to safe ones.
Objectors to this approach do so on the grounds that minors do not have the capacity to decide for themselves. There is evidence that demonstrates adolescents 15 and older being competent enough to provide informed consent on a procedure like abortion. Further, there are already pediatric processes in place that do not require parental consent.
Adolescents that do not disclose to their parents do so for many reasons. Requiring consent from a parent may discourage adolescents and instead lead them to seek an unsafe option, and there could also be consequences of seeking the parental consent. A harm reduction lens ethically calls for expanded access of safe abortion care for this population. One way to expand access is to allow adolescents the authority to seek an abortion without parental consent beginning at age 15, rather than 18
Development and Assessment of a Housing Composite Score to Explain Variation in Emergency Room Admissions in Maryland
Get PDFBackground: Social determinants of health (SDOH) have considerable impact on an individual’s health. SDOH factors are increasingly used for risk stratification and population health management efforts. New efforts have focused on aggregating existing geo-derived SDOH factors to develop summarized geo-derived SDOH scores or ratios. Housing conditions are a fundamental SDOH factor affecting a significant portion of the population, and subcategories of housing may have different impacts on population health outcomes. However, the few existing housing composite scores have overlooked the specific contributions of these housing subcategories. This study aimed to develop a housing composite score along with individual scores for various housing subcategories and assess whether these housing composite scores can explain variation in emergency room (ER) admissions.
Methods: This is a retrospective cross-sectional study using the U.S. Census Bureau data. We included 3198 counties and 25,979 zip codes and selected a total of 18 housing variables in the analysis. We applied the principal components analysis approach to develop housing composite scores. Then, we used the Healthcare Cost and Utilization Project data to assess the prediction performance. We included 1,506,719 unique adult patients in Maryland and constructed predictive models for ER admissions using Logistic regression on different combinations of predictors.
Results: Based on principal components analysis, we classified the housing variables into four distinct subgroups: financial, ownership, building quality, and others. A higher score in each domain indicated a better relative condition. Financial score and building quality score have a similar pattern with the scores being higher in urban areas compared to rural areas. However, the ownership score shows a different pattern, with the score being higher in rural areas. Other housing subcategory scores did not show a specific pattern. Adding housing composite scores into base model improved around 0.2% of the AUC. Zip code level composite scores showed a slightly better improvement of ER predictive models’ performance than county level composite scores. In addition, the prediction model performed best for white individuals, but housing composite scores improved the model’s performance the most in the Black population.
Discussion: We developed a new set of housing composite scores using publicly available geo-derived SDOH data. We demonstrated that housing composite scores can statistically significantly improve the performance of predicting ER admission, although such improvements are modest at best. This study provided a conceptual methodology for future investigations into the housing and health relationship
On the diagnosable, efficient, scalable and effective learning
Get PDFDeep learning is a cornerstone in the quest for Artificial General Intelligence (AGI), offering a transformative approach to machines' comprehension of and interaction with the world. At its core, deep learning mimics the complexity and adaptability of human intelligence by processing and learning from vast amounts of data through neural networks. This approach enables machines to perform a wide range of tasks, such as computer vision, natural language processing, and speech recognition, with unprecedented flexibility and learning capacity.
A mature deep learning system comprises a reliable evaluation method, a strong model, and an effective training approach. In this dissertation, I focus on developing algorithms from these three perspectives. For the evaluation, (i) we propose PatchAttack, a black-box adversarial attack algorithm, to diagnose whether a deep network is robust against localized changes that do not confuse humans. For the model, (ii) we introduce the Lite Vision Transformer (LVT), a mobile transformer designed to tackle the problem of insufficient representation capability with limited model parameters. (iii) We propose a model series called MOAT to explore the design principles of scalable vision transformers that benefit both upstream and downstream tasks. For the training, (iv) we design a knowledge distillation algorithm to train neural networks across multiple generations using the same architecture, aiming to explore the benefits of self-supervision. (v) We further propose the Snapshot Distillation algorithm to condense the knowledge distillation process across multiple model generations into a single generation. (vi) Finally, we study the training algorithm for the vision-language model and propose the Information Gain (IG) Captioner, a multimodal GPT model, to explore the benefits of the information gain training method on zero-shot tasks
Assessing the Use of Philadelphia's Bike Share Program Before and After COVID-19
Get PDFBackground: Alternative transportation methods aim to enable car-free mobility by allowing users to travel beyond typical walking distances with ease. There remains limited information on the specific impact of COVID-19 on the utilization of bike share programs in the United States. This study aims to evaluate associations between neighborhood-level environmental and sociodemographic characteristics and bike share program use in Philadelphia, before and after the declaration of the COVID-19 pandemic.
Methods: We conducted a retrospective ecological study of 268 operational bike share stations in Philadelphia that reported usage between January 2016 and December 2023. Linear regression models were used to estimate multivariable adjusted differences (95% confidence intervals [CI]) in daily station ridership comparing the pre-COVID (January 2016 – February 2020) and post-COVID (March 2020 – December 2023) time periods. These values were also estimated across categories of environmental (number of bus stops, metro stations, national walkability index, job density) and sociodemographic (median household income, racial/ethnic composition, educational attainment) characteristics.
Results: Characteristics of highly active bike share stations include public transportation proximity, higher job density, higher median income, higher educational attainment, and primarily white neighborhoods. Stations in neighborhoods with less than 25% white residents saw a decrease of 0.78 (95% CI: 0.69, 0.87) daily rides per station post-COVID, while those in areas with ≥75% white residents experienced a more substantial decrease of 4.96 (95% CI: -4.76, -5.15) daily rides per station. Similarly, the highest categories of educational attainment and median income were associated with a decrease of 4.04 (95% CI: 3.88, 4.2) and 3.04 (95% CI: 2.91, 3.17) respectively. Stations with higher job density, bus stops, metro stations, and walkability were associated with a decrease of 2.39 (95% CI: 2.26, 2.53), 2.58 (2.48, 2.68), 2.37 (95% CI: 2.24, 2.51), and 2.25 (95% CI: 2.16, 2.33) across COVID periods.
Conclusions: This study examined the utilization patterns of Philadelphia’s bike share program before and after the COVID-19 pandemic, revealing persistent disparities across neighborhoods with differing sociodemographic profiles and built environment characteristics. Despite the program’s overall growth and expansion, the findings indicate that the potential benefits of bike sharing are likely not equitable among all population groups
A scalable, non-invasive, high-frequency stimulator for tactile sensory feedback
Get PDFTranscutaneous Electrical Nerve Stimulation (TENS) based tactile sensory feedback has proven effective in enhancing the functionality of prosthetics for amputees, enabling them to perceive and manipulate objects through restored sensations of touch, grasp, and pressure. However, existing TENS stimulators often require high-voltage stimulation to overcome significant skin impedance, limiting their practicality for everyday use. Moreover, the lack of stimulators capable of high-density output further restricts the richness of sensory feedback.
In this work, we developed a scalable, high-density, battery-powered TENS stimulator that has a hardware-based current control with a maximum output current of 6.6 mA. 10 kHz biphasic square waves are generated to significantly reduce the skin impedance and voltage required (32 V). More importantly, our stimulator features a scalable electrode switch module that facilitates arbitrary electrode selection from a 16-site high-density electrode array with minimal crosstalk (2.8\%) and short delays (0.5 ms).
Chapter 1 offers a comprehensive review on tactile sensory feedback, TENS technology, and existing stimulators. Chapter 2 details the design of the device and the associated software programming. Followed by the outline of device characterization and human experiments in chapter 3 and 4 respectively. The results from the device characterization tests are presented and analyzed in Chapter 5
IMAGE-GUIDED ROBOT-ASSISTED ANKLE JOINT REDUCTION: SYSTEM INTEGRATION AND PRECLINICAL TRIAL
Get PDFSyndesmotic ankle fractures, often accompanied by joint dislocations, are common orthopedic trauma injuries that often require surgical intervention. However, conventional fluoroscopy-guided reduction may suffer from limitations in accuracy and precision, which can lead to complications and prolonged recovery times. To address these challenges, we propose an integrated image-guided robot-assisted ankle surgery system aimed at improving the accuracy, precision, and, ultimately, safety of ankle joint reduction. The system consists of a custom-designed robot and a multi-body 3D-2D registration-based image guidance. Leveraging pre- and intra-operative x-ray imaging, our approach enables accurate manipulation of the dislocated bone with closed-loop feedback. The robot, designed in-house, offers three degrees of freedom (DoF), while the image guidance system utilizes automatic segmentation and 3D-2D registration to localize bone and robot positions in fluoroscopic images. Through end-to-end evaluations and pre-clinical trials on ankle cadaver samples, we demonstrate the effectiveness of our system in achieving accurate reduction outcomes. Our evaluations highlight the robustness and reliability of the multi-body registration algorithm, even in the presence of metal implants and varying imaging resolutions. Furthermore, we developed a GUI interface that facilitates seamless interaction between the surgeon and the robotic platform, allowing for efficient planning and execution of reduction targets. By overcoming the limitations of traditional methods, our proposed system has the potential to significantly improve patient outcomes by reducing the risk of complications associated with ankle trauma surgery. Future directions include clinical validation studies to further assess the system's efficacy and integration into routine surgical practice
EPIGENETIC MODIFIER MUTATIONS AND HOW THEY RESHAPE THE DNA METHYLATION LANDSCAPE IN ACUTE MYELOID LEUKEMIA
No full textAcute Myeloid Leukemia (AML) is a deadly and genetically complex hematologic malignancy with older adults having only 15% 5-year survival rate. Despite recent efforts, there is still a fundamental lack of understanding about how some genetic mutations contribute to disease initiation, progression and relapse.
Up to 40% of AML patients have genetic disruptions in epigenetic modifiers (DNMT3A, TET2, IDH1/2, MLL), therefore we curated a cross-institutional cohort of primary AML samples from Stanford University and Johns Hopkins Hospital. The basis of inclusion in our cohort of 35 primary AML was the presence of a mutually exclusive mutation in one of the following epigenetic machines: DNMT3A, TET2, IDH1/2, or MLL. We performed and analyzed whole genome bisulfite sequencing on this cohort of de novo AML blasts. We identified canonical cancer methylation signatures across AML mutant groups such as broad hypomethylation over heterochromatic regions and focal hypermethylation of CpG shores. Mutant-specific signatures also followed some previously established patterns where DNMT3A and MLL-r AML showed stronger global hypomethylation phenotype compared to IDH1/2- and TET2-mutant AML. Interestingly, mutant-specific differentially methylated regions showed an independent convergence on regulatory regions near similar TF targets such as GATA1, GATA2, and RUNX1 albeit in opposing directions. TET2-mutant AML showed almost no unique methylation signature. There were multiple mutation specific motifs and gene sets enriched in each mutant family DMRs as well, which could suggest unique TF network disruptions and targeted therapy approaches.
In addition to conventional directionally intuitive DMRs (hypo and hypermethylated), we also used an information-theoretic model of the DNA methylation landscape (informME)to identify more sophisticated changes to AML methylome. This model incorporates methylation pattern variability and entropy to identify epigenetically disrupted regions. We found that this method revealed various convergently disrupted genes that have been previously reported such as FLI1 and EGLF7. Using our information theoretic metrics we also found splicing factor SRSF9 and DNA repair gene BRCA1 binding sites among the most disrupted in the genome. This beckons further research as to how methylation alters splicing dynamics and DNA repair in AML
FROM COUNTERCULTURE TO CLINICAL CARE: NAVIGATING THE ETHICAL, LEGAL, AND REGULATORY JOURNEY OF PSYCHEDELICS IN AMERICA
Get PDFPsychedelic drugs, though classified as Schedule I substances in the United States, are rapidly emerging as powerful tools for treating mental health conditions. This growing recognition has sparked a resilient demand for their rescheduling and integration into American healthcare. As the psychedelic renaissance gains momentum, the U.S. faces a pivotal opportunity to reclassify these substances, break away from outdated practices, and confront the mental health crisis. However, the medicalization of psychedelics is fraught with complex ethical, legal, and regulatory challenges that must be addressed before they can be implemented effectively.
This study delves into the history of psychedelic drug scheduling, exposing the deep-rooted stigmas fueled by racism and political agendas that have long hindered scientific progress. By analyzing current research and discerning key concerns within the field, this study identifies the critical issues surrounding the medicalization of psychedelics in the United States. Through an extensive literature review and international policy analysis, this project presents a blueprint for safely and ethically advancing the field of psychedelic medicine. Key recommendations include strategies to enhance informed consent processes, rectify cultural appropriation, and develop regulatory frameworks that prioritize harm reduction. Central to these recommendations is a deep respect and acknowledgment of Indigenous cultures, which have long
recognized and harnessed the therapeutic potential of psychedelics.
This project’s holistic approach aims to integrate psychedelics into American healthcare, revolutionizing mental health treatment while upholding safety, equity, and reverence for the cultural origins of these transformative substances
SVCFit: INFERRING STRUCTURAL VARIANT CELLULAR FRACTION IN TUMORS
No full textThe dynamic nature of the cancer genome, characterized by intratumor heterogeneity and multiple cell subpopulations (clones), underscores the importance of reconstructing tumor phylogeny to understand the evolutionary trajectories of cancer. This study introduces a novel approach to estimating the structural variant cellular fraction (SVCF) in tissue samples where tumor and normal cells are mixed. By applying variant allele frequency (VAF) adjustments for different types of structural variants (SVs), including deletions, tandem duplications, and inversions, my method SVCFit aims to achieve more accurate SVCF estimation. Preliminary results demonstrate improved SVCF estimation compared to a comparable published method, SVclone, particularly for deletions and inversions across various levels of tumor purity (proportion of tumor cells in the sample). The method, however, has limitations related to assumptions of constant read depth and heterozygous SVs, as well as challenges in detecting certain SV types due to the constraints of short-read sequencing. Future work will utilize long-read sequencing data to address these limitations and incorporate read-depth variation and SV haplotype information. This research represents a significant step forward in accurately reconstructing tumor phylogeny by considering the cellular fraction of somatic SVs, thereby enhancing our understanding of tumor evolution and its implications for therapeutic outcomes