DSpace@RPI (Rensselaer Polytechnic Institute)
Not a member yet
6809 research outputs found
Sort by
Deep neural networks for mri applications
Get PDFMay 2022School of EngineeringMagnetic resonance imaging (MRI) has been widely used for clinical disease diagnosis and neuroscience research since its invention. Compared with other commonly used medical imaging modalities like computed tomography and ultrasound, MRI has advantages in showing soft tissue in rich contrast, introducing no ionizing radiation during the scanning, and presenting either anatomical or physiological information through the flexible configuration of the scanning pulse sequence. It is estimated there are around 40 million MRI scans conducted in the United States every year. Despite MRI having achieved great success in recent years, it still has some limitations like long scan time and low signal-to-noise ratio. With the emergence of powerful GPU-based computing systems and the collection of large-scale open-access datasets, deep learning has achieved great progress and profoundly changed the world in multiple regions over the past decade, ranging from pattern recognition to healthcare. Numerous studies have shown the performance of neural networks superior to humans on specific tasks. However, studies on working principles of deep learning like network interpretability, generalization, and stability are still insufficient, raising safety concerns and limiting its actual deployment in real-world applications. This dissertation addresses problems in three aspects: 1) using deep learning to overcome existing MRI shortcomings; 2) using deep learning to broaden the application scope of MRI; and 3) exploring unsolved deep learning problems. For existing MRI shortcomings, we aim at 1) shortening MRI scan time without sacrificing image quality, 2) reducing cine cardiac MRI motion artifacts, and 3) correcting the error caused by T2 defocusing happened in current MRI reconstruction. Through conducting experiments on single-contrast and multi-contrast super-resolution and proposing neural networks with advanced architectures, objective functions, and optimization algorithms, we can successfully shorten MRI scan time by at least 4-fold without significant image quality degradation. In the cine cardiac MRI study, we propose a recurrent neural network with convolutions and long short-term memory. According to the feedback of collaborating clinicians, our results greatly mitigate motion artifacts and can better be used for cardiovascular diseases diagnosis. In the study trying to correct the error caused by T2 defocusing that happened in current MRI reconstruction, we propose a deep learning framework integrating the MRI data acquisition process and image reconstruction process, and optimizing both the pulse sequence and the reconstruction scheme seamlessly. According to our pilot simulation results, better MR images can be obtained. For the extension of MRI application, we propose a deep learning-based end-to-end framework that can classify brain metastases based on their primary organ sites from whole-brain MRI scans with minimal human intervention. Through designing neural networks for metastases segmentation, image modality transformation, and classification, we can classify brain metastases into five categories with high accuracy. Our study shows the potential to use a whole-brain MRI scan instead of biopsy for metastases classification in the future, which also contributes to the early diagnosis of cancer. For the exploration of unsolved deep learning problems, we conduct a study on neural network uncertainty. We propose a method to estimate neural network uncertainty so that the reliability of network results can be quantified. We then apply the proposed method to some MRI applications and show that quantifying network result uncertainty can deliver better diagnostic performance and make medical AI imaging more explainable and trustworthy.Ph
DeFi Survival Analysis: Insights into Risks and User Behavior
No full textWe propose a decentralized finance (DeFi) survival analysis approach for discovering and characterizing user behavior and risks
in lending protocols. We demonstrate how to gather and prepare DeFi transaction data for survival analysis. We demonstrate our approach using transactions in AAVE, one of the largest lending protocols. We develop a DeFi survival analysis pipeline which first prepares transaction data for survival analysis through the selection of different index events (or transactions) and associated outcome events. Then we apply survival analysis statistical and visualization methods such as median survival times, Kaplan–Meier survival curves, and Cox hazard regression to gain insights into usage patterns and risks within the protocol. We show how by varying the index and outcome events, we can utilize DeFi survival analysis to answer three different questions. What do users do after a deposit? How long until borrows are first repaid or liquidated? How does coin type influence liquidation risk? The proposed DeFi survival analysis can easily be generalized to other DeFi lending protocols. By defining appropriate index and outcome events, DeFi survival analysis can be applied to any cryptocurrency protocol with transactions
High-efficiency strontium thiogallate based green phosphor for lighting and display
No full textMay 2022School of EngineeringHumans perceive color with the help of photosensitive tissues found at the back of their eyes. The rod and cone cells present in the human retina act as photoreceptors that detect the incoming light and transmit signals to the brain with the assistance of optical nerves. Typically, under photopic vision, the human eye has maximum sensitivity for visible light, corresponding to the greenish-yellow region of the visible spectra. Phosphor plays an indispensable role in modern-day lighting applications and a coating of phosphor can be found in almost all modern fluorescent and solid-state lighting devices. Furthermore, it is very difficult to achieve high efficiency direct green emission from LED’s, a challenge often referred to as the “Green Gap” by the industry; consequently, synthesizing efficient green-emitting phosphors is crucial for the lighting industry. Inspired and motivated by the above facts, the primary objective of my research is to synthesize a phosphor that would generate an intense green emission upon blue or ultraviolet (UV) excitation.
The Europium (Eu2+) doped Strontium Thiogallate (SrGa2S4) phosphor (SrGa2S4:Eu2+) is a well-known high luminescence efficiency phosphor that emits green light in the 530-550 nm range when excited by blue or ultraviolet light in the 200-450 nm range. However, the biggest challenge with this material is its rapid luminescence degradation when left under ambient condition for several weeks. This limits the use of this phosphor for LED and display devices. Coating the surface with protective barrier layers has been used to provide better stability. However, they are not adequate for shelf-life of 20-25 years which is necessary for lighting and display applications. The origin of the luminescence degradation can be traced to the presence of multiphase crystallites in the synthesized powder using the traditional solid-state growth method. The multiphase crystallites consist of compositions that have poor luminescence efficiency and high reactivity with oxygen. In this research, a new synthesis process has been developed to reduce or eliminate the unwanted low-efficiency phases from the phosphor powder matrix. Followed by the solid-state synthesis, a high-temperature flux crystal growth was used. Phosphors generated using the high-temperature flux growth step exhibit enhancement in luminescence efficiency by 100-150% compared to the solid-state synthesized phosphors. In addition, luminescence degradation has been found to be significantly reduced. Phosphors left under ambient conditions for months showed minimal change in luminescence efficiency. An intense green luminescence with an emission peak at 535 nm was found in the synthesized phosphors indicating that the primary phase of the phosphor composition was unaltered by the flux growth step. An additional benefit of the process used in this research is the elimination of hazardous Hydrogen Sulfide gas that is being used in the industry for synthesizing these phosphors. The synthesis process used in this research is based on solid-state starting materials. Future research efforts are necessary to optimize the flux growth temperature and crystal growth rate (growth duration) to study the trade-off between crystal phases and luminescence efficiency and degradation (shelf-life) of this compound. The effect of higher flux growth temperature on the decomposition of the high-efficiency thiogallate phase needs to be researched and understood.M
Lakes responses to regional and global environmental changes
No full textAugust 2021School of ScienceFreshwater is a vital natural resource currently under threat from multiple anthropogenic stressors. The effects of watershed and atmospheric disturbances often present themselves in lakes, given their deep-lying position within a watershed. Water transparency in many north-temperate lakes is rapidly decreasing due to increasing loads of terrestrial dissolved organic matter (“DOM”); commonly referred to as lake browning. At the same time, the planet is undergoing rapid and widespread changes in climate. Lake surface temperatures are warming at record rates following increases in air temperatures. Further, wide-spread declines in surface wind speeds have been observed across much of the northern hemisphere; a phenomenon referred to as “atmospheric stilling”. Watershed and atmospheric changes can fundamentally alter lake ecosystem structure and function, which in turn will reduce many of the ecosystem services lakes provide such as safe drinking water, angling, and recreational opportunities. This dissertation investigates how changes in water transparency and climate interact to alter physical and chemical lake features, and further examines how these changes will ultimately alter habitat availability and growth rates of common sport fish.
Though a clear pattern has been shown between lake surface temperatures and climate warming, less is known about how the multiple factors of climate change affect summer thermal stratification. Summer stratification depth and strength are two characteristics that regulate many in-lake processes including the creation of a distinct vertical thermal gradient, which strongly regulates the distribution of many organisms. Understanding the role climate plays on stratification depth and strength can be difficult because non-climate factors like water transparency can strongly influence stratification features. In the first study, I utilized a lake in a protected National Park as a case study in order to isolate the effects of climate change on stratification depth and strength. Crater Lake is a near-pristine lake which has exhibited little changes in water transparency and watershed land use over the last 25 years. I examined long-term trends in water temperature profiles and meteorological conditions to determine how summer stratification characteristics changed from 1993 to 2017. I next calibrated a hydrodynamic model and performed scenarios to investigate how changes in climate variables (mainly air temperature and windspeed) alter stratification characteristics. Summer depth and strength of stratification were regulated by different climate variables. I found that the depth of stratification decreased by 55% across the 25-year period, and that this decline was most likely driven by a decrease in wind speeds. While there was no clear long-term pattern in summer stratification strength, I found variability in stratification strength was largely driven by variation in air temperature, with warmer air temperatures resulting in stronger stratification. Notably, I found that spring time conditions in both wind speed and air temperature strongly influenced summer stratification characteristics. Since wind speeds are declining elsewhere and air temperatures are warming across the globe, other large lakes may be experiencing similar stratification changes.
North temperate lakes have been browning across North America and Europe following recovery from acid deposition, and changes in precipitation and land use practices. Browning associated decreases in water transparency can strongly influence lake productivity due to the light absorbing nature of DOM. However, limiting nutrients (i.e., phosphorus and nitrogen) could be associated with DOM, which may help compensate for the negative light absorbing effects of DOM on lake productivity. I utilized two spatial surveys across the United States and a long-term lake survey of 28 lakes in the Adirondacks to understand how limiting nutrients and DOM characteristics were related. Across space, limiting nutrients were strongly positively related with DOM concentrations (quantified here as dissolved organic carbon concentration; “DOC”) and DOM specific absorbance. Adirondack lakes strongly increased in DOC concentration and became browner from 1994 to 2012, but limiting nutrients did not increase. Instead, phosphorus concentrations largely stayed the same while nitrogen concentrations decreased. Further, modeling of lake photosynthetic potential in each of the 28 lakes indicates that most lakes have likely decreased in whole-lake productivity from 1994-2012. Contrasting trends in DOM and limiting nutrients suggests that the strongest effect of lake browning will likely be a decrease in lake productivity through time.
Many north temperate lakes are warming and browning at the same time. Both warming temperatures and decreases in water transparency have the potential to threaten lake dissolved oxygen (“DO”) levels. The relative importance of each driver and the combined effects of warming and browning are not well understood. The third study investigates how warming air temperatures and browning affect DO levels via hydrodynamic and biogeochemical modeling scenarios across a 30-year period (1990 – 2019). I calibrated a hydrodynamic and coupled biogeochemical model to Lake Giles in Pennsylvania. Lake Giles has a robust long-term data record and is a small oligotrophic lake, which is representative of many north temperate lakes. After model validation, I recreated Lake Giles across three different initial DOC concentrations to represent a wide initial range of DOC. For each of the three lakes, I simulated trends in browning and climate warming across the 30-year period and calculated annual summer DO metrics. Lakes with more DOC tended to have less DO. Browning tended to reduce DO in clear lakes quicker than in lakes with higher initial DOC concentrations. Climate warming had a smaller negative effect on DO concentrations. Browning increased the prevalence of anoxia and hypoxia in the summer, and the combined browning and warming scenarios generated the highest levels of summer anoxia and hypoxia. An increase in DOC from 1 to 5 mg L-1 shifted the onset of anoxia and hypoxia up to one and half months earlier. This study has major water quality implications, as the onset of anoxia and hypoxia can release nutrients stored in the sediments, which can fuel algae blooms in the surface waters.
Browning traps more heat at the lake surface, which leads to warmer surface waters and cooler deep waters in the summer. Fish are ectotherms, and as such many functions including metabolism and growth are temperature dependent. Browning therefore has the potential to alter growth rates in fish. In this last study, I again used Lake Giles as a case study. I used a hydrodynamic model to recreate Lake Giles at different DOC concentrations from 1 to 15 mg L-1 during a typical year. I then examined how the growth of two common fish with different temperature preferences varied in response to browning-induced changes in water temperature. Largemouth Bass (Micropterus salmoides) are a common warm-water species present in many north temperate lakes. As DOC concentrations increased, summer surface temperatures warmed and shifted closer to bass optimal temperatures. Thus, increases in DOC led to larger growth rates for bass, most noticeable in the 1 to 3 mg L-1 range. Brook Trout (Salvelinus fontinalis) are a common cold-water sport fish that inhabits deeper parts of the lake in the summer. Browning led to a slight decrease in deep water temperatures, especially in the spring. As a result, trout growth in the spring slightly reduced with browning, but the overall affects were minimal. Though trout growth was minimally affected by DOC, I estimated that trout would have to move shallower in the water column to track their preferred temperature, which could lead to changes in behavior and interspecific competition. Overall, I found differential effects of lake browning on fish growth via temperature depending on fish type (i.e., warm-water vs cold-water fish). Lakes may become dominated by warm water species as lakes continue to brown if food supplies remain suitable.Ph
Machine learning in health informatics
No full textAugust 2022School of ScienceThis dissertation concerns applications of Machine learning in health informatics. With an eye on personalized care as a main driver, we contribute on the following topics, (i) interpretable AI with applications in health
(ii) scalable healthcare by mimicking the experts
(iii) evaluations of health interventions. For interpretable analysis at observational data, we contribute two models, the Supervised Gaussian Mixture of Experts (SGMM) and the Supervised Bernoulli Mixture of Experts (SBMM). We demonstrate by applying them in the Statewide Planning and Research Cooperative System (SPARCS) data, their ability to outperform most of the black box machine learning models, while providing interpretable solutions as well as highly predictable subpopulations, which could be the basis for actionable policy.
For scaling up the reach of healthcare experts, we consider complex care management (CCM).
Complex care management aims to effectively assist patients to manage medical conditions and reduce hospitalizations. Doctors and health plan providers are responsible for assigning eligible patients to a CCM program. Some subjects eligible for the program don't get enrolled due to capacity constrains and overload of the physicians. We provide a decision support framework to assist doctors and health plan providers in this referral process. We view the medical condition of a patient as a sequence of "healthy" (H) and "sick" (S) events, where the sick condition indicates that the patient needs to be in CCM. Label bias, incorrect labels, and unbalanced classes are some of the challenges in using supervised learning to predict the H/S states. We solve this problem using a modification of hidden markov models, the HMM-BOOST model. The framework is general and can be applied in a multitude of problems of the same nature. We test the model with propriatery data from a local Health Maintenance Organization (HMO), Electroencephalogram (EEG) data from Boston Children's Hospital as well as in simulation using synhthetic data.
For evaluation of health interventions, our primary goal is to develop a theory for estimating effect in non targeted trials, which is a common setting for health programs
aiming to improve participants lifestyle by changing their everyday habits. We apply our methodology to evaluate a health program in proprietary data from a local HMO. In a non targeted trial the inclusion criteria are loose. This leads to application of the intervention widely producing a heterogeneous treated population. When most patients treated are healthy, this can obscure the true effects of an intervention. We develop an asymptotically consistent non parametric method (PCM) that provably recovers heterogeneous population effects in such a non targeted trial setting.Ph
Improving parallelism of scientific and engineering applications on heterogeneous supercomputers
No full textAugust 2021School of ScienceThe rising usage of heterogeneous supercomputers introduces both opportunities for increased parallelism and challenges for efficient usage of the available hardware. Applications running on heterogeneous supercomputers must adopt new methods to achieve performance across two levels of parallelism. Inter-process parallelism defines coordination between processes and intra-process parallelism within each process. This thesis presents research towards improving inter-process and intra-process parallelism for applications that use complex data structures such as distributed unstructured meshes. Inter-process parallelism is defined by the coupled costs of the partition of load between processes and the communications between processes required as a result of the partition. To achieve optimal performance, partitions must divide computational load evenly between processes while minimizing the additional costs of communications. This thesis addresses improving inter-process parallelism using multicriteria partition improvement multicriteria methods on a generalized structure for a broad set of potential applications. The partition improvement methods are applied to different unstructured mesh setups with partitions up to half a million processes. In the case of heterogeneous supercomputers, intra-process parallelism is dictated by the parallel hardware available to each process for performing computations. For most of the current and next generation US systems, Graphic Processing Units (GPUs) are the parallel hardware available on each node. This thesis addresses methods for intra-process parallelism in the scope of particle-in-cell simulations with a novel approach to the storage of the unstructured mesh and the particles for optimized performance on GPUs while utilizing performance-portable methods for performance on future hardware. Scaling studies of these methods are presented up to 4096 nodes of the Summit supercomputer with over a trillion particles simulated.Ph
Multiscale dynamics in complex materials and interfaces
No full textAugust 2021School of ScienceMaterial response to light irradiation is at the heart of a wide range of emergent technologies that are in the nano- to quantum length scale. These technologies include interferencenanolithography, optoelectronics, and spin-based quantum information processing devices.
Efficient exploitation of material response is the key to the full realization of these technologies. In the context of these applications, this means investigating the charge, spin, and
coupled optical dynamics with kinetic reactions in a photoresist at timescales ranging from
femto- to microseconds. To this end, first-principles methods have emerged as powerful tools
in simulating the dynamics with the required level of detail in disentangling multiple pro-
cesses in the microscopic and macroscopic regimes. In this talk, I will present my studies of
these dynamics from first-principles with three specific aims: (1) charge carrier (electron and
hole) dynamics in plasmonic materials for finding key parameters affecting plasmonic hot
carrier device efficiency, (2) spin carrier (electron and hole) dynamics in materials promising
for spin-based technologies, and (3) coupled optical dynamics with kinetic reactions in a
photoresist to explore parameters for quality nanopatterning.
First, searching for new promising plasmonic materials, we find that transition metal
nitrides (TMNs) and Ag-Au alloys have hot carrier properties (e.g., lifetimes and mean
free paths) comparable to Au. This coupled with their optical tunability and stability as
small nanoparticles in extreme conditions make them highly favorable candidates for hot-
carrier applications operating in infrared to ultraviolet regimes. Second, probing the hot hole
injection in Au on p-type GaN, we predict that ∼90% of hot holes can cross the interfacial
barrier, suggesting hot-hole driven devices to perform similar to hot-electron driven devices
for optoelectronics. A similar investigation of hot hole injection in Cu on p-GaN suggests that
harnessing hot holes with p-type semiconductors is a promising strategy for plasmon-driven
photodetection across the visible and ultraviolet regimes.
Next, searching for materials with suitable spin properties, first, we establish a new, ac-
curate and universal first-principles methodology based on Lindbladian dynamics of density
matrices to calculate spin dynamics in solids with arbitrary spin mixing and crystal symmetry. Applying this method to calculate spin-phonon relaxation rates in MoS2, in addition to
the excellent agreement with experimental measurements, we gain important insights about the intervalley and intravalley contributions to spin relaxation. We find that intervalley
scattering is dominant for holes at low temperatures because of the large spin-orbit split in
the valence bands. In contrast, for conduction electrons, it is the intravalley scattering that
dominates at all temperatures. For graphene, density-matrix dynamics simulations reveal
that electric fields and substrates strongly reduce spin-phonon relaxation lifetimes to the
nanosecond scale as shown in experiments. We find that hBN increases the out-of-plane
to in-plane lifetime ratio from 1/2 to 0.8, matching experiments, suggesting that intrinsic
spin-phonon relaxation is likely the limiting factor for graphene-based spin technologies at
room temperature.
Lastly, we develop a coupled electromagnetic (EM) and reaction kinetics simulation method that can predict optical dynamics including absorption, diffraction, and intensity
modulations coupled with the photo-activated and inhibited reaction kinetics in a material
in 2 dimensions. This method, simulating a two-color interference lithography technique for
a periodic pattern of lines, shows that diffraction effects are negligible (< 0.1%) for film
depths up to 10 μm. Simplifying the experimental configuration from using two standing
waves for both colors to a plane and a standing wave combination, we can achieve a line
contrast as good as 80% at optimal exposure times. Our EM solver based on perturbation
theory provides a computationally efficient method to be coupled to the kinetic reactions
of any future two-color photoresist candidate for the study of optimal parameters for the
nanopatterning technique.Ph
Machine learning strategies for power systems
No full text2021 AugustSchool of EngineeringAs renewable energy penetration into the United States power system continues to increase, maintaining situational awareness of the complex power grid becomes increasingly challenging. Machine learning models developed from classical optimization theory have been investigated and implemented across many practical disciplines. This dissertation addresses adapting some of these models to create data-driven approaches trained by real power system signals, offering unique advantages and improvements over existing methods. In the first part of the dissertation, the use of constrained machine learning models for power system stability is investigated and a convolutional neural network is developed into a classifier for use in transient stability assessment. The second part of this dissertation deals with the challenge of reliable Th\'evenin equivalent model estimation. An algorithm for estimating equivalent values from existing voltage and current measurements is developed and tested using real power system data. This allows for the supervised training of a recurrent neural network toward Th\'evenin equivalent regression.Ph
EaT-PIM: Substituting Entities in Procedural Instructions Using Flow Graphs and Embeddings
No full textWhen cooking, it can sometimes be desirable to substitute ingredients for purposes such as avoiding allergens, replacing a missing ingredient, or exploring new flavors. More generally, the problem of substituting entities used in procedural instructions is challenging as it requires an understanding of how entities and actions in the instructions interact to produce the final result. To support the task of automatically identifying viable substitutions, we introduce a methodology to (1) parse instructions, using NLP tools and domain-specific ontologies, to generate flow graph representations, (2) train a novel embedding model which captures flow and interaction of entities in each step of the instructions, and (3) utilize the embeddings to identify plausible substitutions. Our embedding strategy aggregates nodes and dynamically computes intermediate results within the flow graphs, which requires learning embeddings for fewer nodes than typical graph embedding models. Our rule-based flow graph generation method shows comparable performance to machine learning-based work, while our embedding model outperforms baselines on a link-prediction task for ingredients in recipes.Robert Bosch LL
Should we tweet this? Generative response modeling for predicting reception of public health messaging on Twitter
No full textThe way people respond to messaging from public health organizations on social media can provide insight into public perceptions on critical health issues, especially during a global crisis such as COVID-19. It could be valuable for high-impact organizations such as the US Centers for Disease Control and Prevention (CDC) or the World Health Organization (WHO) to understand how these perceptions impact reception of messaging on health policy recommendations. We collect two datasets of public health messages and their responses from Twitter relating to COVID-19 and Vaccines, and introduce a predictive method which can be used to explore the potential reception of such messages. Specifically, we harness a generative model (GPT-2) to directly predict probable future responses and demonstrate how it can be used to optimize expected reception of important health guidance. Finally, we introduce a novel evaluation scheme with extensive statistical testing which allows us to conclude that our models capture the semantics and sentiment found in actual public health responses