Doctor of Philosophy, Case Western Reserve University, 2024, EECS - System and Control Engineering

This dissertation investigates the application of digital pathology for developing diagnostic and prognostic tools for 2 diseases: Biliary tract adenocarcinoma and Papillary Thyroid Carcinoma (PTC). We explore the realms of cytopathology, which studies exclusively the morphologies of epithelial cells, and histopathology, which includes the entire tissue region. Bile duct brush specimens are difficult to interpret as they often present inflammatory and reactive backgrounds due to the local effects of stricture, atypical reactive changes, or previously installed stents, and often have low to intermediate cellularity. As a result, diagnosis of biliary adenocarcinomas is challenging and often results in large interobserver variability and low sensitivity. In this dissertation, we first used computational image analysis to evaluate the role of nuclear morphological and texture features of epithelial cell clusters to predict the presence of biliary tract adenocarcinoma on digitized brush cytology specimens. We improved the sensitivity of diagnosis with a machine learning approach from 46% to 68% when atypical cases were included and treated as nonmalignant false negatives. The specificity of our model was 100% within the atypical category. PTC is the most prevalent form of thyroid cancer, with the classical form and the follicular variant representing the majority of cases. Despite generally favorable prognoses, approximately 10% of patients experience recurrence post- surgery and radioactive iodine therapy. Attempts to stratify risk of recurrence have relied on gene expression-based prognostic and predictive signatures with a focus on mutations of well-known driver genes, while hallmarks of tumor morphology have been ignored. In this dissertation, we introduce a new computational pathology approach to develop prognostic gene signatures for thyroid cancer that is informed by quantitative features of tumor and immune cell morphology. We show that integrating gene express (open full item for complete abstract)

Committee: Kenneth Loparo (Committee Chair); Anant Madabhushi (Advisor); Satish Viswanath (Committee Member); Sylvia Asa (Committee Member); Aparna Harbhajanka (Committee Member) Subjects: Artificial Intelligence; Biomedical Engineering; Biomedical Research; Biostatistics; Computer Engineering; Medical Imaging; Oncology; Systems Design

Master of Arts (MA), Bowling Green State University, 2024, Mathematics/Mathematical Statistics

There are 195 countries in the world with many measurable characteristics. Life expectancy, gross domestic product (GDP) per capita, population, and many other features relay meaningful information and serve as a catalyst for statistical analysis. There are three questions of importance we wish to answer from an open and publicly available data set of country data. First, partitioning nations into different levels of power helps answer the natural question of existence of superpower nations that can easily influence global affairs. This is a clustering type of question that we will investigate using k-means clustering. Next, high economic prosperity is associated with other virtuous attributes like high life expectancy and high living standards. A logical question that arises is what the relationship between a country's economic prosperity might be, represented by GDP per capita, to other features like minimum wage, quality of healthcare, and level of industrialization. Linear models and decision trees are very interpretable models that will help uncover the relation between the response and the predictors. Finally, minimum wage reflects and relates to many economic indexes and factors and is therefore of importance to stamp the development stage of a country. There are a handful of nations with no minimum wage regulations, and the intuitive question arises to predict the minimum wage for these countries. We apply several prediction methods, and the neural network and random forests methods yield the lowest validation prediction error. We also compute 95% confidence intervals on each prediction for further interpretation of the results. Throughout, model performance will be measured by different metrics depending on the model at hand. This includes the percentage of variance explained for principal components analysis, R-squared for linear models, and root mean square deviation for prediction accuracy.

Committee: Junfeng Shang Ph.D. (Committee Chair); Riddhi Ghosh Ph.D. (Committee Member); Umar Islambekov Ph.D. (Committee Member) Subjects: Statistics

Doctor of Philosophy, The Ohio State University, 2020, Chemical Engineering

Quantitative Structure-Activity Relationships (QSARs) are empirical or semi-empirical models which correlate the structure of chemical compounds with their biological activities. QSAR analysis frequently finds application in drug development and environmental and human health protection. It is here that these models are employed to predict pharmacological endpoints for candidate drug molecules or to assess the toxicological potential of chemical ingredients found in commercial products, respectively. Fields such as drug design and health regulation share the necessity of managing a plethora of chemicals in which sufficient experimental data as to their application-relevant profiles is often lacking; the time and resources required to conduct the necessary in vitro and in vivo tests to properly characterize these compounds make a pure experimental approach impossible. QSAR analysis successfully alleviates the problems posed by these data gaps through interpretation of the wealth of information already contained in existing databases. This research involves the development of a novel QSAR workflow utilizing a local modeling strategy. By far the most common QSAR models reported in the literature are “global” models; they use all available training molecules and a single set of chemical descriptors to learn the relationship between structure and the endpoint of interest. Additionally, accepted QSAR models frequently use linear transformations such as principal component analysis or partial least squares regression to reduce the dimensionality of complex chemical data sets. To contrast these conventional approaches, the proposed methodology uses a locality-defining radius to identify a subset of training compounds in proximity to a test query to learn an individual model for that query. Furthermore, descriptor selection is utilized to isolate the subset of available chemical descriptors tailored specifically to explain the activity of each test compound. Finally, this (open full item for complete abstract)

Committee: James Rathman (Advisor); Bhavik Bakshi (Committee Member); Jeffrey Chalmers (Committee Member) Subjects: Chemical Engineering

Doctor of Philosophy, The Ohio State University, 2017, Computer Science and Engineering

Clustering is the task of organizing data into natural groups, or clusters. A central goal in developing a theory of clustering is the derivation of correctness guarantees which ensure that clustering methods produce the right results. In this dissertation, we analyze the setting in which the data are sampled from some underlying probability distribution. In this case, an algorithm is "correct" (or consistent) if, given larger and larger data sets, its output converges in some sense to the ideal cluster structure of the distribution. In the first part, we study the setting in which data are drawn from a probability density supported on a subset of a Euclidean space. The natural cluster structure of the density is captured by the so-called high density cluster tree, which is due to Hartigan (1981). Hartigan introduced a notion of convergence to the density cluster tree, and recent work by Chaudhuri and Dasgupta (2010) and Kpotufe and von Luxburg (2011) has contructed algorithms which are consistent in this sense. We will show that Hartigan's notion of consistency is in fact not strong enough to ensure that an algorithm recovers the density cluster tree as we would intuitively expect. We identify the precise deficiency which allows this, and introduce a new, stronger notion of convergence which we call consistency in merge distortion. Consistency in merge distortion implies Hartigan's consistency, and we prove that the algorithm of Chaudhuri and Dasgupta (2010) satisfies our new notion. In the sequel, we consider the clustering of graphs sampled from a very general, non-parametric random graph model called a graphon. Unlike in the density setting, clustering in the graphon model is not well-studied. We therefore rigorously analyze the cluster structure of a graphon and formally define the graphon cluster tree. We adapt our notion of consistency in merge distortion to the graphon setting and identify efficient, consistent algorithms.

Committee: Mikhail Belkin PhD (Advisor); Yusu Wang PhD (Advisor); Facundo Mémoli PhD (Committee Member); Vincent Vu PhD (Committee Member) Subjects: Artificial Intelligence; Computer Science; Statistics

Doctor of Philosophy, The Ohio State University, 2011, Statistics

Over the recent years, machine learning techniques have breathed a new life in to the classical regression framework. The primary focus in these techniques has often been the predictive performance of the estimated models and the models themselves have developed in to sophisticated non-linear predictive machines. In this development, the ubiquitous “kernel-trick” has played a very important role by providing a means to compute the inner products in the unwieldy high-dimensional spaces via simple and easily computable functions on the low-dimensional covariate domains, called as kernels. The domain knowledge of data dictates the collection of kernels suitable for the specific application. In “learning the kernel” paradigm, current state of the art is to use some optimization method to select the best kernel for the data at hand from this collection. The work in this dissertation assumes the existence of a “true” underlying process, a Gaussian Process, (defined by a fully specified covariance kernel) for the given data. The Gaussian Process itself is considered as a prior on the reproducing kernel Hilbert space of functions characterized by the associated kernel. The goal is to make suggestions towards developing some diagnostic tools which can be used to hasten the kernel learning process. In particular, the setup for computational experimentation is restricted to a Gaussian Process Regression framework with some “mild stationarity” and “closure” type of assumptions on the possible family of kernels. Tools are developed based on the generalized cross validation and the functional norm of the estimated functions. The sign-change behaviors of these tools are exploited for diagnostic purposes. For the tool based on generalized cross validation, a result is conjectured based on computational evidence, and partially proved, which attempts to justify the observed sign-change patterns. Complete proofs for the said result are given under some spec (open full item for complete abstract)

Committee: Prem Goel (Advisor); Tao Shi (Committee Member); Radu Herbei (Committee Member) Subjects: Statistics

Master of Science in Mechanical Engineering, Cleveland State University, 2022, Washkewicz College of Engineering

Tissue indentation response is an important metric for understanding how different musculoskeletal regions respond to loading and is a function of the tissue's form. Modern imaging techniques provide information about the internal structures of human tissue. Ultrasound remains one of the most common imaging techniques performed, given its portability and low costs. Prior work and data collection on 100 patients involved the collection of ultrasound images at eight different locations across the musculoskeletal extremities. Given the tissue structure information that the medical imaging provided, it was hypothesized that the mechanical properties of the tissue could be predicted from this data. This work aimed to incorporate various forms of patient data into different machine learning models for the prediction of tissue indentation response. These surrogate models would be capable of prediction of tissue compliance once input features are provided, potentially making them relevant in the clinical domain. Eight different surrogate models were developed, with four statistics models built and four deep learning models built to assess which method and which input factors were most suitable for accurately predicting indentation mechanics. The first four models were informed by tissue thicknesses and indentation region. The statistics surrogate models consist of two pure statistical models, while the other two models were based on a physics-based interpretation of two springs in series. The statistical models showed reasonable capability of predicting tissue surface stiffness, with the mean absolute percent difference v ranging from 25.4% to 29.7% across the four models. The deep learning approach was divided between two separate forms of deep learning. The first model was fed only demographic features, while a second model of demographics and manually extracted tissue thicknesses. These models also showed reasonable capability of predicting tissue (open full item for complete abstract)

Committee: Ahmet Erdemir (Advisor); Antonie van den Bogert (Committee Member); Shawn Ryan (Committee Member); Brian Davis (Committee Member) Subjects: Mechanical Engineering

Master of Science, The Ohio State University, 2022, Materials Science and Engineering

Recent increases in computational power have led to growing enthusiasm about the volume of data that can be collected and analyzed for many applications. However, the amount of data some physical/virtual systems generate is so great that an increased reliance on mathematical, statistical, and algorithmic based approaches to analyze and make decisions from the data is required. Application of these computational tools can lead to sharper decision making and vast amounts of knowledge discovered. The abstraction of the scientific decision-making process has led many researchers to consider observing systems with more tunable experimental parameters. This makes traditional experimentation, which is based on human researchers conducting the experiment and using their intuition to drive the next set of experiments, intractable for these applications. Autonomous experimentation (AE) systems, which are also a byproduct of the computational explosion, are able to address this issue and have found use across the fields of biology, chemistry, and materials science. AE systems are typically capable of conducting certain types of experiments with lower and more reliable turnaround times as opposed to their human counterparts. The automated execution of experiments naturally leads one to think about how those experiments can be parallelized and otherwise completed faster due to the lack of human presence in the experimentation environment. Therefore, AE systems are considered when designing many high-throughput experimentation (HTE) efforts. This thesis presents an overview of the current state-of-the-art for AE systems in Chapter 1, a framework developed to increase the independence of AE systems from human assistance in Chapter 2, and a machine-learning (ML) data processing pipeline that automates the image post-processing phase of the analysis of backscattered-electron scanning electron microscope images in Chapter 3.

Committee: Stephen Niezgoda (Advisor); Joerg Jinschek (Advisor); Sriram Vijayan (Other); Gopal Viswanathan (Committee Member); Oksana Chkrebtii (Committee Member) Subjects: Business Administration; Computer Science; Engineering; Experiments; Industrial Engineering; Information Science; Information Systems; Information Technology; Metallurgy; Operations Research; Robotics; Statistics

Doctor of Philosophy, The Ohio State University, 2021, Environmental Science

Climate change is threatening food security as it is generally perceived to have negative impacts on agricultural production. Understanding this impact is central to adaptations to reduce potential yield loss. However, yield responses to changes in climate are complicated and have not been well understood. This project aims to characterize yield responses to the changing climate by utilizing modeling approaches, which in turn will help develop decision-supporting tools to inform policy or adaptation strategies. In this dissertation, we address several questions in modeling the impact of climate change on crop yield. First, in Chapter 2, we reviewed and synthesized current progress and findings from studies in the last 21 years using data-driven approaches. We found that previous studies generally agree that warming will negatively affect crop yields. For example, maize, wheat, soybean, and rice yield could be reduced by 7.5 ± 5.3%, 6.0 ± 3.3%, 6.8 ± 5.9%, and 1.2 ± 5.2% with 1 °C warming. Climate change could account for 37% of yield variability across the world. We also identified challenges and issues in previous studies, and thus developed a Bayesian model framework in Chapter 3 to overcome part of these challenges. The proposed Bayesian model framework was used in Chapter 4 to characterize spatial variations in yield responses to changes in climate variables with response curves. These response curves could help us identify what threats crop yield of a county is facing or will face and inform adaptation strategies to deal with these threats. If without adaptions, projected climate conditions of more than 36 climate models under four Coupled Model Intercomparison Project 5 (CMIP5) scenarios would benefit crops in some areas but could also cause severe yield loss in others. These yield changes are location- and scenario-specific. The Henry County in northern Ohio, for example, would have a yield increase of 1.2% and 0.7% under RCP 2.6 and 6.0 (both scenarios ar (open full item for complete abstract)

Committee: Kaiguang Zhao Dr. (Advisor); Gil Bohrer Dr. (Committee Member); Jay Martin Dr. (Committee Member); Yanlan Liu Dr. (Committee Member) Subjects: Agricultural Engineering; Agriculture; Climate Change; Ecology; Environmental Science

Doctor of Philosophy, University of Akron, 2020, Integrated Bioscience

Monitoring and maintaining freshwater resources are crucial for humans and animals; we simply would not be able to call Earth home without water flowing over its vast surface. Nevertheless, water resources are stressed by human activities all around the world. We pour pollution into rivers, streams, and lakes with both foreseeable and unforeseeable consequences. Monitoring and understanding the effects of these on water resources is an organized complex problem, where the whole is greater than the sum of the parts. Hence, a complete comprehension of various parts is needed in understanding the system. Addressing such a problem can be quite complicated and possibly fruitless. However, simplifying the challenge into small crucial compartments can help. To this end, we developed a monitoring system to predict pH and dissolved oxygen using classical time series and machine learning methods, as well as, developed tools to identify and track sources of pollutions in waterways. Our models showed classical time series analysis can give better estimate for our short-term predictions than more complicated machine learning models (chapter 2 and 3). Our tool showed it is possible to develop open source hardware and software to acquire accurate water monitoring data that can be used by citizen scientists (chapter 4).

Committee: Francisco Moore (Advisor); Hunter King (Advisor); Zhong-Hui Duan (Committee Member); Yingcai Xiao (Committee Member); John Huss (Committee Member); Christopher Miller (Committee Member) Subjects: Computer Science; Ecology

PHD, Kent State University, 2019, College of Arts and Sciences / Department of Psychological Sciences

In the present study, the author employed tools and principles from the domain of machine learning to investigate four questions related to the generalizability of statistical prediction in psychological assessment. First, to what extent do predictive methods common to psychology research and machine learning tend to produce generalizable predictions; that is, how well do calibrated prediction models actually predict new data points in new settings? Second, how well do the methods considered compare with one another with respect to prediction generalizability? Third, to what extent does a model benefit from incorporation of more or fewer predictors; in other words, how should we value parsimony in applied prediction? Fourth, what is the most effective way to select predictors for model inclusion when attempting to maximize generalizable predictive power in psychological assessment? To address these questions, the author developed numerous predictive models, using multiple prediction criteria, in a calibration sample drawn from an inpatient psychiatric population at a county hospital, then externally validated those models by applying them to one or two clinical samples drawn from other settings. Model generalizability was then evaluated based on prediction accuracy in the external validation samples. Noteworthy findings from the present study include 1) statistical models generally demonstrate observable performance shrinkage across settings regardless of modeling approach, though they may nevertheless retain non-negligible predictive power in new settings; 2) of the modeling approaches considered, regularized (penalized) regression methods appear to produce the most consistently robust predictions across settings; 3) models appear to produce more accurate predictions when allowed to incorporate more, rather than fewer potentially important predictors, indicating parsimony may be over-valued in psychology research; and 4) multivariate models whose predictors (open full item for complete abstract)

Committee: Yossef Ben-Porath Ph.D. (Committee Chair); Mary Beth Spitznagel Ph.D. (Committee Member); Manfred van Dulmen Ph.D. (Committee Member); Richard Serpe Ph.D. (Committee Member); Michael Ensley Ph.D. (Committee Member) Subjects: Clinical Psychology; Psychobiology; Quantitative Psychology

Master of Science, The Ohio State University, 2017, Materials Science and Engineering

The transition of newly developed materials from the laboratory to the manufacturing floor is often hindered by the task of quantifying the material's inherit variability which spans from the atomistic to macroscale. This impedance is coupled with the task of linking this variability observed at these length scales and ultimately correlating this multidimensional variance to the macroscale performance of the material. This issue has lead to the development of statistical and mathematical frameworks for evaluating material variability. In this work, the author employs one such methodology for the purpose of mesoscale variability quantification with the goal to further explore and enhance this framework while simultaneously presenting the pathway as a computational design tool. This stochastic representation of microstructure allows for the delineation of materials to be highly dependent upon the topology of the material's structure and allows for digital representation via statistical volume elements (SVEs). Quantification of the topology of these SVEs can be achieved through utilization of statistical microstructural descriptors (SMDs), which inevitably leads to an extremely high order data set for each microstructure realization. This high order data set can then be dimensionally reduced via kernel principal component analysis (KPCA), thus allowing for the variance of the microstructure to be observed through the generation of microstructure visualizations. Enhancement of these visualizations can then be achieved through the use of the 1-way multivariate analysis of variance (1-way MANOVA). The reduced order SMD data set can then be combined with property results determined via finite element analysis (FEA) producing microstructure-property maps, thus allowing for both the microstructure and property variance to be observed graphically. Lastly, predictive models can be trained on the reduced order SMD data sets and property results utilizing the machine learning te (open full item for complete abstract)

Committee: Stephen Niezgoda Dr. (Advisor); Dennis Dimiduk Dr. (Committee Member); Soheil Soghrati Dr. (Committee Member) Subjects: Materials Science; Mathematics; Statistics

Master of Science in Biomedical Engineering (MSBME), Wright State University, 2016, Biomedical Engineering

This thesis introduces the concept of implementing greyscale analysis, also known as intensity analysis, on endobronchial ultrasound (EBUS) images for the purposes of diagnosing peripheral lung tumors. The statistical methodology of using greyscale and histogram analysis allows the characterization of lung tissue in EBUS images. Regions of interest (ROI) will be analyzed in MATLAB and a feature vector will be created. A feature vector of first-order, second-order and histogram greyscale analysis will be created and used for the classification of malignant vs benign peripheral lung tumors. The tools that were implemented were MedCalc for the initial statistical analysis of receiver operating curves (ROC), Multiple Regression and MATLAB for the machine learning and ROI collection. Feature analysis, multiple regression and machine learning methods were used to better classify the malignant and benign EBUS images. The classification is assessed with a confusion matrix, ROC curve, accuracy, sensitivity and specificity. It was found that minimum pixel value, contrast and energy are the best determining factors to discriminate between benign and malignant EBUS images.

Committee: Ulas Sunar Ph.D. (Advisor); Jason Parker Ph.D. (Committee Member); Jaime Ramirez-Vick Ph.D. (Committee Member) Subjects: Biomedical Engineering; Biomedical Research; Biostatistics; Computer Engineering; Engineering; Health Care; Medical Imaging

PhD, University of Cincinnati, 2011, Engineering and Applied Science: Mechanical Engineering

Feature selection is an active research topic in the community of machine learning and knowledge discovery in databases (KDD). It contributes to making the data mining model more comprehensible to domain experts, improving the prediction performance and robustness of the model, and reducing model training. This dissertation aims to provide solutions to three issues that are overlooked by many current feature selection researchers. These issues are feature interaction, data imbalance, and multiple subsets of features. Most of extant filter feature selection methods are pair-wise comparison methods which test each pair of variables, i.e., one predictor variable and the response variable, and provide a correlation measure for each feature associated with the response variable. Such methods cannot take into account feature interactions. Data imbalance is another issue in feature selection. Without considering data imbalance, the features selected will be biased towards the majority class. In high dimensional datasets with sparse data samples, there will be many different feature sets that are highly correlated with the output. Domain experts usually expect us to identify multiple feature sets for them so that they can evaluate them based on their domain knowledge. This dissertation aims to solve these three issues based on a criterion called minimum expected cost of misclassification (MECM). MECM is a model independent evaluation measure. It evaluates the classification power of the tested feature subset as a whole. MECM has adjustable weights to deal with imbalanced datasets. A number of case studies showed that MECM had some favorable properties for searching a compact subset of interacting features. In addition, an algorithm and corresponding data structure were developed to produce multiple feature subsets. The success of this research will have broad applications ranging from engineering, business, to bioinformatics, such as credit card fraud detection, email f (open full item for complete abstract)

Committee: Hongdao Huang PhD (Committee Chair); Sundararaman Anand PhD (Committee Member); Jaroslaw Meller PhD (Committee Member); David Thompson PhD (Committee Member); Michael Wagner PhD (Committee Member) Subjects: Information Systems

Doctor of Philosophy, Case Western Reserve University, 2011, EECS - Computer and Information Sciences

Medical prognostic models are designed by clinicians to predict the future course or outcome of disease progression after diagnosis or treatment. The data, which are used when these clinical models are developed, are required to contain a high number of events per variable (EPV) for the resulting model to be reliable. If our objective is to optimize predictive performance by some criterion, we can often achieve a reduced model that has a little bias with low variance, but whose overall performance is improved. To accomplish this goal, we propose a new variable selection approach that combines Stepwise Tuning in the Maximum Concordance Index (STMC) and Forward Nested Subset Selection (FNSS) in two stages. In the first stage, the proposed variable selection is employed to identify the best subset of risk factors optimized with the concordance index using inner cross validation for optimism correction in the outer loop of cross validation, yielding potentially different final models for each of the folds. We then feed the intermediate results of the prior stage into another selection method in the second stage to resolve the overfitting problem and to select a final model from the variation of predictors in the selected models. Two case studies on relatively different sized survival data sets as well as a simulation study demonstrate that the proposed approach is able to select an improved and reduced average model under a sufficient sample and event size compared to other selection methods such as stepwise selection using the likelihood ratio test, Akaike Information Criterion (AIC), and least absolute shrinkage and selection operator (lasso). Finally, we achieve improved final models in each dataset as compared full models according to most criteria. These results of the model selection models and the final models were analyzed in a systematic scheme through validation for independent performance evaluation. For the second part of this dissertation, we build prognos (open full item for complete abstract)

Committee: Michael Kattan (Advisor); Mehmet Koyuturk (Committee Chair); Andy Podgurski (Committee Member); Soumya Ray (Committee Member) Subjects: Computer Science