Jongeun Choi - Profile on Academia.edu (original) (raw)

Papers by Jongeun Choi

arXiv (Cornell University), Oct 5, 2021

Deep reinforcement learning has shown its effectiveness in various applications, providing a prom... more Deep reinforcement learning has shown its effectiveness in various applications, providing a promising direction for solving tasks with high complexity. However, naively applying classical RL for learning a complex long-horizon task with a single control policy is inefficient. Thus, policy modularization tackles this problem by learning a set of modules that are mapped to primitives and properly orchestrating them. In this study, we further expand the discussion by incorporating simultaneous activation of the skills and structuring them into multiple hierarchies in a recursive fashion. Moreover, we sought to devise an algorithm that can properly orchestrate the skills with different action spaces via multiplicative Gaussian distributions, which highly increases the reusability. By exploiting the modularity, interpretability can also be achieved by observing the modules that are used in the new task if each of the skills is known. We demonstrate how the proposed scheme can be employed in practice by solving a pick and place task with a 6 DoF manipulator, and examine the effects of each property from ablation studies.

This paper presents a novel class of self-organizing multi-agent systems that form a swarm and le... more This paper presents a novel class of self-organizing multi-agent systems that form a swarm and learn a spatiotemporal process through noisy measurements from neighbors for various global goals. The physical spatio-temporal process of interest is modeled by a spatio-temporal Gaussian process. Each agent maintains its own posterior predictive statistics of the Gaussian process based on measurements from neighbors. A set of biologically inspired navigation strategies are identified from the posterior predictive statistics. A unified way to prescribe a global goal for the group of agents is presented. A reference trajectory state that guides agents to achieve the maximum of the objective function is proposed. A switching protocol is proposed for achieving the global maximum of a spatiotemporal Gaussian process over the surveillance region. The usefulness of the proposed multi-agent system with respect to various global goals is demonstrated by several numerical examples.

Microsystems & Nanoengineering, Mar 20, 2023

This study presents a new technology that can detect and discriminate individual chemical vapors ... more This study presents a new technology that can detect and discriminate individual chemical vapors to determine the chemical vapor composition of mixed chemical composition in situ based on a multiplexed DNA-functionalized graphene (MDFG) nanoelectrode without the need to condense the original vapor or target dilution. To the best of our knowledge, our artificial intelligence (AI)-operated arrayed electrodes were capable of identifying the compositions of mixed chemical gases with a mixed ratio in the early stage. This innovative technology comprised an optimized combination of nanodeposited arrayed electrodes and artificial intelligence techniques with advanced sensing capabilities that could operate within biological limits, resulting in the verification of mixed vapor chemical components. Highly selective sensors that are tolerant to high humidity levels provide a target for "breath chemovapor fingerprinting" for the early diagnosis of diseases. The feature selection analysis achieved recognition rates of 99% and above under low-humidity conditions and 98% and above under humid conditions for mixed chemical compositions. The 1D convolutional neural network analysis performed better, discriminating the compositional state of chemical vapor under low-and high-humidity conditions almost perfectly. This study provides a basis for the use of a multiplexed DNA-functionalized graphene gas sensor array and artificial intelligence-based discrimination of chemical vapor compositions in breath analysis applications.

arXiv (Cornell University), Oct 5, 2021

Deep reinforcement learning has shown its effectiveness in various applications, providing a prom... more Deep reinforcement learning has shown its effectiveness in various applications, providing a promising direction for solving tasks with high complexity. However, naively applying classical RL for learning a complex long-horizon task with a single control policy is inefficient. Thus, policy modularization tackles this problem by learning a set of modules that are mapped to primitives and properly orchestrating them. In this study, we further expand the discussion by incorporating simultaneous activation of the skills and structuring them into multiple hierarchies in a recursive fashion. Moreover, we sought to devise an algorithm that can properly orchestrate the skills with different action spaces via multiplicative Gaussian distributions, which highly increases the reusability. By exploiting the modularity, interpretability can also be achieved by observing the modules that are used in the new task if each of the skills is known. We demonstrate how the proposed scheme can be employed in practice by solving a pick and place task with a 6 DoF manipulator, and examine the effects of each property from ablation studies.

IEEE Robotics and Automation Letters

Deep reinforcement learning has shown its effectiveness in various applications, providing a prom... more Deep reinforcement learning has shown its effectiveness in various applications, providing a promising direction for solving tasks with high complexity. However, naively applying classical RL for learning a complex long-horizon task with a single control policy is inefficient. Thus, policy modularization tackles this problem by learning a set of modules that are mapped to primitives and properly orchestrating them. In this study, we further expand the discussion by incorporating simultaneous activation of the skills and structuring them into multiple hierarchies in a recursive fashion. Moreover, we sought to devise an algorithm that can properly orchestrate the skills with different action spaces via multiplicative Gaussian distributions, which highly increases the reusability. By exploiting the modularity, interpretability can also be achieved by observing the modules that are used in the new task if each of the skills is known. We demonstrate how the proposed scheme can be employed in practice by solving a pick and place task with a 6 DoF manipulator, and examine the effects of each property from ablation studies.

Background: To develop a warning system that can prevent or minimize laser exposure resulting in ... more Background: To develop a warning system that can prevent or minimize laser exposure resulting in kidney and ureter damage during retrograde intrarenal surgery (RIRS) for urolithiasis. Our study builds on the hypothesis that shock waves of different degrees are delivered to the hand of the surgeon depending on whether the laser hits the stone or tissue. Methods: A surgical environment was simulated for RIRS by filling the body of a raw whole chicken with water and stones from the human body. We developed an acceleration measurement system that recorded the power signal data for a number of hours, yielding distinguishable characteristics among three different states (idle state, stones, and tissue–laser interface) by conducting fast Fourier transform (FFT) analysis. A discrete wavelet transform (DWT) was used for feature extraction, and a random forest classification algorithm was applied to classify the current state of the laser–tissue interface. Results: The result of the FFT showe...

RSC Advances, 2020

A two-step machine learning (ML) algorithm for coronary artery decision making is introduced, to ... more A two-step machine learning (ML) algorithm for coronary artery decision making is introduced, to increase the data quality by providing flow characteristics and biometric features by aid of computational fluid dynamics (CFD).

and control of a thermal stabilizing system for a MEMS

Journal of Dynamic Systems Measurement and Control-transactions of The Asme, Oct 21, 2014

This tutorial paper presents the expositions of stochastic optimal feedback control theory and Ba... more This tutorial paper presents the expositions of stochastic optimal feedback control theory and Bayesian spatiotemporal models in the context of robotics applications. The presented material is self-contained so that readers can grasp the most important concepts and acquire knowledge needed to jump-start their research. To facilitate this, we provide a series of educational examples from robotics and mobile sensor networks.

Journal of Medical Internet Research, Jan 13, 2023

Background: Osteoporosis is one of the diseases that requires early screening and detection for i... more Background: Osteoporosis is one of the diseases that requires early screening and detection for its management. Common clinical tools and machine-learning (ML) models for screening osteoporosis have been developed, but they show limitations such as low accuracy. Moreover, these methods are confined to limited risk factors and lack individualized explanation. Objective: The aim of this study was to develop an interpretable deep-learning (DL) model for osteoporosis risk screening with clinical features. Clinical interpretation with individual explanations of feature contributions is provided using an explainable artificial intelligence (XAI) technique. Methods: We used two separate data sets: the National Health and Nutrition Examination Survey data sets from the United States (NHANES) and South Korea (KNHANES) with 8274 and 8680 respondents, respectively. The study population was classified according to the T-score of bone mineral density at the femoral neck or total femur. A DL model for osteoporosis diagnosis was trained on the data sets and significant risk factors were investigated with local interpretable model-agnostic explanations (LIME). The performance of the DL model was compared with that of ML models and conventional clinical tools. Additionally, contribution ranking of risk factors and individualized explanation of feature contribution were examined. Results: Our DL model showed area under the curve (AUC) values of 0.851 (95% CI 0.844-0.858) and 0.922 (95% CI 0.916-0.928) for the femoral neck and total femur bone mineral density, respectively, using the NHANES data set. The corresponding AUC values for the KNHANES data set were 0.827 (95% CI 0.821-0.833) and 0.912 (95% CI 0.898-0.927), respectively. Through the LIME method, significant features were induced, and each feature's integrated contribution and interpretation for individual risk were determined. Conclusions: The developed DL model significantly outperforms conventional ML models and clinical tools. Our XAI model produces high-ranked features along with the integrated contributions of each feature, which facilitates the interpretation of individual risk. In summary, our interpretable model for osteoporosis risk screening outperformed state-of-the-art methods.

Journal of Dynamic Systems Measurement and Control-transactions of The Asme, Oct 21, 2014

Explainable deep learning-based clinical decision support engine for MRI-based automated diagnosis of temporomandibular joint anterior disk displacement

Computer Methods and Programs in Biomedicine, May 1, 2023

Image and Vision Computing, Jul 1, 2015

This paper considers visual feature selection to implement position estimation using an omnidirec... more This paper considers visual feature selection to implement position estimation using an omnidirectional camera. The localization is based on a maximum likelihood estimation (MLE) with a map from optimally selected visual features using Gaussian Process (GP) regression. In particular, the collection of selected features over a surveillance region is modeled by a multivariate GP with unknown hyperparameters. The hyperparameters are identified through the learning process by an MLE, which are used for prediction in an empirical Bayes fashion. To select features, we apply a backward sequential elimination technique in order to improve the quality of the position estimation with compressed features for efficient localization. The excellent results of the proposed algorithm are illustrated by the experimental studies with different visual features under both indoor and outdoor real-world scenarios.

IEEE Transactions on Control Systems and Technology, Mar 1, 2015

In this paper, we present a set of techniques for finding a cost function to the time-invariant L... more In this paper, we present a set of techniques for finding a cost function to the time-invariant Linear Quadratic Regulator (LQR) problem in both continuous-and discrete-time cases. Our methodology is based on the solution to the inverse LQR problem, which can be stated as: does a given controller K describe the solution to a time-invariant LQR problem, and if so, what weights Q and R produce K as the optimal solution? Our motivation for investigating this problem is the analysis of motion goals in biological systems. We first describe an efficient Linear Matrix Inequality (LMI) method for determining a solution to the general case of this inverse LQR problem when both the weighting matrices Q and R are unknown. Our first LMI-based formulation provides a unique solution when it is feasible. Additionally, we propose a gradientbased, least-squares minimization method that can be applied to approximate a solution in cases when the LMIs are infeasible. This new method is very useful in practice since the estimated gain matrix K from the noisy experimental data could be perturbed by the estimation error, which may result in the infeasibility of the LMIs. We also provide an LMI minimization problem to find a good initial point for the minimization using the proposed gradient descent algorithm. We then provide a set of examples to illustrate how to apply our approaches to several different types of problems. An important result is the application of the technique to human subject posture control when seated on a moving robot. Results show that we can recover a cost function which may provide a useful insight on the human motor control goal.

A body of literature garners the importance of the intraluminal thrombus layer (ILT) as a key fac... more A body of literature garners the importance of the intraluminal thrombus layer (ILT) as a key factor for the abdominal aortic aneurysm associated with its growth and rupture. There are, however, multiple, often opposite views, on the role of ILT regarding to the pathological, biosolid, and biofluid aspects. Hence, there is a significant need to understand how the ILT is associated with the growth and the rupture and how an aneurysm, with a given patient, predicts its aneurysmal growth. In this work, we use existing patients’ CT data and investigate the association of the ILT in the aneurysm growth rate, with general and localized trends, and its role of biomechanics. We discuss a statistical framework using a computational model of aneurysm growth, which will empower the prediction capability for the clinical management. Method The first set of 39 CT images from 9 patients is used to generate multiple hypotheses for AAA expansion and biomechanics associated with ILT. For this purpos...

Physics Letters A, 2021

Electron scattering cross sections have been acquired both theoretically and experimentally over ... more Electron scattering cross sections have been acquired both theoretically and experimentally over the last few decades. By combining scattering data with machine learning, this work is designed to provide physics benefits: AI assisted incorrect-data screening, cross section data generation, and inverse design. As a basic task before undertaking these applications, we present essential training procedures in this paper. We trained electron-collision data to train a neural network with the type of each collision. The neural network with two hidden layer was implemented using multilayer perceptrons, the earliest deep learning model. Monte Carlo cross-validation was employed to ensure the reliability of the test results, and the optimal model structure was obtained by performing Bayesian optimization. We evaluated our model through the performance indicators within a confidence interval. The results indicate that the data were well-learned, except for the attachment class. Furthermore, feature investigation was carried out to ensure the decision-making process.

Journal of Biomechanics, 2020

Performance during seated balancing is often used to assess trunk neuromuscular control, includin... more Performance during seated balancing is often used to assess trunk neuromuscular control, including evaluating impairments in back pain populations. Balancing in less challenging environments allows for flexibility in control, which may not depend on health status but instead may reflect personal preferences. To make assessment less ambiguous, trunk neuromuscular control should be maximally challenged. Thirty-four healthy subjects balanced on a robotic seat capable of adjusting rotational stiffness. Subjects balanced while rotational stiffness was gradually reduced. The rotational stiffness at which subjects could no longer maintain balance, defined as critical stiffness (k Crit), was used to quantify the subjects' trunk neuromuscular control. A higher k Crit reflects poorer control, as subjects require a more stable base to balance. Subjects were tested on three days separated by 24 hours to assess test-retest reliability. Anthropometric (height and

IEEE Transactions on Industrial Informatics, 2019

An approximate closed-form formula for calculating the ohmic resistance of a circular multiloop c... more An approximate closed-form formula for calculating the ohmic resistance of a circular multiloop coil with unequal pitches is presented. Skin effect and proximity effect are included in the formula. The proximity effect is expressed as a proximity factor obtained using transverse magnetic fields applied to a wire from the rest of the wires. For verification, the optimum dimension for minimum resistance of wires with an equal pitch is compared with the previous results, and both results agree. The formula is applied to calculate the ohmic resistance of helical and spiral coils and is verified by a 2-D finite-element-method simulation. Both calculation and simulation results are consistent as well. As a practical application, a spiral coil with unequal pitches is designed for uniform mutual inductance, and it is optimized for the lowest resistance using the formula. The measured ohmic resistance of the designed coil also agrees with the calculated and simulated results. The results show that the formula can be well applied to designing circular multiloop coils with minimum ohmic loss in wireless-power-transfer systems.

IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2019

To study complex neuromuscular control pathways in human movement, biomechanical parametric model... more To study complex neuromuscular control pathways in human movement, biomechanical parametric models and system identification methods are employed. Although test-retest reliability is widely used to validate outcomes of motor control tasks, it was not incorporated in system identification methods. This study investigates the feasibility of incorporating test-retest reliability in our previously published method of selecting sensitive parameters. We consider the selected parameters via this novel approach to be the key neuromuscular parameters because they meet three criteria: reduced variability, improved goodness of fit, and excellent reliability. These criteria ensure that parameter variability is below a user-defined value, the number of these parameters is maximized to enhance goodness of fit, and their test-retest reliability is above a user-defined value. We measured variability, goodness of fit, and reliability using Fisher information matrix, variance accounted for, and intraclass correlation, respectively. We also incorporated model diversity as a fourth optional criterion to narrow down the solution space of key parameters. We applied this approach to head position tracking in axial rotation and flexion/extension. Forty healthy subjects performed the tasks during two visits. With variability and reliability measures ≤0.35 and ≥0.75 respectively, we selected three key parameters out of twelve with goodness of fit >69%. The key

IEEE transactions on bio-medical engineering, Jan 2, 2018

An Abdominal Aortic Aneurysm (AAA) is a form of vascular disease causing focal enlargement of the... more An Abdominal Aortic Aneurysm (AAA) is a form of vascular disease causing focal enlargement of the abdominal aorta. It affects a large part of the population, and in case of rupture, has up to 90% mortality rate. Recent clinical recom- mendations suggest that people with small aneurysms should be examined 3-36 months depending on the size, to monitor morphological changes. While advances in biomechanics provide state-of-the-art spatial estimates of stress distributions of AAAs, there are still limitations in modeling its time evolution and uncertainty qualification. To date, there are a few biomechanical frameworks that utilize longitudinal medical images, which would aid physicians in detecting small aneurysms with high risk of rupture. In this study, we use longitudinal computer tomography (CT) scans of AAAs that are captured at different times to predict the spatio-temporal evolution of AAAs' shape in future time. We consider a surface of 3D AAA as a manifold embedded in a sca...

arXiv (Cornell University), Oct 5, 2021

Deep reinforcement learning has shown its effectiveness in various applications, providing a prom... more Deep reinforcement learning has shown its effectiveness in various applications, providing a promising direction for solving tasks with high complexity. However, naively applying classical RL for learning a complex long-horizon task with a single control policy is inefficient. Thus, policy modularization tackles this problem by learning a set of modules that are mapped to primitives and properly orchestrating them. In this study, we further expand the discussion by incorporating simultaneous activation of the skills and structuring them into multiple hierarchies in a recursive fashion. Moreover, we sought to devise an algorithm that can properly orchestrate the skills with different action spaces via multiplicative Gaussian distributions, which highly increases the reusability. By exploiting the modularity, interpretability can also be achieved by observing the modules that are used in the new task if each of the skills is known. We demonstrate how the proposed scheme can be employed in practice by solving a pick and place task with a 6 DoF manipulator, and examine the effects of each property from ablation studies.

This paper presents a novel class of self-organizing multi-agent systems that form a swarm and le... more This paper presents a novel class of self-organizing multi-agent systems that form a swarm and learn a spatiotemporal process through noisy measurements from neighbors for various global goals. The physical spatio-temporal process of interest is modeled by a spatio-temporal Gaussian process. Each agent maintains its own posterior predictive statistics of the Gaussian process based on measurements from neighbors. A set of biologically inspired navigation strategies are identified from the posterior predictive statistics. A unified way to prescribe a global goal for the group of agents is presented. A reference trajectory state that guides agents to achieve the maximum of the objective function is proposed. A switching protocol is proposed for achieving the global maximum of a spatiotemporal Gaussian process over the surveillance region. The usefulness of the proposed multi-agent system with respect to various global goals is demonstrated by several numerical examples.

Microsystems & Nanoengineering, Mar 20, 2023

This study presents a new technology that can detect and discriminate individual chemical vapors ... more This study presents a new technology that can detect and discriminate individual chemical vapors to determine the chemical vapor composition of mixed chemical composition in situ based on a multiplexed DNA-functionalized graphene (MDFG) nanoelectrode without the need to condense the original vapor or target dilution. To the best of our knowledge, our artificial intelligence (AI)-operated arrayed electrodes were capable of identifying the compositions of mixed chemical gases with a mixed ratio in the early stage. This innovative technology comprised an optimized combination of nanodeposited arrayed electrodes and artificial intelligence techniques with advanced sensing capabilities that could operate within biological limits, resulting in the verification of mixed vapor chemical components. Highly selective sensors that are tolerant to high humidity levels provide a target for "breath chemovapor fingerprinting" for the early diagnosis of diseases. The feature selection analysis achieved recognition rates of 99% and above under low-humidity conditions and 98% and above under humid conditions for mixed chemical compositions. The 1D convolutional neural network analysis performed better, discriminating the compositional state of chemical vapor under low-and high-humidity conditions almost perfectly. This study provides a basis for the use of a multiplexed DNA-functionalized graphene gas sensor array and artificial intelligence-based discrimination of chemical vapor compositions in breath analysis applications.

arXiv (Cornell University), Oct 5, 2021

Deep reinforcement learning has shown its effectiveness in various applications, providing a prom... more Deep reinforcement learning has shown its effectiveness in various applications, providing a promising direction for solving tasks with high complexity. However, naively applying classical RL for learning a complex long-horizon task with a single control policy is inefficient. Thus, policy modularization tackles this problem by learning a set of modules that are mapped to primitives and properly orchestrating them. In this study, we further expand the discussion by incorporating simultaneous activation of the skills and structuring them into multiple hierarchies in a recursive fashion. Moreover, we sought to devise an algorithm that can properly orchestrate the skills with different action spaces via multiplicative Gaussian distributions, which highly increases the reusability. By exploiting the modularity, interpretability can also be achieved by observing the modules that are used in the new task if each of the skills is known. We demonstrate how the proposed scheme can be employed in practice by solving a pick and place task with a 6 DoF manipulator, and examine the effects of each property from ablation studies.

IEEE Robotics and Automation Letters

Deep reinforcement learning has shown its effectiveness in various applications, providing a prom... more Deep reinforcement learning has shown its effectiveness in various applications, providing a promising direction for solving tasks with high complexity. However, naively applying classical RL for learning a complex long-horizon task with a single control policy is inefficient. Thus, policy modularization tackles this problem by learning a set of modules that are mapped to primitives and properly orchestrating them. In this study, we further expand the discussion by incorporating simultaneous activation of the skills and structuring them into multiple hierarchies in a recursive fashion. Moreover, we sought to devise an algorithm that can properly orchestrate the skills with different action spaces via multiplicative Gaussian distributions, which highly increases the reusability. By exploiting the modularity, interpretability can also be achieved by observing the modules that are used in the new task if each of the skills is known. We demonstrate how the proposed scheme can be employed in practice by solving a pick and place task with a 6 DoF manipulator, and examine the effects of each property from ablation studies.

Background: To develop a warning system that can prevent or minimize laser exposure resulting in ... more Background: To develop a warning system that can prevent or minimize laser exposure resulting in kidney and ureter damage during retrograde intrarenal surgery (RIRS) for urolithiasis. Our study builds on the hypothesis that shock waves of different degrees are delivered to the hand of the surgeon depending on whether the laser hits the stone or tissue. Methods: A surgical environment was simulated for RIRS by filling the body of a raw whole chicken with water and stones from the human body. We developed an acceleration measurement system that recorded the power signal data for a number of hours, yielding distinguishable characteristics among three different states (idle state, stones, and tissue–laser interface) by conducting fast Fourier transform (FFT) analysis. A discrete wavelet transform (DWT) was used for feature extraction, and a random forest classification algorithm was applied to classify the current state of the laser–tissue interface. Results: The result of the FFT showe...

RSC Advances, 2020

A two-step machine learning (ML) algorithm for coronary artery decision making is introduced, to ... more A two-step machine learning (ML) algorithm for coronary artery decision making is introduced, to increase the data quality by providing flow characteristics and biometric features by aid of computational fluid dynamics (CFD).

and control of a thermal stabilizing system for a MEMS

Journal of Dynamic Systems Measurement and Control-transactions of The Asme, Oct 21, 2014

This tutorial paper presents the expositions of stochastic optimal feedback control theory and Ba... more This tutorial paper presents the expositions of stochastic optimal feedback control theory and Bayesian spatiotemporal models in the context of robotics applications. The presented material is self-contained so that readers can grasp the most important concepts and acquire knowledge needed to jump-start their research. To facilitate this, we provide a series of educational examples from robotics and mobile sensor networks.

Journal of Medical Internet Research, Jan 13, 2023

Background: Osteoporosis is one of the diseases that requires early screening and detection for i... more Background: Osteoporosis is one of the diseases that requires early screening and detection for its management. Common clinical tools and machine-learning (ML) models for screening osteoporosis have been developed, but they show limitations such as low accuracy. Moreover, these methods are confined to limited risk factors and lack individualized explanation. Objective: The aim of this study was to develop an interpretable deep-learning (DL) model for osteoporosis risk screening with clinical features. Clinical interpretation with individual explanations of feature contributions is provided using an explainable artificial intelligence (XAI) technique. Methods: We used two separate data sets: the National Health and Nutrition Examination Survey data sets from the United States (NHANES) and South Korea (KNHANES) with 8274 and 8680 respondents, respectively. The study population was classified according to the T-score of bone mineral density at the femoral neck or total femur. A DL model for osteoporosis diagnosis was trained on the data sets and significant risk factors were investigated with local interpretable model-agnostic explanations (LIME). The performance of the DL model was compared with that of ML models and conventional clinical tools. Additionally, contribution ranking of risk factors and individualized explanation of feature contribution were examined. Results: Our DL model showed area under the curve (AUC) values of 0.851 (95% CI 0.844-0.858) and 0.922 (95% CI 0.916-0.928) for the femoral neck and total femur bone mineral density, respectively, using the NHANES data set. The corresponding AUC values for the KNHANES data set were 0.827 (95% CI 0.821-0.833) and 0.912 (95% CI 0.898-0.927), respectively. Through the LIME method, significant features were induced, and each feature's integrated contribution and interpretation for individual risk were determined. Conclusions: The developed DL model significantly outperforms conventional ML models and clinical tools. Our XAI model produces high-ranked features along with the integrated contributions of each feature, which facilitates the interpretation of individual risk. In summary, our interpretable model for osteoporosis risk screening outperformed state-of-the-art methods.

Journal of Dynamic Systems Measurement and Control-transactions of The Asme, Oct 21, 2014

Explainable deep learning-based clinical decision support engine for MRI-based automated diagnosis of temporomandibular joint anterior disk displacement

Computer Methods and Programs in Biomedicine, May 1, 2023

Image and Vision Computing, Jul 1, 2015

This paper considers visual feature selection to implement position estimation using an omnidirec... more This paper considers visual feature selection to implement position estimation using an omnidirectional camera. The localization is based on a maximum likelihood estimation (MLE) with a map from optimally selected visual features using Gaussian Process (GP) regression. In particular, the collection of selected features over a surveillance region is modeled by a multivariate GP with unknown hyperparameters. The hyperparameters are identified through the learning process by an MLE, which are used for prediction in an empirical Bayes fashion. To select features, we apply a backward sequential elimination technique in order to improve the quality of the position estimation with compressed features for efficient localization. The excellent results of the proposed algorithm are illustrated by the experimental studies with different visual features under both indoor and outdoor real-world scenarios.

IEEE Transactions on Control Systems and Technology, Mar 1, 2015

In this paper, we present a set of techniques for finding a cost function to the time-invariant L... more In this paper, we present a set of techniques for finding a cost function to the time-invariant Linear Quadratic Regulator (LQR) problem in both continuous-and discrete-time cases. Our methodology is based on the solution to the inverse LQR problem, which can be stated as: does a given controller K describe the solution to a time-invariant LQR problem, and if so, what weights Q and R produce K as the optimal solution? Our motivation for investigating this problem is the analysis of motion goals in biological systems. We first describe an efficient Linear Matrix Inequality (LMI) method for determining a solution to the general case of this inverse LQR problem when both the weighting matrices Q and R are unknown. Our first LMI-based formulation provides a unique solution when it is feasible. Additionally, we propose a gradientbased, least-squares minimization method that can be applied to approximate a solution in cases when the LMIs are infeasible. This new method is very useful in practice since the estimated gain matrix K from the noisy experimental data could be perturbed by the estimation error, which may result in the infeasibility of the LMIs. We also provide an LMI minimization problem to find a good initial point for the minimization using the proposed gradient descent algorithm. We then provide a set of examples to illustrate how to apply our approaches to several different types of problems. An important result is the application of the technique to human subject posture control when seated on a moving robot. Results show that we can recover a cost function which may provide a useful insight on the human motor control goal.

A body of literature garners the importance of the intraluminal thrombus layer (ILT) as a key fac... more A body of literature garners the importance of the intraluminal thrombus layer (ILT) as a key factor for the abdominal aortic aneurysm associated with its growth and rupture. There are, however, multiple, often opposite views, on the role of ILT regarding to the pathological, biosolid, and biofluid aspects. Hence, there is a significant need to understand how the ILT is associated with the growth and the rupture and how an aneurysm, with a given patient, predicts its aneurysmal growth. In this work, we use existing patients’ CT data and investigate the association of the ILT in the aneurysm growth rate, with general and localized trends, and its role of biomechanics. We discuss a statistical framework using a computational model of aneurysm growth, which will empower the prediction capability for the clinical management. Method The first set of 39 CT images from 9 patients is used to generate multiple hypotheses for AAA expansion and biomechanics associated with ILT. For this purpos...

Physics Letters A, 2021

Electron scattering cross sections have been acquired both theoretically and experimentally over ... more Electron scattering cross sections have been acquired both theoretically and experimentally over the last few decades. By combining scattering data with machine learning, this work is designed to provide physics benefits: AI assisted incorrect-data screening, cross section data generation, and inverse design. As a basic task before undertaking these applications, we present essential training procedures in this paper. We trained electron-collision data to train a neural network with the type of each collision. The neural network with two hidden layer was implemented using multilayer perceptrons, the earliest deep learning model. Monte Carlo cross-validation was employed to ensure the reliability of the test results, and the optimal model structure was obtained by performing Bayesian optimization. We evaluated our model through the performance indicators within a confidence interval. The results indicate that the data were well-learned, except for the attachment class. Furthermore, feature investigation was carried out to ensure the decision-making process.

Journal of Biomechanics, 2020

Performance during seated balancing is often used to assess trunk neuromuscular control, includin... more Performance during seated balancing is often used to assess trunk neuromuscular control, including evaluating impairments in back pain populations. Balancing in less challenging environments allows for flexibility in control, which may not depend on health status but instead may reflect personal preferences. To make assessment less ambiguous, trunk neuromuscular control should be maximally challenged. Thirty-four healthy subjects balanced on a robotic seat capable of adjusting rotational stiffness. Subjects balanced while rotational stiffness was gradually reduced. The rotational stiffness at which subjects could no longer maintain balance, defined as critical stiffness (k Crit), was used to quantify the subjects' trunk neuromuscular control. A higher k Crit reflects poorer control, as subjects require a more stable base to balance. Subjects were tested on three days separated by 24 hours to assess test-retest reliability. Anthropometric (height and

IEEE Transactions on Industrial Informatics, 2019

An approximate closed-form formula for calculating the ohmic resistance of a circular multiloop c... more An approximate closed-form formula for calculating the ohmic resistance of a circular multiloop coil with unequal pitches is presented. Skin effect and proximity effect are included in the formula. The proximity effect is expressed as a proximity factor obtained using transverse magnetic fields applied to a wire from the rest of the wires. For verification, the optimum dimension for minimum resistance of wires with an equal pitch is compared with the previous results, and both results agree. The formula is applied to calculate the ohmic resistance of helical and spiral coils and is verified by a 2-D finite-element-method simulation. Both calculation and simulation results are consistent as well. As a practical application, a spiral coil with unequal pitches is designed for uniform mutual inductance, and it is optimized for the lowest resistance using the formula. The measured ohmic resistance of the designed coil also agrees with the calculated and simulated results. The results show that the formula can be well applied to designing circular multiloop coils with minimum ohmic loss in wireless-power-transfer systems.

IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2019

To study complex neuromuscular control pathways in human movement, biomechanical parametric model... more To study complex neuromuscular control pathways in human movement, biomechanical parametric models and system identification methods are employed. Although test-retest reliability is widely used to validate outcomes of motor control tasks, it was not incorporated in system identification methods. This study investigates the feasibility of incorporating test-retest reliability in our previously published method of selecting sensitive parameters. We consider the selected parameters via this novel approach to be the key neuromuscular parameters because they meet three criteria: reduced variability, improved goodness of fit, and excellent reliability. These criteria ensure that parameter variability is below a user-defined value, the number of these parameters is maximized to enhance goodness of fit, and their test-retest reliability is above a user-defined value. We measured variability, goodness of fit, and reliability using Fisher information matrix, variance accounted for, and intraclass correlation, respectively. We also incorporated model diversity as a fourth optional criterion to narrow down the solution space of key parameters. We applied this approach to head position tracking in axial rotation and flexion/extension. Forty healthy subjects performed the tasks during two visits. With variability and reliability measures ≤0.35 and ≥0.75 respectively, we selected three key parameters out of twelve with goodness of fit >69%. The key

IEEE transactions on bio-medical engineering, Jan 2, 2018

An Abdominal Aortic Aneurysm (AAA) is a form of vascular disease causing focal enlargement of the... more An Abdominal Aortic Aneurysm (AAA) is a form of vascular disease causing focal enlargement of the abdominal aorta. It affects a large part of the population, and in case of rupture, has up to 90% mortality rate. Recent clinical recom- mendations suggest that people with small aneurysms should be examined 3-36 months depending on the size, to monitor morphological changes. While advances in biomechanics provide state-of-the-art spatial estimates of stress distributions of AAAs, there are still limitations in modeling its time evolution and uncertainty qualification. To date, there are a few biomechanical frameworks that utilize longitudinal medical images, which would aid physicians in detecting small aneurysms with high risk of rupture. In this study, we use longitudinal computer tomography (CT) scans of AAAs that are captured at different times to predict the spatio-temporal evolution of AAAs' shape in future time. We consider a surface of 3D AAA as a manifold embedded in a sca...