Skip to main content

Jinah Park

Korea Advanced Institute of Science and Technology, School of Computing, Faculty Member

Followers

6

Following

2

Public Views

Full Professor in the School of Computing, KAIST ICT Endowed Chair Professor

less

Daniel D. Hutto

University of Wollongong

Harvard University

Canadian University of Dubai

University of California, Merced

Louis de Saussure

University of Neuchâtel

Armando Marques-Guedes

UNL - New University of Lisbon

University of Exeter

Stefan Kristensen

Université de Strasbourg

Case Western Reserve University

Universidade Federal do Rio Grande do Sul

Interests

Uploads

Papers by Jinah Park

WG-VITON: Wearing-Guide Virtual Try-On for Top and Bottom Clothes

arXiv (Cornell University), May 10, 2022

Studies of virtual try-on (VITON) have been shown their effectiveness in utilizing the generative... more Studies of virtual try-on (VITON) have been shown their effectiveness in utilizing the generative neural network for virtually exploring fashion products, and some of recent researches of VITON attempted to synthesize human image wearing given multiple types of garments (e.g., top and bottom clothes). However, when replacing the top and bottom clothes of the target human, numerous wearing styles are possible with a certain combination of the clothes. In this paper, we address the problem of variation in wearing style when simultaneously replacing the top and bottom clothes of the model. We introduce Wearing-Guide VITON (i.e., WG-VITON) which utilizes an additional input binary mask to control the wearing styles of the generated image. Our experiments show that WG-VITON effectively generates an image of the model wearing given top and bottom clothes, and create complicated wearing styles such as partly tucking in the top to the bottom.

<title>Interactive visualization of fuzzy set operations</title>

Visualization and Data Analysis 2010, 2010

Fuzzy sets refer to the datasets which do not have separate, distinct clusters, and they contain ... more Fuzzy sets refer to the datasets which do not have separate, distinct clusters, and they contain data elements whose membership values are between 0 and 1. Often each element belongs to several groups. We design a disk diagram with an augmented frequency graph to visualize elements in a fuzzy set. It shows the distribution of the elements within the set with respect to the membership function value. Fuzzy set operations-like intersection-are used to find the elements of interests with certain degree of confidence. However, few techniques exist for visualization of fuzzy set operations whose process and result provide both the relationship and the distribution of individual data point between sets. With the proposed disk diagram representing a fuzzy set, we suggest an interactive visualization system to analyze fuzzy data and grasp the relationship among the sets. By interactively overlapping the disk diagrams, a user can not only bring out the individual elements of interests, but also overview the whole data elements associated with the sets. Furthermore, we investigated two different visualization scenarios for intuitive interpretation of the visualization results. We demonstrate the proposed visualization system with a terrorist analysis output dataset.

Model-based Analysis of Cardiac Motion from

We develop a new method for analyzing the motion of the left ventricle (LV) of a heart from tagge... more We develop a new method for analyzing the motion of the left ventricle (LV) of a heart from tagged MRI data. Our technique is based on the development of a new class of physics-based deformable models whose parameters are functions allowing the definition of new parameterized primitives and parameterized deformations. These parameter functions improve the accuracy of shape description through the use of a few intuitive parameters such as functional twisting. Furthermore, these parameters require no complex postprocessing in order to be used by a physician. Using a physics-based approach, we convert these geometric models into deformable models that deform due to forces exerted from the datapoints and conform to the given dataset. We present experiments involving the extraction of shape and motion of the LV from MRI-SPAMM data based on a few parameter functions. Furthermore, by plotting the variations over time of the extracted model parameters from normal and abnormal heart data we are able to characterize quantitatively their differences. Comments

Localizing Human Keypoints beyond the Bounding Box

2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), 2021

Since human pose is one of the most effective and popular sources for understanding human in vari... more Since human pose is one of the most effective and popular sources for understanding human in various applications, there have been numerous researches on detecting keypoints of human body from the image source. However, when a human body is shown partially in the source image, estimation range is also restricted causing performance degradation in locating keypoints of human body. In this paper, we propose 'Position Puzzle' network and augmentation to leverage the performance of detecting keypoints including those outside the bounding box. Specifically, Position Puzzle Network expands the spatial range of keypoint localization by refining the position and the scale of the target's bounding box, and Position Puzzle Augmentation improves the performance of keypoint detector using the partial image in training. We prepare data by cropping COCO dataset and utilize them in training and evaluation. Under the prepared dataset, the proposed method enhances the performance of baseline network up to 37.6% and 30.6% in mAP and mAR, respectively, and effectively localizes keypoints positioned not only inside but also outside the bounding box. We also verify that the proposed method can localize keypoints beyond the bounding box in the original COCO dataset.

A Study on Virtual Assembly Simulation Using Virtual Reality Technology

Real-time soft material volume deformation using shape retaining chain linked model

International Conference on …, 2002

... In addition, since the haptic update rate (1kHz) is faster than the graphic update rate (30Hz... more

Visualization and Data Analysis 2008

The papers included in this volume were part of the technical conference cited on the cover and t... more The papers included in this volume were part of the technical conference cited on the cover and title page. Papers were selected and subject to review by the editors and conference program committee. Some conference presentations may not be available for publication. The papers published in these proceedings reflect the work and thoughts of the authors and are published herein as submitted. The publishers are not responsible for the validity of the information or for any outcomes resulting from reliance thereon.

Real-Time Area-Based Haptic Rendering for a Palpation Simulator

Lecture Notes in Computer Science, 2006

In palpation procedure, medical doctors push and rub the organ's surface and they are provided th... more In palpation procedure, medical doctors push and rub the organ's surface and they are provided the sensation of distributed pressure and contact force (reflecting force) for discerning doubtable portion. This paper suggests a real-time area-based haptic rendering model to describe distributed pressure and contact force simultaneously and present a haptic interface system to generate surface property in accordance with the haptic rendering algorithm. We represent the haptic model using the shape-retaining chain link (or S-Chain) framework for a fast and stable computation of the contact force and distributed pressure from a volumetric virtual object. In addition, we developed a compact pin-array type of tactile display unit and attached it to PHANToM TM haptic device to complement each other. In order to evaluate the performance of the proposed scheme, related experiments have been conducted with non-homogenous volumetric cubic objects consisting of approximately 500,000 volume elements at a haptic update rate of 1000 Hz. The experimental results show that compared to the point-contact the area-contact provides the users with more precise perception of the shape and softness of the object's composition, and that our proposed system satisfies the real-time and realism constraints to be useful for virtual reality applications.

Automatic Hippocampal Multimodal Assessment for Studies of Stroke and Small Vessel Disease

We propose a framework for assessing the hippocampi on stroke patients and studies of small vesse... more We propose a framework for assessing the hippocampi on stroke patients and studies of small vessel disease, where sclerosis, perivascular spaces and infarcts on this structure are common. It includes hippocampal and cavity segmentations, hippocampal shape modelling, feature characterisation and statistical analyses, all which have been particularly developed for assessing extreme abnormalities on this small brain structure on clinical MRI datasets. The hippocampal segmentation uses FSL tools. We apply this approach to 48 datasets from a study of mild stroke and assess its relevance on a larger dataset of 189 stroke patients. Using shape similarity metrics we show that the hippocampal shape models generated by our method are accurate on this dataset. We estimate the prevalence and distribution of the features analysed on the samples to discuss the usefulness of our approach.

A SPH-based dissolution behavior model for real-time fluid-solid interaction

Interactions among fluids and solids are frequently observed around our daily life and simulated ... more Interactions among fluids and solids are frequently observed around our daily life and simulated for various purposes. However, real-time simulation of dissolution phenomena, including erosion and liquefying, is restrictively allowed due to the complexity of the behavior model. To lessen this limitation, we suggest a particle-based dissolution model for real-time interaction between the fluids and the soluble solid objects. In our system, both of the fluids and the solid objects consist of the particles based on the unified particle model. Dissolution occurs when the solid object is submerged into the fluid. The concentration of the solid particle is transferred to the adjacent fluid particles during the dissolution. In this process, the total amount of the solute is preserved. Completely dissolved solid particles are detached from the object, and the inertia tensor of the object is renewed to reflect the shape changes. With the updated properties, the system solves rigid body dynam...

BiopathwayBuilder: Nested 3D Visualization System for Complex Molecular Interactions

Genome Informatics, 2002

In order to gain a full understanding of a biological process, we must be able to augment the kno... more In order to gain a full understanding of a biological process, we must be able to augment the known molecular interactions with discovered knowledge. We believe that a visualization system works as a means for accomplishing this task, as it provides an intuitive base for necessary information, among others. However, reported implementations have further problems: (1) The size of the information is not only enormous, but also grows very fast, which makes scalability and elision essential properties [5]; (2) the available information is not only incomplete, but also unreliable; and (3) the usual information in the field, such as protein modification [2], is inherently complex, which makes it very difficult to make the resulting visualization intuitive enough for end users as well as field experts. We address all the problems above with a 3D visualization system.

Domain-Robust Mitotic Figure Detection with Style Transfer

ArXiv, 2021

We propose a new training scheme for domain generalization in mitotic figure detection. Mitotic f... more We propose a new training scheme for domain generalization in mitotic figure detection. Mitotic figures show different characteristics for each scanner. We consider each scanner as a ’domain’ and the image distribution specified for each domain as ’style’. The goal is to train our network to be robust on scanner types by using various ’style’ images. To expand the style variance, we transfer a style of the training image into arbitrary styles, by defining a module based on StarGAN. Our model with the proposed training scheme shows positive performance on MIDOG Preliminary Test-Set containing scanners never seen before.

Domain-Robust Mitotic Figure Detection with StyleGAN

We propose a new training scheme for domain generalization in mitotic figure detection. By consid... more We propose a new training scheme for domain generalization in mitotic figure detection. By considering the image variance due to different scanner types as different image styles, we have trained our detection network to be robust on scanner types. To expand the image variance, domain of training image is transferred into arbitrary domain. The proposed style transfer module generates different styled images from an input image with random code, eventually generating variously styled images. Our model with the proposed training scheme shows good performance on MIDOG Preliminary Test-Set containing scanners never seen before.

Deep Regression Network-Assisted Efficient Streamline Generation Method

IEEE Access

Streamlining is one of the most frequently utilized visualization methods to analyze the flow str... more Streamlining is one of the most frequently utilized visualization methods to analyze the flow structure of computational fluid dynamics (CFD) data. However, it is challenging to find a set of streamlines showing the most prominent flow across the entire flow field due to the heavy computation time required to generate bundles of streamlines. In this paper, we propose an efficient streamline generation method that removes several seed candidates that are predicted as less important using a 3D U-net based regression model. We employ 3D line integral convolution (LIC) volumes that depict the entire flow field for training data of the proposed learning model and evaluate our method using a real-world CFD data set. We find using our model that we can obtain quality of visualization results comparable to that of the ground truth even when more than 90% of the seed candidates are truncated while operating 6.6∼17.1 times faster than the competing method. INDEX TERMS Deep regression network, dilated convolutional neural network, flow visualization, line integral convolution, streamline.

Lipid Droplet Tracking Method using Optical Flow in 3D Cell Image Data

KIISE Transactions on Computing Practices

Organ Shape Modeling Based on the Laplacian Deformation Framework for Surface-Based Morphometry Studies

Journal of Computing Science and Engineering

Recently, shape analysis of human organs has achieved much attention, owing to its potential to l... more Recently, shape analysis of human organs has achieved much attention, owing to its potential to localize structural abnormalities. For a group-wise shape analysis, it is important to accurately restore the shape of a target structure in each subject and to build the inter-subject shape correspondences. To accomplish this, we propose a shape modeling method based on the Laplacian deformation framework. We deform a template model of a target structure in the segmented images while restoring subject-specific shape features by using Laplacian surface representation. In order to build the inter-subject shape correspondences, we implemented the progressive weighting scheme for adaptively controlling the rigidity parameter of the deformable model. This weighting scheme helps to preserve the relative distance between each point in the template model as much as possible during model deformation. This area-preserving deformation allows each point of the template model to be located at an anatomically consistent position in the target structure. Another advantage of our method is its application to human organs of non-spherical topology. We present the experiments for evaluating the robustness of shape modeling against large variations in shape and size with the synthetic sets of the second cervical vertebrae (C2), which has a complex shape with holes.

Real-time Simulation Technique for Visual-Haptic Interaction between SPH-based Fluid Media and Soluble Solids

Journal of the Korean Society of Visualization

Interaction between fluid and a rigid object is frequently observed in everyday life. However, it... more Interaction between fluid and a rigid object is frequently observed in everyday life. However, it is difficult to simulate their interaction as the medium and the object have different representations. One of the challenging issues arises especially in handling deformation of the object visually as well as rendering haptic feedback. In this paper, we propose a real-time simulation technique for multimodal interaction between particle-based fluids and soluble solids. We have developed the dissolution behavior model of solids, which is discretized based on the idea of smoothed particle hydrodynamics, and the changes in physical properties accompanying dissolution is immediately reflected to the object. The user is allowed to intervene in the simulation environment anytime by manipulating the solid object, where both visual and haptic feedback are delivered to the user on the fly. For immersive visualization, we also adopt the screen space fluid rendering technique which can balance realism and performance.

Hippocampal morphology and cognitive functions in community-dwelling older people: the Lothian Birth Cohort 1936

Neurobiology of aging, Apr 1, 2017

Structural measures of the hippocampus have been linked to a variety of memory processes and also... more Structural measures of the hippocampus have been linked to a variety of memory processes and also to broader cognitive abilities. Gross volumetry has been widely used, yet the hippocampus has a complex formation, comprising distinct subfields which may be differentially sensitive to the deleterious effects of age, and to different aspects of cognitive performance. However, a comprehensive analysis of multidomain cognitive associations with hippocampal deformations among a large group of cognitively normal older adults is currently lacking. In 654 participants of the Lothian Birth Cohort 1936 (mean age = 72.5, SD = 0.71 years), we examined associations between the morphology of the hippocampus and a variety of memory tests (spatial span, letter-number sequencing, verbal recall, and digit backwards), as well as broader cognitive domains (latent measures of speed, fluid intelligence, and memory). Following correction for age, sex, and vascular risk factors, analysis of memory subtests ...

Associations between hippocampal morphology, diffusion characteristics, and salivary cortisol in older men

Psychoneuroendocrinology, Apr 27, 2017

High, unabated glucocorticoid (GC) levels are thought to selectively damage certain tissue types.... more High, unabated glucocorticoid (GC) levels are thought to selectively damage certain tissue types. The hippocampus is thought to be particularly susceptible to such effects, and though findings from animal models and human patients provide some support for this hypothesis, evidence for associations between elevated GCs and lower hippocampal volumes in older age (when GC levels are at greater risk of dysregulation) is inconclusive. To address the possibility that the effects of GCs in non-pathological ageing may be too subtle for gross volumetry to reliably detect, we analyse associations between salivary cortisol (diurnal and reactive measures), hippocampal morphology and diffusion characteristics in 88 males, aged ∼73 years. However, our results provide only weak support for this hypothesis. Though nominally significant peaks in morphology were found in both hippocampi across all salivary cortisol measures (standardised β magnitudes<0.518, puncorrected>0.0000003), associations...

Morphological and Microstructural Changes of the Hippocampus in Early MCI: A Study Utilizing the Alzheimer's Disease Neuroimaging Initiative Database

Journal of clinical neurology (Seoul, Korea), Jan 25, 2017

With the aim of facilitating the early detection of Alzheimer's disease, the Alzheimer's ... more With the aim of facilitating the early detection of Alzheimer's disease, the Alzheimer's Disease Neuroimaging Initiative proposed two stages based on the memory performance: early mild cognitive impairment (EMCI) and late mild cognitive impairment (LMCI). The current study was designed to investigate structural differences in terms of surface atrophy and microstructural changes of the hippocampus in EMCI and LMCI. Hippocampal shape modeling based on progressive template surface deformation was performed on T1-weighted MRI images obtained from 20 cognitive normal (CN) subjects, 17 EMCI patients, and 20 LMCI patients. A template surface in CN was used as a region of interest for diffusion-tensor imaging (DTI) voxel-based morphometry (VBM) analysis. Cluster-wise group comparison was performed based on DTI indices within the hippocampus. Linear regression was performed to identify correlations between DTI metrics and clinical scores. The hippocampal surface analysis showed signi...

WG-VITON: Wearing-Guide Virtual Try-On for Top and Bottom Clothes

arXiv (Cornell University), May 10, 2022

Studies of virtual try-on (VITON) have been shown their effectiveness in utilizing the generative... more Studies of virtual try-on (VITON) have been shown their effectiveness in utilizing the generative neural network for virtually exploring fashion products, and some of recent researches of VITON attempted to synthesize human image wearing given multiple types of garments (e.g., top and bottom clothes). However, when replacing the top and bottom clothes of the target human, numerous wearing styles are possible with a certain combination of the clothes. In this paper, we address the problem of variation in wearing style when simultaneously replacing the top and bottom clothes of the model. We introduce Wearing-Guide VITON (i.e., WG-VITON) which utilizes an additional input binary mask to control the wearing styles of the generated image. Our experiments show that WG-VITON effectively generates an image of the model wearing given top and bottom clothes, and create complicated wearing styles such as partly tucking in the top to the bottom.

<title>Interactive visualization of fuzzy set operations</title>

Visualization and Data Analysis 2010, 2010

Fuzzy sets refer to the datasets which do not have separate, distinct clusters, and they contain ... more Fuzzy sets refer to the datasets which do not have separate, distinct clusters, and they contain data elements whose membership values are between 0 and 1. Often each element belongs to several groups. We design a disk diagram with an augmented frequency graph to visualize elements in a fuzzy set. It shows the distribution of the elements within the set with respect to the membership function value. Fuzzy set operations-like intersection-are used to find the elements of interests with certain degree of confidence. However, few techniques exist for visualization of fuzzy set operations whose process and result provide both the relationship and the distribution of individual data point between sets. With the proposed disk diagram representing a fuzzy set, we suggest an interactive visualization system to analyze fuzzy data and grasp the relationship among the sets. By interactively overlapping the disk diagrams, a user can not only bring out the individual elements of interests, but also overview the whole data elements associated with the sets. Furthermore, we investigated two different visualization scenarios for intuitive interpretation of the visualization results. We demonstrate the proposed visualization system with a terrorist analysis output dataset.

Model-based Analysis of Cardiac Motion from

We develop a new method for analyzing the motion of the left ventricle (LV) of a heart from tagge... more We develop a new method for analyzing the motion of the left ventricle (LV) of a heart from tagged MRI data. Our technique is based on the development of a new class of physics-based deformable models whose parameters are functions allowing the definition of new parameterized primitives and parameterized deformations. These parameter functions improve the accuracy of shape description through the use of a few intuitive parameters such as functional twisting. Furthermore, these parameters require no complex postprocessing in order to be used by a physician. Using a physics-based approach, we convert these geometric models into deformable models that deform due to forces exerted from the datapoints and conform to the given dataset. We present experiments involving the extraction of shape and motion of the LV from MRI-SPAMM data based on a few parameter functions. Furthermore, by plotting the variations over time of the extracted model parameters from normal and abnormal heart data we are able to characterize quantitatively their differences. Comments

Localizing Human Keypoints beyond the Bounding Box

2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), 2021

Since human pose is one of the most effective and popular sources for understanding human in vari... more Since human pose is one of the most effective and popular sources for understanding human in various applications, there have been numerous researches on detecting keypoints of human body from the image source. However, when a human body is shown partially in the source image, estimation range is also restricted causing performance degradation in locating keypoints of human body. In this paper, we propose 'Position Puzzle' network and augmentation to leverage the performance of detecting keypoints including those outside the bounding box. Specifically, Position Puzzle Network expands the spatial range of keypoint localization by refining the position and the scale of the target's bounding box, and Position Puzzle Augmentation improves the performance of keypoint detector using the partial image in training. We prepare data by cropping COCO dataset and utilize them in training and evaluation. Under the prepared dataset, the proposed method enhances the performance of baseline network up to 37.6% and 30.6% in mAP and mAR, respectively, and effectively localizes keypoints positioned not only inside but also outside the bounding box. We also verify that the proposed method can localize keypoints beyond the bounding box in the original COCO dataset.

A Study on Virtual Assembly Simulation Using Virtual Reality Technology

Real-time soft material volume deformation using shape retaining chain linked model

International Conference on …, 2002

... In addition, since the haptic update rate (1kHz) is faster than the graphic update rate (30Hz... more

Visualization and Data Analysis 2008

The papers included in this volume were part of the technical conference cited on the cover and t... more The papers included in this volume were part of the technical conference cited on the cover and title page. Papers were selected and subject to review by the editors and conference program committee. Some conference presentations may not be available for publication. The papers published in these proceedings reflect the work and thoughts of the authors and are published herein as submitted. The publishers are not responsible for the validity of the information or for any outcomes resulting from reliance thereon.

Real-Time Area-Based Haptic Rendering for a Palpation Simulator

Lecture Notes in Computer Science, 2006

In palpation procedure, medical doctors push and rub the organ's surface and they are provided th... more In palpation procedure, medical doctors push and rub the organ's surface and they are provided the sensation of distributed pressure and contact force (reflecting force) for discerning doubtable portion. This paper suggests a real-time area-based haptic rendering model to describe distributed pressure and contact force simultaneously and present a haptic interface system to generate surface property in accordance with the haptic rendering algorithm. We represent the haptic model using the shape-retaining chain link (or S-Chain) framework for a fast and stable computation of the contact force and distributed pressure from a volumetric virtual object. In addition, we developed a compact pin-array type of tactile display unit and attached it to PHANToM TM haptic device to complement each other. In order to evaluate the performance of the proposed scheme, related experiments have been conducted with non-homogenous volumetric cubic objects consisting of approximately 500,000 volume elements at a haptic update rate of 1000 Hz. The experimental results show that compared to the point-contact the area-contact provides the users with more precise perception of the shape and softness of the object's composition, and that our proposed system satisfies the real-time and realism constraints to be useful for virtual reality applications.

Automatic Hippocampal Multimodal Assessment for Studies of Stroke and Small Vessel Disease

We propose a framework for assessing the hippocampi on stroke patients and studies of small vesse... more We propose a framework for assessing the hippocampi on stroke patients and studies of small vessel disease, where sclerosis, perivascular spaces and infarcts on this structure are common. It includes hippocampal and cavity segmentations, hippocampal shape modelling, feature characterisation and statistical analyses, all which have been particularly developed for assessing extreme abnormalities on this small brain structure on clinical MRI datasets. The hippocampal segmentation uses FSL tools. We apply this approach to 48 datasets from a study of mild stroke and assess its relevance on a larger dataset of 189 stroke patients. Using shape similarity metrics we show that the hippocampal shape models generated by our method are accurate on this dataset. We estimate the prevalence and distribution of the features analysed on the samples to discuss the usefulness of our approach.

A SPH-based dissolution behavior model for real-time fluid-solid interaction

Interactions among fluids and solids are frequently observed around our daily life and simulated ... more Interactions among fluids and solids are frequently observed around our daily life and simulated for various purposes. However, real-time simulation of dissolution phenomena, including erosion and liquefying, is restrictively allowed due to the complexity of the behavior model. To lessen this limitation, we suggest a particle-based dissolution model for real-time interaction between the fluids and the soluble solid objects. In our system, both of the fluids and the solid objects consist of the particles based on the unified particle model. Dissolution occurs when the solid object is submerged into the fluid. The concentration of the solid particle is transferred to the adjacent fluid particles during the dissolution. In this process, the total amount of the solute is preserved. Completely dissolved solid particles are detached from the object, and the inertia tensor of the object is renewed to reflect the shape changes. With the updated properties, the system solves rigid body dynam...

BiopathwayBuilder: Nested 3D Visualization System for Complex Molecular Interactions

Genome Informatics, 2002

In order to gain a full understanding of a biological process, we must be able to augment the kno... more In order to gain a full understanding of a biological process, we must be able to augment the known molecular interactions with discovered knowledge. We believe that a visualization system works as a means for accomplishing this task, as it provides an intuitive base for necessary information, among others. However, reported implementations have further problems: (1) The size of the information is not only enormous, but also grows very fast, which makes scalability and elision essential properties [5]; (2) the available information is not only incomplete, but also unreliable; and (3) the usual information in the field, such as protein modification [2], is inherently complex, which makes it very difficult to make the resulting visualization intuitive enough for end users as well as field experts. We address all the problems above with a 3D visualization system.

Domain-Robust Mitotic Figure Detection with Style Transfer

ArXiv, 2021

We propose a new training scheme for domain generalization in mitotic figure detection. Mitotic f... more We propose a new training scheme for domain generalization in mitotic figure detection. Mitotic figures show different characteristics for each scanner. We consider each scanner as a ’domain’ and the image distribution specified for each domain as ’style’. The goal is to train our network to be robust on scanner types by using various ’style’ images. To expand the style variance, we transfer a style of the training image into arbitrary styles, by defining a module based on StarGAN. Our model with the proposed training scheme shows positive performance on MIDOG Preliminary Test-Set containing scanners never seen before.

Domain-Robust Mitotic Figure Detection with StyleGAN

We propose a new training scheme for domain generalization in mitotic figure detection. By consid... more We propose a new training scheme for domain generalization in mitotic figure detection. By considering the image variance due to different scanner types as different image styles, we have trained our detection network to be robust on scanner types. To expand the image variance, domain of training image is transferred into arbitrary domain. The proposed style transfer module generates different styled images from an input image with random code, eventually generating variously styled images. Our model with the proposed training scheme shows good performance on MIDOG Preliminary Test-Set containing scanners never seen before.

Deep Regression Network-Assisted Efficient Streamline Generation Method

IEEE Access

Streamlining is one of the most frequently utilized visualization methods to analyze the flow str... more Streamlining is one of the most frequently utilized visualization methods to analyze the flow structure of computational fluid dynamics (CFD) data. However, it is challenging to find a set of streamlines showing the most prominent flow across the entire flow field due to the heavy computation time required to generate bundles of streamlines. In this paper, we propose an efficient streamline generation method that removes several seed candidates that are predicted as less important using a 3D U-net based regression model. We employ 3D line integral convolution (LIC) volumes that depict the entire flow field for training data of the proposed learning model and evaluate our method using a real-world CFD data set. We find using our model that we can obtain quality of visualization results comparable to that of the ground truth even when more than 90% of the seed candidates are truncated while operating 6.6∼17.1 times faster than the competing method. INDEX TERMS Deep regression network, dilated convolutional neural network, flow visualization, line integral convolution, streamline.

Lipid Droplet Tracking Method using Optical Flow in 3D Cell Image Data

KIISE Transactions on Computing Practices

Organ Shape Modeling Based on the Laplacian Deformation Framework for Surface-Based Morphometry Studies

Journal of Computing Science and Engineering

Recently, shape analysis of human organs has achieved much attention, owing to its potential to l... more Recently, shape analysis of human organs has achieved much attention, owing to its potential to localize structural abnormalities. For a group-wise shape analysis, it is important to accurately restore the shape of a target structure in each subject and to build the inter-subject shape correspondences. To accomplish this, we propose a shape modeling method based on the Laplacian deformation framework. We deform a template model of a target structure in the segmented images while restoring subject-specific shape features by using Laplacian surface representation. In order to build the inter-subject shape correspondences, we implemented the progressive weighting scheme for adaptively controlling the rigidity parameter of the deformable model. This weighting scheme helps to preserve the relative distance between each point in the template model as much as possible during model deformation. This area-preserving deformation allows each point of the template model to be located at an anatomically consistent position in the target structure. Another advantage of our method is its application to human organs of non-spherical topology. We present the experiments for evaluating the robustness of shape modeling against large variations in shape and size with the synthetic sets of the second cervical vertebrae (C2), which has a complex shape with holes.

Real-time Simulation Technique for Visual-Haptic Interaction between SPH-based Fluid Media and Soluble Solids

Journal of the Korean Society of Visualization

Interaction between fluid and a rigid object is frequently observed in everyday life. However, it... more Interaction between fluid and a rigid object is frequently observed in everyday life. However, it is difficult to simulate their interaction as the medium and the object have different representations. One of the challenging issues arises especially in handling deformation of the object visually as well as rendering haptic feedback. In this paper, we propose a real-time simulation technique for multimodal interaction between particle-based fluids and soluble solids. We have developed the dissolution behavior model of solids, which is discretized based on the idea of smoothed particle hydrodynamics, and the changes in physical properties accompanying dissolution is immediately reflected to the object. The user is allowed to intervene in the simulation environment anytime by manipulating the solid object, where both visual and haptic feedback are delivered to the user on the fly. For immersive visualization, we also adopt the screen space fluid rendering technique which can balance realism and performance.

Hippocampal morphology and cognitive functions in community-dwelling older people: the Lothian Birth Cohort 1936

Neurobiology of aging, Apr 1, 2017

Structural measures of the hippocampus have been linked to a variety of memory processes and also... more Structural measures of the hippocampus have been linked to a variety of memory processes and also to broader cognitive abilities. Gross volumetry has been widely used, yet the hippocampus has a complex formation, comprising distinct subfields which may be differentially sensitive to the deleterious effects of age, and to different aspects of cognitive performance. However, a comprehensive analysis of multidomain cognitive associations with hippocampal deformations among a large group of cognitively normal older adults is currently lacking. In 654 participants of the Lothian Birth Cohort 1936 (mean age = 72.5, SD = 0.71 years), we examined associations between the morphology of the hippocampus and a variety of memory tests (spatial span, letter-number sequencing, verbal recall, and digit backwards), as well as broader cognitive domains (latent measures of speed, fluid intelligence, and memory). Following correction for age, sex, and vascular risk factors, analysis of memory subtests ...

Associations between hippocampal morphology, diffusion characteristics, and salivary cortisol in older men

Psychoneuroendocrinology, Apr 27, 2017

High, unabated glucocorticoid (GC) levels are thought to selectively damage certain tissue types.... more High, unabated glucocorticoid (GC) levels are thought to selectively damage certain tissue types. The hippocampus is thought to be particularly susceptible to such effects, and though findings from animal models and human patients provide some support for this hypothesis, evidence for associations between elevated GCs and lower hippocampal volumes in older age (when GC levels are at greater risk of dysregulation) is inconclusive. To address the possibility that the effects of GCs in non-pathological ageing may be too subtle for gross volumetry to reliably detect, we analyse associations between salivary cortisol (diurnal and reactive measures), hippocampal morphology and diffusion characteristics in 88 males, aged ∼73 years. However, our results provide only weak support for this hypothesis. Though nominally significant peaks in morphology were found in both hippocampi across all salivary cortisol measures (standardised β magnitudes<0.518, puncorrected>0.0000003), associations...

Morphological and Microstructural Changes of the Hippocampus in Early MCI: A Study Utilizing the Alzheimer's Disease Neuroimaging Initiative Database

Journal of clinical neurology (Seoul, Korea), Jan 25, 2017

With the aim of facilitating the early detection of Alzheimer's disease, the Alzheimer's ... more With the aim of facilitating the early detection of Alzheimer's disease, the Alzheimer's Disease Neuroimaging Initiative proposed two stages based on the memory performance: early mild cognitive impairment (EMCI) and late mild cognitive impairment (LMCI). The current study was designed to investigate structural differences in terms of surface atrophy and microstructural changes of the hippocampus in EMCI and LMCI. Hippocampal shape modeling based on progressive template surface deformation was performed on T1-weighted MRI images obtained from 20 cognitive normal (CN) subjects, 17 EMCI patients, and 20 LMCI patients. A template surface in CN was used as a region of interest for diffusion-tensor imaging (DTI) voxel-based morphometry (VBM) analysis. Cluster-wise group comparison was performed based on DTI indices within the hippocampus. Linear regression was performed to identify correlations between DTI metrics and clinical scores. The hippocampal surface analysis showed signi...