MCMigyung Cho

Migyung Cho

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Migyung Cho has a total of 45 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Data-Driven Surrogate Pretraining and Policy Transfer for Rotary Inverted Pendulum Control: A Gazebo-Based Study

    by Gyeongdong Lee, Junhyung Kim, Migyung Cho, Jaesool Shim

    Learning control policies for underactuated systems can require extensive interaction with a physics simulator. This paper presents a unified surrogate-to-simulator policy-initialization framework for rotary inverted pendulum control. A mixed dataset of policy-driven and structured excitation signals, comprising more than 2.5 million transitions from 3000 Gazebo episodes, was used to train an action-conditioned long short-term memory (LSTM) dynamics surrogate. The learned four-state predictor achieved R2 values of 0.98, 0.96, 0.95, and 0.97 for motor angle, motor angular velocity, pendulum angle, and pendulum angular velocity, respectively, under the reported evaluation protocol. The surrogate was embedded as an interactive Soft Actor–Critic (SAC) environment for actor and value-estimation pretraining, followed by network transfer and fine-tuning in Gazebo. A quadratic regulation reward with a 10-point step reference yields a theoretical episodic upper bound of 10,000 over 1000 steps. The recorded Gazebo learning histories show earlier high-return episodes after surrogate pretraining than during training from scratch. The central contribution is an integrated controller-development method that links sequence-based system identification, learned-environment policy optimization, and target-simulator adaptation.

  • Open Access

    Lithium-Ion Battery Life Prediction Using Deep Transfer Learning

    by Wen Zhang, R. S. B. Pranav, Rui Wang, Cheonghwan Lee, Jie Zeng, Migyung Cho, Jaesool Shim

    Lithium-ion batteries are critical components of various advanced devices, including electric vehicles, drones, and medical equipment. However, their performance degrades over time, and unexpected failures or discharges can lead to abrupt operational interruptions. Therefore, accurate prediction of the remaining useful life is essential to ensure device safety and reliability. Conventional RUL prediction methods typically rely on regression analysis, signal processing, and machine learning techniques to assess battery conditions such as charge/discharge cycles, voltage, temperature, and durability. Although effective, these approaches are constrained by their dependence on large amounts of labeled data and the necessity for complex feature engineering to capture battery physical characteristics. In this study, we propose an approach that employs deep transfer learning to address these limitations. By leveraging pretrained model weights, the proposed method significantly improves the efficiency and accuracy of RUL prediction even under limited training data conditions. Furthermore, we investigate the impact of external environmental factors and physical battery characteristics on RUL prediction precision, thereby contributing to a more robust and reliable prediction framework.


  • Open Access

    Segmentation of Variants of Nuclei on Whole Slide Images by Using Radiomic Features

    by Taimoor Shakeel Sheikh, Migyung Cho

    The histopathological segmentation of nuclear types is a challenging task because nuclei exhibit distinct morphologies, textures, and staining characteristics. Accurate segmentation is critical because it affects the diagnostic workflow for patient assessment. In this study, a framework was proposed for segmenting various types of nuclei from different organs of the body. The proposed framework improved the segmentation performance for each nuclear type using radiomics. First, we used distinct radiomic features to extract and analyze quantitative information about each type of nucleus and subsequently trained various classifiers based on the best input sub-features of each radiomic feature selected by a LASSO operator. Second, we inputted the outputs of the best classifier to various segmentation models to learn the variants of nuclei. Using the MoNuSAC2020 dataset, we achieved state-of-the-art segmentation performance for each category of nuclei type despite the complexity, overlapping, and obscure regions. The generalized adaptability of the proposed framework was verified by the consistent performance obtained in whole slide images of different organs of the body and radiomic features.


  • Open Access

    Clustering-Based Cancer Diagnosis Model for Whole Slide Image

    by Taimoor Shakeel Sheikh, Jaesool Shim, Migyung Cho

    Automated classification of Whole Slide Images (WSIs) is of great significance for early diagnosis of cancer. Existing approaches are trained on a specific level which affects the analysis performance due to weak supervision of patches and variants. Additionally, it is difficult to distinguish cancer subtype patches accurately from different magnification levels of WSIs. However, this can be improved by employing artificial intelligence models to address these problems, we propose a novel clustering-based cancer diagnosis (CBCD) method for WSI classification. The CBCD constructs three modules: first, we extracted patches from each magnification level of WSIs with respective cancer sub-types. Second, we employed two features (global and local) to learn discriminative and salient information of each patch. Then we find the meaningful cluster regions based on these features to quantify (select) the best patches of salient cancer subtypes by only relying on the collective characteristics of patches from different magnification levels. The clustering techniques used are k-means, gaussian mixture model, and agglomerative clustering. The quality of each clustering technique was determined using adjusted rand, and calinski harabasz scores. Later we used five state-of-the-art (SOTA) deep learning models to learn and classify cancer subtype regions of WSIs based on two types of features of patches. We also showed the results with no clustering techniques in an end-to-end supervised way by directly extracting patches from WSIs. Our method is evaluated on the public WSI dataset (KBSMC) for cancer sub-types classification and achieves better performance and great interpretability compared with the SOTA methods.

  • Open Access

    Detection of brain regions responsible for chronic pain in osteoarthritis: an fMRI-based neuroimaging study using deep learning

    by Indranath Chatterjee, Lea Baumgartner, Migyung Cho

    Introduction: Chronic pain is a multifaceted condition that has yet to be fully comprehended. It is frequently linked with a range of disorders, particularly osteoarthritis (OA), which arises from the progressive deterioration of the protective cartilage that cushions the bone endings over time.Methods: In this paper, we examine the impact of chronic pain on the brain using advanced deep learning (DL) algorithms that leverage resting-state functional magnetic resonance imaging (fMRI) data from both OA pain patients and healthy controls. Our study encompasses fMRI data from 51 pain patients and 20 healthy subjects. To differentiate chronic pain-affected OA patients from healthy controls, we introduce a DL-based computer-aided diagnosis framework that incorporates Multi-Layer Perceptron and Convolutional Neural Networks (CNN), separately.Results: Among the examined algorithms, we discovered that CNN outperformed the others and achieved a notable accuracy rate of nearly 85%. In addition, our investigation scrutinized the brain regions affected by chronic pain and successfully identified several regions that have not been mentioned in previous literature, including the occipital lobe, the superior frontal gyrus, the cuneus, the middle occipital gyrus, and the culmen.Discussion: This pioneering study explores the applicability of DL algorithms in pinpointing the differentiating brain regions in OA patients who experience chronic pain. The outcomes of our research could make a significant contribution to medical research on OA pain patients and facilitate fMRI-based pain recognition, ultimately leading to enhanced clinical intervention for chronic pain patients.


  • Open Access

    Multi-Joint Unit Network for COVID-19 Diagnosis using X-Ray and CT Images

    by Taimoor Shakeel Sheikh, Migyung Cho

    The global COVID-19 pandemic has caused a health crisis globally. Automated diagnostic methods can control the spread of the pandemic, as well as assists physicians to tackle high workload conditions through the quick treatment of affected patients. Owing to the scarcity of medical images and from different resources, the present image heterogeneity has raised challenges for achieving effective approaches to network training and effectively learning robust features. We propose a multi-joint unit network for the diagnosis of COVID-19 using the joint unit module, which leverages the receptive fields from multiple resolutions for learning rich representations. Existing approaches usually employ a large number of layers to learn the features, which consequently requires more computational power and increases the network complexity. To compensate, our joint unit module extracts low-, same-, and high-resolution feature maps simultaneously using different phases. Later, these learned feature maps are fused and utilized for classification layers. We observed that our model helps to learn sufficient information for classification without a performance loss and with faster convergence. We used three public benchmark datasets to demonstrate the performance of our network. Our proposed network consistently outperforms existing state-of-the-art approaches by demonstrating better accuracy, sensitivity, and specificity and F1-score across all datasets.


  • Open Access

    Unsupervised Learning Based on Multiple Descriptors for WSIs Diagnosis

    by Taimoor Shakeel Sheikh, Jee-Yeon Kim, Jaesool Shim, Migyung Cho

    An automatic pathological diagnosis is a challenging task because histopathological images with different cellular heterogeneity representations are sometimes limited. To overcome this, we investigated how the holistic and local appearance features with limited information can be fused to enhance the analysis performance. We propose an unsupervised deep learning model for whole-slide image diagnosis, which uses stacked autoencoders simultaneously feeding multiple-image descriptors such as the histogram of oriented gradients and local binary patterns along with the original image to fuse the heterogeneous features. The pre-trained latent vectors are extracted from each autoencoder, and these fused feature representations are utilized for classification. We observed that training with additional descriptors helps the model to overcome the limitations of multiple variants and the intricate cellular structure of histopathology data by various experiments. Our model outperforms existing state-of-the-art approaches by achieving the highest accuracies of 87.2 for ICIAR2018, 94.6 for Dartmouth, and other significant metrics for public benchmark datasets. Our model does not rely on a specific set of pre-trained features based on classifiers to achieve high performance. Unsupervised spaces are learned from the number of independent multiple descriptors and can be used with different variants of classifiers to classify cancer diseases from whole-slide images. Furthermore, we found that the proposed model classifies the types of breast and lung cancer similar to the viewpoint of pathologists by visualization. We also designed our whole-slide image processing toolbox to extract and process the patches from whole-slide images.


  • An NLP and LSTM Based Stock Prediction and Recommender System for KOSDAQ and KOSPI

    by Indranath Chatterjee, Jeon Gwan, Yong Jin Kim, Min Seok Lee, Migyung Cho


  • Multivariate discriminant analysis for branching classification of colonic tubular adenoma glands

    by Hyekyung Lee, Jong O. Kim, Jaesool Shim, Migyung Cho

    Background Many morphologic findings of histology can be translated into mathematically computerized data, and identifying important parameters is primarily pathologists' task as users. Shape-specific parameters based on computational geometry properties of glands can be used in the field of pathology. We evaluated the diagnostic utility of three shape-specific parameters: the chord intersection ratio, convexity ratio, and maximum concave area ratio for branching classification of glands. Methods Seven cases of tubular adenoma were studied. After image analysis, segmented neoplastic glands were classified into nonbranching, mild branching, and moderate branching. Using image analysis formulae for the three shape-specific parameters, we compared the values of the parameters with the branching classification results for colonic tubular adenoma. Results Multivariate discriminant analysis was used to classify the branching groups. Classification accuracies of nonbranching, mild branching, and moderate branching group based on the three shape-specific parameters were 98, 94, and 95%, respectively. More branching growth exhibited a higher chord intersection ratio and maximum concave area ratio but lower convexity ratio. We found a statistically significant difference in chord intersection ratio, maximum concave area ratio, and convexity ratio between mild, moderate, and nonbranching groups. Among the three features, the chord intersection ratio was the most significant parameter. Conclusions Shape-based parameters of chord intersection ratio, convexity ratio, and maximum concave area ratio are valid assessment parameters for irregular branching structures. For the understanding of spatial relationships of histology, the holistic geometric approach using shape-based parameters can be useful.