A5068701548- Advanced MRI Techniques and Applications
- Radiomics and Machine Learning in Medical Imaging
- Medical Imaging Techniques and Applications
- Advanced Neuroimaging Techniques and Applications
- COVID-19 diagnosis using AI
- MRI in cancer diagnosis
- Domain Adaptation and Few-Shot Learning
- AI in cancer detection
- Medical Image Segmentation Techniques
- Colorectal Cancer Screening and Detection
- Gastrointestinal Bleeding Diagnosis and Treatment
- Advanced X-ray Imaging Techniques
- Sparse and Compressive Sensing Techniques
- Mathematical Biology Tumor Growth
- Advanced X-ray and CT Imaging
- Artificial Intelligence in Healthcare and Education
- Underwater Acoustics Research
- Ultrasound Imaging and Elastography
- Fetal and Pediatric Neurological Disorders
- Medical Imaging and Analysis
- Image and Video Quality Assessment
- Advanced Electron Microscopy Techniques and Applications
- Gastric Cancer Management and Outcomes
- Gastrointestinal motility and disorders
- Color Science and Applications
Yonsei University
2014-2025
Purpose To demonstrate accurate MR image reconstruction from undersampled k‐space data using cross‐domain convolutional neural networks (CNNs) Methods Cross‐domain CNNs consist of 3 components: (1) a deep CNN operating on the (KCNN), (2) an domain (ICNN), and (3) interleaved consistency operations. These components are alternately applied, each is trained to minimize loss between reconstructed corresponding fully sampled k‐spaces. The final obtained by forward‐propagating through entire...
Recent advances in deep learning-based medical image segmentation studies achieve nearly human-level performance fully supervised manner. However, acquiring pixel-level expert annotations is extremely expensive and laborious imaging fields. Unsupervised domain adaptation (UDA) can alleviate this problem, which makes it possible to use annotated data one modality train a network that successfully perform on target with no labels. In work, we propose SDC-UDA, simple yet effective volumetric...
With increasing fields of application for neural networks and the development networks, ability to explain deep learning models is also becoming increasingly important. Especially, prior practical applications, it crucial analyze a model's inference process generating results. A common explanation method Class Activation Mapping(CAM) based where often used understand last layer convolutional popular in field Computer Vision. In this paper, we propose novel CAM named Relevance-weighted...
Magnetic resonance imaging (MRI) can provide diagnostic information with high-resolution and high-contrast images. However, MRI requires a relatively long scan time compared to other medical techniques, where might occur patient's discomfort limit the increase in resolution of magnetic (MR) image. In this study, we propose Joint Deep Model-based MR Image Coil Sensitivity Reconstruction Network, called Joint-ICNet, which jointly reconstructs an image coil sensitivity maps from undersampled...
Abstract Black-blood (BB) imaging is used to complement contrast-enhanced 3D gradient-echo (CE 3D-GRE) for detecting brain metastases, requiring additional scan time. In this study, we proposed deep-learned BB with an auto-labelling technique and convolutional neural networks metastases detection without scan. Patients were randomly selected training (29 sets) testing (36 sets). Two neuroradiologists independently evaluated original images, assessing the degree of blood vessel suppression...
Purpose To develop and evaluate a method of parallel imaging time‐of‐flight (TOF) MRA using deep multistream convolutional neural networks (CNNs). Methods A network (“DPI‐net”) was developed to reconstruct 3D multichannel from undersampled data. It comprises 2 deep‐learning networks: CNNs for extracting feature maps images reconstruction reconstructing the output maps. The were evaluated normalized root mean square error (NRMSE), peak signal‐to‐noise ratio (PSNR), structural similarity...
Recent advances in deep learning-based medical image segmentation studies achieve nearly human-level performance when fully supervised condition. However, acquiring pixel-level expert annotations is extremely expensive and laborious imaging fields. Unsupervised domain adaptation can alleviate this problem, which makes it possible to use annotated data one modality train a network that successfully perform on target with no labels. In work, we propose self-training based unsupervised...
By automatically classifying the stomach, small bowel, and colon, reading time of wireless capsule endoscopy (WCE) can be reduced. In addition, it is an essential first preprocessing step to localize bowel in order apply automated lesion detection algorithms based on deep learning. The purpose study was develop method from long untrimmed videos captured WCE. Through this, stomach colon also distinguished. proposed a convolutional neural network (CNN) with temporal filtering predicted...
With the advances of deep learning, many medical image segmentation studies achieve human-level performance when in fully supervised condition. However, it is extremely expensive to acquire annotation on every data fields, especially magnetic resonance images (MRI) that comprise different contrasts. Unsupervised methods can alleviate this problem; however, drop inevitable compared methods. In work, we propose a self-training based unsupervised-learning framework performs automatic Vestibular...
Motivation: Scanning for multi-contrast MR images is time-consuming. To reduce scan time, it beneficial to explore methods efficiently synthesizing target contrast from existing scans. Goal(s): address the stability issues encountered when dealing with image domains individually, we propose a methodology effectively while incorporating domains. Approach: Our model novel unified diffusion (UDM) that improves synthesis of detailed anatomical structures in through an ensemble method. Results:...
Knee effusion, a common and important indicator of joint diseases such as osteoarthritis, is typically more discernible on magnetic resonance imaging (MRI) scans compared to radiographs. However, the use radiographs for early detection knee effusion remains promising due their cost-effectiveness accessibility. This multi-center prospective study collected total 1413 from four hospitals between February 2022 March 2023, which 1281 were analyzed after exclusions. To automatically detect...
Image adjustment methods are one of the most widely used post-processing techniques for enhancing image quality and improving visual preference human system (HVS). However, assessment adjusted images has been mainly dependent on subjective evaluations. Also, recently developed automatic have focused evaluating distorted degraded by compression or noise. The effects colorfulness, contrast, sharpness adjustments overlooked. In this study, we propose a fully method that evaluates...
Purpose To develop an effective method that can suppress noise in successive multiecho T 2 (*)‐weighted magnetic resonance (MR) brain images while preventing filtering artifacts. Materials and Methods For the simulation experiments, we used multiple ‐weighted of anatomical phantom. vivo MR were acquired from five healthy subjects using a gradient‐recalled‐echo (MGRE) sequence with 3T MRI system. Our denoising is nonlinear filter whose weights are determined by tissue characteristics among...
Multiple-echo magnetic resonance images are a series of acquired at different echo times. Especially, phase multiple-echo often used to measure susceptibility tissues. But these tend have low SNR because high sampling rates reduce the scan time. Conventional filters can effectively noise, but introduce spatial artifacts. To obtain high-quality without artifacts, we developed new denoising method based on tissue relaxation properties. As result, could high-SNR and high-resolution image mask.
This paper introduces a stereoscopy in Computed Tomography (CT), and two projection methods to create stereoscopy. One is summing the other Maximum Intensity Projection (MIP). And then, comparison of stereoscopies using MIP presented. Numerical experiments from real CT data were performed for techniques. Also Anaglyph 3D method used offer vivid visualization effect readers, provides advantages disadvantages each conclusion.