Publications by authors named "Beomgu Kang"
Magn Reson Med
November 2024
Article Synopsis
- The study aims to create a quick denoising method for high-dimensional MRI data using self-supervised learning that doesn’t need perfectly clean images.
- The new deep learning framework leverages redundancies in MRI data to effectively reduce noise, addressing the limitations posed by signal-to-noise ratio (SNR) issues and complex signal models.
- Results showed significant improvements in denoising performance and quantification accuracy when tested on different MRI datasets, suggesting the MD-S2S technique could enhance various multi-dimensional MRI applications.
Article Synopsis
- MRI scan times can be reduced by using fewer sampled k-space lines, utilizing data from multiple receiver coils, with the goal of improving clinical application efficiency.
- This study introduces a deep-learning method to enhance parallel MRI imaging by using a cycle interpolator network that reconstructs data even with a small number of auto-calibration signals (ACS) in noisy environments.
- Results indicate that this method effectively reconstructs images with fewer ACS lines, reduces noise, eliminates artifacts, and outperforms traditional methods like GRAPPA, showing promise for improved MRI accuracy.
Article Synopsis
- The study aims to create a fast, deep-learning method for quantitative magnetization-transfer contrast (MTC)-MR fingerprinting (MRF) that estimates tissue parameters while correcting variations in magnetic field strength (B) and B.
- A recurrent neural network was developed to quickly quantify tissue parameters across different MRF acquisition schedules, using measured B and B maps for accurate parameter mapping.
- Results showed that this approach successfully estimates tissue parameters even with significant B and B discrepancies, improving the accuracy of brain-tissue maps and could work alongside existing MRF techniques.
Article Synopsis
- - Magnetization transfer contrast MR fingerprinting (MTC-MRF) is an advanced imaging technique that measures both free bulk water and macromolecule parameters through complex scan settings.
- - The study presents a new method called learning-based optimization of the acquisition schedule (LOAS) which simplifies the scan parameters in MRF acquisitions to improve tissue parameter determination.
- - Tests indicated that LOAS provided better quantification accuracy and efficiency in imaging compared to older methods, suggesting its potential as a valuable tool for designing MRF pulse sequences.
Magn Reson Med
April 2021
Article Synopsis
- The study aims to create a rapid, quantitative 3D magnetization transfer contrast (MTC) framework using unsupervised learning for better imaging techniques like CEST and nuclear Overhauser enhancement.
- Pseudo-random RF saturation and relaxation delays were utilized with MR fingerprinting to generate signals, while a convolutional neural network was developed to solve the MTC Bloch equation, outperforming traditional methods.
- The unsupervised learning approach showed strong agreement with conventional methods in estimating MTC parameters, while significantly reducing computation time by about 1000 times, indicating its potential for efficient imaging.
