Broad research interests of magnetic resonance imaging (MRI) laboratory at Korea Advanced Institute of Science and Technology (KAIST) lie in (i) developing new MRI techniques of image acquisition, processing, and analysis, (ii) revealing brain waste clearance pathway and mechanism through neurofluid MRI, (iii) applying deep learning algorithms to MRI acceleration, image analysis and synthesis, and clinical diagnosis. These technical developments can make big differences in understanding baseline and functional brain physiology, and also diagnosis / prognosis / treatment of our body.
1) MR Imaging of Neurofluid to Reveal Brain Waste Clearance Pathways and Mechanisms
Dementia research has traditionally focused on toxic proteins such as beta-amyloid and tau, yet the exact mechanisms through which the brain expels these toxic waste materials remain unclear. While for centuries it was believed that the brain lacked lymphatic vessels, the discovery of meningeal lymphatic vessels in the brain, revealed in a 2015 Nature paper, has sparked global interest and activation of research into brain waste clearance. Although it is generally anticipated that brain waste is transported from the brain's interior lymphatic system to cerebrospinal fluid and then expelled through meningeal lymphatic vessels, several models exist to explain this process. However, challenges persist in accurately describing the mechanism due to numerous inconsistencies in experimental data and flow dynamics perspectives. One of the difficulties in studying brain waste clearance is the slow flow of the associated lymphatic system, cerebrospinal fluid, and lymphatic system, making visualization challenging. Additionally, quantifying flow or pulsatility to understand the waste clearance function is an even more challenging task. Specifically, direct contrast agent injection through cerebrospinal fluid is not permitted in many countries, including South Korea and the United States, making the development of non-invasive imaging techniques crucial. Overcoming these limitations through the development of new non-invasive imaging techniques can provide a precise understanding of brain waste clearance mechanisms in humans. This advancement has the potential to be utilized in the early detection and treatment of neurodegenerative brain conditions, such as dementia. For this, we are developing new non-Invasive imaging techniques for systematic and quantitative measurement of brain waste clearance in humans and animals. Specifically, we develop non-invasive imaging techniques enabling quantitative measurement of the functions of the meningeal lymphatic system, cerebrospinal fluid, and glymphatic system.
2) Physiological, Metabolic, and Functional MRI
Noninvasive visualization of blood vessels, blood flow, blood permeability, and blood oxygenation level dependence is an important step toward understanding sources of brain diseases and brain function. We are developing methods to image both arteries and veins within a single MRI acquisition without compromising their image qualities, map blood perfusion in brain and kidney noninvasively, and map brain function with high spatial resolutions. These techniques have been used to understand diseases such as brain tumors and stroke, map renal function noninvasively, and understand signal sources of high-resolution functional MRI. Also, we have been improving the image quality of single-shot echo planar imaging (EPI), which is typically used for the physiological, metabolic, and functional MRI.
3) Deep Learning for MRI Acceleration
Deep learning has been of interest to many research areas and one of the hottest applications is biomedical imaging. We are trying to develop and apply deep learning algorithms for MRI artifacts suppression, denoising, and fast MRI data acquisition. Simulation-based physics-informed neural network is an interesting approach that does not require a separate measurement of ground truth, and thus has high generalization ability and wide applicability. Also, quantitative mapping of tissue parameters require long processing time, which can be significantly reduced using deep learning.
4) Deep Learning for Image Segmentation and Synthesis and Disease Diagnosis
Deep learning finds applications in image analysis including segmentation of organs and lesions, image synthesis, and diagnosis of various diseases. Our focus involves developing and applying deep learning algorithms for segmentation, diagnosis/prognosis/treatment planning and monitoring of various diseases. Notably, Huan Minh Luu won first place in the 2021 Brain Tumor Segmentation Challenge (BRATS) among 2200 competitors worldwide.
5) Biomedical Applications
We have tried to integrate all these imaging techniques for multi-parametric assessment of various brain diseases. We believe combination of the various aspects of the brain would provide much more accurate information than conventional approaches. Also we expect this approach can stimulate interdisciplinary collaborative studies involving engineers, neuroscientists, and clinicians.