Joint Research Laboratory | mai-laboratory

MAI-LAB @

History of JRL Yonsei-UCSD

2015, Started research supervisor communication
2016, graduate student participation
2017, Yonsei-UCSD Joint Research Laboratory Agreement
2017, Published results of research to international journals
2018, Started deep research interchange and published SCI papers
2019, Visited UCSD
2019, Installed AI (Artificial Intelligence) program developed from Yonsei Univ. in UCSD laboratory
2019, Submitted joint research results to RSNA (Radiological Society of North America)
2019, Research meeting in Montreal, Canada

Intervertebral disc segmentation in MR images is challenging owing to their complex shapes and non-uniform intensity.
This study introduces a robust deep-learning segmentation network which can successfully segment intervertebral discs with complex boundaries.

MAI-LAB @

History of JRL Yonsei-Martinos Center

2022, Yohan Jun, an MAI-LAB alumnus, joined the Martinos Center as a Postdoctoral Researcher
2024, Yohan Jun was promoted to instructor in Radiology at Harvard Medical School
2025, Jinho Joo, a Ph.D. student at MAI-LAB, began collaborative research with the Martinos Center
2026, MAI-LAB and the Martinos Center established the Yonsei-Martinos Center Joint Research Laboratory (JRL)
2026, Yohan Jun was appointed Assistant Professor at Korea University
2026, Jinho Joo joined the Martions Center as a Visiting Researcher through the JRL collaboration

Vertebral body segmentation on MR images provides clinically useful information including quantitative biomarkers, volume, and shape.
This study proposes a semi-automatic algorithm for the segmentation of vertebral bodies in magnetic resonance images of the human lumbar spine.
Our method achieved a 90% dice similarity coefficient and significantly reduces the user’s role.

In a translational diagnostic workflow of the knee MRI involving quantitative sequences, there is a need for segmentation in order to evaluate specific regions.
In the knee, articular cartilage and menisci are among those frequently segmented, to determine quantitative MR (qMR) values such as T2.
This study demonstrates an application for deep learning (DL)-based automated segmentation of cartilage and menisci.