KAIST researchers proposed new technology that reduces MRI (magnetic resonance imaging) acquisition time to less than a sixth of the conventional method. They made a reconstruction method using machine learning of multilayer perception (MLP) algorithm to accelerate imaging time.
High-quality image can be reconstructed from subsampled data using the proposed method. This method can be further applied to various k-space subsampling patterns in a phase encoding direction, and its processing can be performed in real time.
The research, led by Professor Hyun Wook Park from the Department of Electrical Engineering, was described in Medical Physics as the cover paper last December. Ph.D. candidate Kinam Kwon is the first author.
MRI is an imaging technique that allows various contrasts of soft tissues without using radioactivity. Since MRI could image not only anatomical structures, but also functional and physiological features, it is widely used in medical diagnoses. However, one of the major shortcomings of MRI is its long imaging time. It induces patients’ discomfort, which is closely related to voluntary and involuntary motions, thereby deteriorating the quality of the MR images. In addition, lengthy imaging times limit the system’s throughput, which results in the long waiting times of patients as well as the increased medical expenses.
To reconstruct MR images from subsampled data, the team applied the MLP to reduce aliasing artifacts generated by subsampling in k-space. The MLP is learned from training data to map aliased input images into desired alias-free images. The input of the MLP is all voxels in the aliased lines of multichannel real and imaginary images from the subsampled k-space data, and the desired output is all voxels in the corresponding alias-free line of the root-sum-of-squares of multichannel images from fully sampled k-space data. Aliasing artifacts in an image reconstructed from subsampled data were reduced by line-by-line processing of the learned MLP architecture.
Reconstructed images from the proposed method are better than those from compared methods in terms of normalized root-mean-square error. The proposed method can be applied to image reconstruction for any k-space subsampling patterns in a phase encoding direction. Moreover, to further reduce the reconstruction time, it is easily implemented by parallel processing.
To address the aliasing artifact phenomenon, the team employed a parallel imaging technique using several receiver coils of various sensitivities and a compressed sensing technique using sparsity of signals.
Existing methods are heavily affected by sub-sampling patterns, but the team’s technique is applicable for various sub-sampling patterns, resulting in superior reconstructed images compared to existing methods, as well as allowing real-time reconstruction.
Professor Park said, "MRIs have become essential equipment in clinical diagnosis. However, the time consumption and the cost led to many inconveniences." He continued, "This method using machine learning could greatly improve the patients’ satisfaction with medical service." This research was funded by the Ministry of Science and ICT.
(Firgure 1. Cover of Medical Physics for December 2017)
(Figure 2. Concept map for the suggested network)
(Figure 3. Concept map for conventional MRI image acquisition and accelerated image acquisiton)
<(From left) Dr. Dongjo Yoon, Professor Je-Kyun Park from the Department of Bio and Brain Engineering, (upper right) Professor Yoonkey Nam, Dr. Soo Jee Kim> Existing three-dimensional (3D) neuronal culture technology has limitations in brain research due to the difficulty of precisely replicating the brain's complex multilayered structure and the lack of a platform that can simultaneously analyze both structure and function. A KAIST research team has successfully developed an integrated
2025-07-16<(From left) Ph.D candidate See-On Park, Professor Jongwon Lee, and Professor Shinhyun Choi> In the midst of the co-development of artificial intelligence and robotic advancements, developing technologies that enable robots to efficiently perceive and respond to their surroundings like humans has become a crucial task. In this context, Korean researchers are gaining attention for newly implementing an artificial sensory nervous system that mimics the sensory nervous system of living org
2025-07-16<(From Left) Ph.D. Candidate Hyunmin Eun, Distinguished Professor Sang Yup Lee, , Dr. Cindy Pricilia Surya Prabowo> The application of systems metabolic engineering strategies, along with the construction of an electron channeling system, has enabled the first gram-per-liter scale production of lutein from Corynebacterium glutamicum, providing a viable alternative to plant-derived lutein production. A research group at KAIST has successfully engineered a microbial strain capable o
2025-07-14<Picture1.(From Left) Prof. Seungbum Hong, Ph.D candidate Youngwoo Choi> Steel alloys used in automobiles and machinery parts are typically manufactured through a melting process at high temperatures. The phenomenon where the components remain unchanged during melting is called “congruent melting.” KAIST researchers have now addressed this process—traditionally only possible through high-temperature experiments—using artificial intelligence (AI). This study dra
2025-07-14<Photo1. (From left) Ph.D candidate Yong-hoo Kwon, M.S candidate Do-hwan Kim, Professor Jung-woo Choi, Dr. Dong-heon Lee> 'Acoustic separation and classification technology' is a next-generation artificial intelligence (AI) core technology that enables the early detection of abnormal sounds in areas such as drones, fault detection of factory pipelines, and border surveillance systems, or allows for the separation and editing of spatial audio by sound source when producing AR/VR conten
2025-07-13