FastMRI Challenge

FastMRI Challenge

FastMRI Challenge

FastMRI Challenge

SNU FastMRI

SNU FastMRI

SNU FastMRI

SNU FastMRI

Reduced MRI scan time, clearer images and improved MRI access.

Reduced MRI scan time, clearer images and improved MRI access.

Brain MRI Images
Brain MRI Images

By Klim ventures

By Klim ventures

Published Apr 28, 2021

Published Apr 28, 2021

What is an MRI?

MRI(Magnetic Resonance Imaging) is a device that uses magnetic fields to create a cross-section image of the human body using signals from hydrogen in the body(the inventors of MRI equipment, Dr. Lauterbur and Mansfield, won the Nobel Prize in Medicine in 2003). MRI relies solely on magnetic fields when creating images, so it is safe from the risk of radiation exposure that other imaging devices(X-ray, CT, PET) have. MRI scans manipulate the system's magnetic field to obtain signals containing data about the human body. However, this information is not image data but frequency data of the image (or Fourier transformed data of the image; Dr. Richard Ernst received the Nobel Prize in Chemistry in 1991 for discovering this fact).


MRI system and MRI brain image


MRI Image Reconstruction

MRI acquires frequency data for these images one line at a time in a two-dimensional domain(K-space). Once enough of these data(for example, 256 lines, 256 points per line) are collected, they are then made into an image via inverse Fourier transform. This process of creating an image is called reconstruction.


(a) Fully obtained frequency domain(K-space) data; (b) the image result of inverse Fourier transformed data in (a); (c) a dataset where every other line is filled with 0 in the entire dataset (a); (d) the image result of inverse Fourier transformed data in (c). We can see that aliasing has occurred in the brain image (d). The red arrow indicates a lesion.

If not enough frequency data are collected, or some frequency data were missed during the reconstruction phase, you will find a part of the reconstructed images overlapping(aliasing, refer to Nyquist Sampling Theorem) as shown in the figure above. Therefore, it is crucial to obtain enough data for MRI scans. However, it generally takes several seconds to obtain each K-space line, so acquiring a large number of lines increases the total MRI scan time. A long scan time increases patient discomfort and decreases the cost-effectiveness of MRI scans(e.g., it would take approximately 4 minutes to obtain 256 lines if the device collects 1 line per second). Moreover, as images of various contrast levels can be acquired and used for diagnosis with MRI instead of an image with single image contrast, MRI often results in more than 30 minutes of scanning.


Contrast images commonly used in hospitals (from left: T2 image, T1 image, Angiography, SWI image). Various contrast images of MRI generally capture 5-6 types of contrast images, including distinct information about the disease. Therefore, the scanning time is often longer than 30 minutes.


What is a FastMRI?

FastMRI is an imaging method that reduces MRI scanning time by acquiring fewer data. Various methods have been proposed to solve the aforementioned aliasing problem(Figure 2(d)) caused by the lack of data, using a high-performance algorithm during the image reconstruction stage(Image reconstruction method using multi-channel phase receivers and compressed sensing are often used; references 1, 2, and 3). 

From left: Reference image -> AIRS Medical SwiftMR™(x4 high-speed scanning) -> Traditional high-speed technique image(x4 high-speed scanning)  


Example results of FastMRI. Image reconstructed to a similar quality level to that of the reference image using deep learning with a dataset undersampled to the quarter of the original requirement.


The Need for FastMRI

MRI scans typically take more than 30 minutes and the patient must lie very still during the scan. Patient movement not only degrade image quality, making diagnosis difficult, but also incur time delay due to the need for re-imaging, increasing medical costs. 

In addition, a significant proportion of patients experience discomfort and anxiety during MRI scans because of the loud noise and tight spaces they have to endure for the length of the procedure. In particular, claustrophobic or pediatric patients often find it extremely difficult. 

If the length of the examination can be shortened without compromising the MRI resolution, it would reduce the level of fear and discomfort during MRI, which will then help obtain clear images by minimizing patient movement. Furthermore, it is expected that the increased speed will lead to cost savings, which then would allow more people to gain access to MRI examinations.


External Links
http://fastmri.snu.ac.kr
https://www.youtube.com/@SNUFastMRIChallenge/featured
https://n.news.naver.com/mnews/article/009/0005016470?sid=105 

What is an MRI?

MRI(Magnetic Resonance Imaging) is a device that uses magnetic fields to create a cross-section image of the human body using signals from hydrogen in the body(the inventors of MRI equipment, Dr. Lauterbur and Mansfield, won the Nobel Prize in Medicine in 2003). MRI relies solely on magnetic fields when creating images, so it is safe from the risk of radiation exposure that other imaging devices(X-ray, CT, PET) have. MRI scans manipulate the system's magnetic field to obtain signals containing data about the human body. However, this information is not image data but frequency data of the image (or Fourier transformed data of the image; Dr. Richard Ernst received the Nobel Prize in Chemistry in 1991 for discovering this fact).


MRI system and MRI brain image


MRI Image Reconstruction

MRI acquires frequency data for these images one line at a time in a two-dimensional domain(K-space). Once enough of these data(for example, 256 lines, 256 points per line) are collected, they are then made into an image via inverse Fourier transform. This process of creating an image is called reconstruction.


(a) Fully obtained frequency domain(K-space) data; (b) the image result of inverse Fourier transformed data in (a); (c) a dataset where every other line is filled with 0 in the entire dataset (a); (d) the image result of inverse Fourier transformed data in (c). We can see that aliasing has occurred in the brain image (d). The red arrow indicates a lesion.

If not enough frequency data are collected, or some frequency data were missed during the reconstruction phase, you will find a part of the reconstructed images overlapping(aliasing, refer to Nyquist Sampling Theorem) as shown in the figure above. Therefore, it is crucial to obtain enough data for MRI scans. However, it generally takes several seconds to obtain each K-space line, so acquiring a large number of lines increases the total MRI scan time. A long scan time increases patient discomfort and decreases the cost-effectiveness of MRI scans(e.g., it would take approximately 4 minutes to obtain 256 lines if the device collects 1 line per second). Moreover, as images of various contrast levels can be acquired and used for diagnosis with MRI instead of an image with single image contrast, MRI often results in more than 30 minutes of scanning.


Contrast images commonly used in hospitals (from left: T2 image, T1 image, Angiography, SWI image). Various contrast images of MRI generally capture 5-6 types of contrast images, including distinct information about the disease. Therefore, the scanning time is often longer than 30 minutes.


What is a FastMRI?

FastMRI is an imaging method that reduces MRI scanning time by acquiring fewer data. Various methods have been proposed to solve the aforementioned aliasing problem(Figure 2(d)) caused by the lack of data, using a high-performance algorithm during the image reconstruction stage(Image reconstruction method using multi-channel phase receivers and compressed sensing are often used; references 1, 2, and 3). 

From left: Reference image -> AIRS Medical SwiftMR™(x4 high-speed scanning) -> Traditional high-speed technique image(x4 high-speed scanning)  


Example results of FastMRI. Image reconstructed to a similar quality level to that of the reference image using deep learning with a dataset undersampled to the quarter of the original requirement.


The Need for FastMRI

MRI scans typically take more than 30 minutes and the patient must lie very still during the scan. Patient movement not only degrade image quality, making diagnosis difficult, but also incur time delay due to the need for re-imaging, increasing medical costs. 

In addition, a significant proportion of patients experience discomfort and anxiety during MRI scans because of the loud noise and tight spaces they have to endure for the length of the procedure. In particular, claustrophobic or pediatric patients often find it extremely difficult. 

If the length of the examination can be shortened without compromising the MRI resolution, it would reduce the level of fear and discomfort during MRI, which will then help obtain clear images by minimizing patient movement. Furthermore, it is expected that the increased speed will lead to cost savings, which then would allow more people to gain access to MRI examinations.


External Links
http://fastmri.snu.ac.kr
https://www.youtube.com/@SNUFastMRIChallenge/featured
https://n.news.naver.com/mnews/article/009/0005016470?sid=105 

What is an MRI?

MRI(Magnetic Resonance Imaging) is a device that uses magnetic fields to create a cross-section image of the human body using signals from hydrogen in the body(the inventors of MRI equipment, Dr. Lauterbur and Mansfield, won the Nobel Prize in Medicine in 2003). MRI relies solely on magnetic fields when creating images, so it is safe from the risk of radiation exposure that other imaging devices(X-ray, CT, PET) have. MRI scans manipulate the system's magnetic field to obtain signals containing data about the human body. However, this information is not image data but frequency data of the image (or Fourier transformed data of the image; Dr. Richard Ernst received the Nobel Prize in Chemistry in 1991 for discovering this fact).


MRI system and MRI brain image


MRI Image Reconstruction

MRI acquires frequency data for these images one line at a time in a two-dimensional domain(K-space). Once enough of these data(for example, 256 lines, 256 points per line) are collected, they are then made into an image via inverse Fourier transform. This process of creating an image is called reconstruction.


(a) Fully obtained frequency domain(K-space) data; (b) the image result of inverse Fourier transformed data in (a); (c) a dataset where every other line is filled with 0 in the entire dataset (a); (d) the image result of inverse Fourier transformed data in (c). We can see that aliasing has occurred in the brain image (d). The red arrow indicates a lesion.

If not enough frequency data are collected, or some frequency data were missed during the reconstruction phase, you will find a part of the reconstructed images overlapping(aliasing, refer to Nyquist Sampling Theorem) as shown in the figure above. Therefore, it is crucial to obtain enough data for MRI scans. However, it generally takes several seconds to obtain each K-space line, so acquiring a large number of lines increases the total MRI scan time. A long scan time increases patient discomfort and decreases the cost-effectiveness of MRI scans(e.g., it would take approximately 4 minutes to obtain 256 lines if the device collects 1 line per second). Moreover, as images of various contrast levels can be acquired and used for diagnosis with MRI instead of an image with single image contrast, MRI often results in more than 30 minutes of scanning.


Contrast images commonly used in hospitals (from left: T2 image, T1 image, Angiography, SWI image). Various contrast images of MRI generally capture 5-6 types of contrast images, including distinct information about the disease. Therefore, the scanning time is often longer than 30 minutes.


What is a FastMRI?

FastMRI is an imaging method that reduces MRI scanning time by acquiring fewer data. Various methods have been proposed to solve the aforementioned aliasing problem(Figure 2(d)) caused by the lack of data, using a high-performance algorithm during the image reconstruction stage(Image reconstruction method using multi-channel phase receivers and compressed sensing are often used; references 1, 2, and 3). 

From left: Reference image -> AIRS Medical SwiftMR™(x4 high-speed scanning) -> Traditional high-speed technique image(x4 high-speed scanning)  


Example results of FastMRI. Image reconstructed to a similar quality level to that of the reference image using deep learning with a dataset undersampled to the quarter of the original requirement.


The Need for FastMRI

MRI scans typically take more than 30 minutes and the patient must lie very still during the scan. Patient movement not only degrade image quality, making diagnosis difficult, but also incur time delay due to the need for re-imaging, increasing medical costs. 

In addition, a significant proportion of patients experience discomfort and anxiety during MRI scans because of the loud noise and tight spaces they have to endure for the length of the procedure. In particular, claustrophobic or pediatric patients often find it extremely difficult. 

If the length of the examination can be shortened without compromising the MRI resolution, it would reduce the level of fear and discomfort during MRI, which will then help obtain clear images by minimizing patient movement. Furthermore, it is expected that the increased speed will lead to cost savings, which then would allow more people to gain access to MRI examinations.


External Links
http://fastmri.snu.ac.kr
https://www.youtube.com/@SNUFastMRIChallenge/featured
https://n.news.naver.com/mnews/article/009/0005016470?sid=105 

FOUNDER'S PARTNER

FOUNDER'S PARTNER

ⓒ 2024 Klim ventures

ⓒ 2024 Klim ventures