A
Figure 1.
Routine brain protocol comparing (top row) conventional reconstruction with 5 mm, 22 slices and (bottom row) AIR™ Recon DL with 3 mm, 42 slices. Conventional reconstructed images: (A) Axial T2w, 0.6 x 0.8 x 5 mm, 1:15 min.; (B) axial T2 FLAIR, 0.7 x 0.8 x 5 mm, 2:05 min; (C) axial T1w, 0.6 x 0.6 x 5 mm, 2:16 min; and (D) axial T2*w, 0.6 x 1 x 5 mm, 1:43 min. AIR™ Recon DL reconstructed images: (E) Axial T2w, 0.6 x 0.8 x 3 mm, 1:12 min; (F) axial T2 FLAIR, 0.7 x 0.8 x 3 mm, 2:33 min (G)axial T1w, 0.6 x 0.6 x 3 mm, 1:58 min; and (H) axial T2*w, 0.6 x 1 x 3 mm, 1:40 min.
B
Figure 1.
Routine brain protocol comparing (top row) conventional reconstruction with 5 mm, 22 slices and (bottom row) AIR™ Recon DL with 3 mm, 42 slices. Conventional reconstructed images: (A) Axial T2w, 0.6 x 0.8 x 5 mm, 1:15 min.; (B) axial T2 FLAIR, 0.7 x 0.8 x 5 mm, 2:05 min; (C) axial T1w, 0.6 x 0.6 x 5 mm, 2:16 min; and (D) axial T2*w, 0.6 x 1 x 5 mm, 1:43 min. AIR™ Recon DL reconstructed images: (E) Axial T2w, 0.6 x 0.8 x 3 mm, 1:12 min; (F) axial T2 FLAIR, 0.7 x 0.8 x 3 mm, 2:33 min (G)axial T1w, 0.6 x 0.6 x 3 mm, 1:58 min; and (H) axial T2*w, 0.6 x 1 x 3 mm, 1:40 min.
C
Figure 1.
Routine brain protocol comparing (top row) conventional reconstruction with 5 mm, 22 slices and (bottom row) AIR™ Recon DL with 3 mm, 42 slices. Conventional reconstructed images: (A) Axial T2w, 0.6 x 0.8 x 5 mm, 1:15 min.; (B) axial T2 FLAIR, 0.7 x 0.8 x 5 mm, 2:05 min; (C) axial T1w, 0.6 x 0.6 x 5 mm, 2:16 min; and (D) axial T2*w, 0.6 x 1 x 5 mm, 1:43 min. AIR™ Recon DL reconstructed images: (E) Axial T2w, 0.6 x 0.8 x 3 mm, 1:12 min; (F) axial T2 FLAIR, 0.7 x 0.8 x 3 mm, 2:33 min (G)axial T1w, 0.6 x 0.6 x 3 mm, 1:58 min; and (H) axial T2*w, 0.6 x 1 x 3 mm, 1:40 min.
D
Figure 1.
Routine brain protocol comparing (top row) conventional reconstruction with 5 mm, 22 slices and (bottom row) AIR™ Recon DL with 3 mm, 42 slices. Conventional reconstructed images: (A) Axial T2w, 0.6 x 0.8 x 5 mm, 1:15 min.; (B) axial T2 FLAIR, 0.7 x 0.8 x 5 mm, 2:05 min; (C) axial T1w, 0.6 x 0.6 x 5 mm, 2:16 min; and (D) axial T2*w, 0.6 x 1 x 5 mm, 1:43 min. AIR™ Recon DL reconstructed images: (E) Axial T2w, 0.6 x 0.8 x 3 mm, 1:12 min; (F) axial T2 FLAIR, 0.7 x 0.8 x 3 mm, 2:33 min (G)axial T1w, 0.6 x 0.6 x 3 mm, 1:58 min; and (H) axial T2*w, 0.6 x 1 x 3 mm, 1:40 min.
E
Figure 1.
Routine brain protocol comparing (top row) conventional reconstruction with 5 mm, 22 slices and (bottom row) AIR™ Recon DL with 3 mm, 42 slices. Conventional reconstructed images: (A) Axial T2w, 0.6 x 0.8 x 5 mm, 1:15 min.; (B) axial T2 FLAIR, 0.7 x 0.8 x 5 mm, 2:05 min; (C) axial T1w, 0.6 x 0.6 x 5 mm, 2:16 min; and (D) axial T2*w, 0.6 x 1 x 5 mm, 1:43 min. AIR™ Recon DL reconstructed images: (E) Axial T2w, 0.6 x 0.8 x 3 mm, 1:12 min; (F) axial T2 FLAIR, 0.7 x 0.8 x 3 mm, 2:33 min (G)axial T1w, 0.6 x 0.6 x 3 mm, 1:58 min; and (H) axial T2*w, 0.6 x 1 x 3 mm, 1:40 min.
F
Figure 1.
Routine brain protocol comparing (top row) conventional reconstruction with 5 mm, 22 slices and (bottom row) AIR™ Recon DL with 3 mm, 42 slices. Conventional reconstructed images: (A) Axial T2w, 0.6 x 0.8 x 5 mm, 1:15 min.; (B) axial T2 FLAIR, 0.7 x 0.8 x 5 mm, 2:05 min; (C) axial T1w, 0.6 x 0.6 x 5 mm, 2:16 min; and (D) axial T2*w, 0.6 x 1 x 5 mm, 1:43 min. AIR™ Recon DL reconstructed images: (E) Axial T2w, 0.6 x 0.8 x 3 mm, 1:12 min; (F) axial T2 FLAIR, 0.7 x 0.8 x 3 mm, 2:33 min (G)axial T1w, 0.6 x 0.6 x 3 mm, 1:58 min; and (H) axial T2*w, 0.6 x 1 x 3 mm, 1:40 min.
G
Figure 1.
Routine brain protocol comparing (top row) conventional reconstruction with 5 mm, 22 slices and (bottom row) AIR™ Recon DL with 3 mm, 42 slices. Conventional reconstructed images: (A) Axial T2w, 0.6 x 0.8 x 5 mm, 1:15 min.; (B) axial T2 FLAIR, 0.7 x 0.8 x 5 mm, 2:05 min; (C) axial T1w, 0.6 x 0.6 x 5 mm, 2:16 min; and (D) axial T2*w, 0.6 x 1 x 5 mm, 1:43 min. AIR™ Recon DL reconstructed images: (E) Axial T2w, 0.6 x 0.8 x 3 mm, 1:12 min; (F) axial T2 FLAIR, 0.7 x 0.8 x 3 mm, 2:33 min (G)axial T1w, 0.6 x 0.6 x 3 mm, 1:58 min; and (H) axial T2*w, 0.6 x 1 x 3 mm, 1:40 min.
H
Figure 1.
Routine brain protocol comparing (top row) conventional reconstruction with 5 mm, 22 slices and (bottom row) AIR™ Recon DL with 3 mm, 42 slices. Conventional reconstructed images: (A) Axial T2w, 0.6 x 0.8 x 5 mm, 1:15 min.; (B) axial T2 FLAIR, 0.7 x 0.8 x 5 mm, 2:05 min; (C) axial T1w, 0.6 x 0.6 x 5 mm, 2:16 min; and (D) axial T2*w, 0.6 x 1 x 5 mm, 1:43 min. AIR™ Recon DL reconstructed images: (E) Axial T2w, 0.6 x 0.8 x 3 mm, 1:12 min; (F) axial T2 FLAIR, 0.7 x 0.8 x 3 mm, 2:33 min (G)axial T1w, 0.6 x 0.6 x 3 mm, 1:58 min; and (H) axial T2*w, 0.6 x 1 x 3 mm, 1:40 min.
A
Figure 2.
Intracerebral hemorrhage with AIR™ Recon DL. (A) Axial T2 FLAIR, 0.5 x 0.6 x 5 mm, 2:33 min; (B) axial T2w, 0.5 x 0.6 x 5 mm 1:05 min; and (C) axial T1w, 0.8 x 0.9 x 5 mm, 52 sec.
B
Figure 2.
Intracerebral hemorrhage with AIR™ Recon DL. (A) Axial T2 FLAIR, 0.5 x 0.6 x 5 mm, 2:33 min; (B) axial T2w, 0.5 x 0.6 x 5 mm 1:05 min; and (C) axial T1w, 0.8 x 0.9 x 5 mm, 52 sec.
C
Figure 2.
Intracerebral hemorrhage with AIR™ Recon DL. (A) Axial T2 FLAIR, 0.5 x 0.6 x 5 mm, 2:33 min; (B) axial T2w, 0.5 x 0.6 x 5 mm 1:05 min; and (C) axial T1w, 0.8 x 0.9 x 5 mm, 52 sec.
A-1
Figure 3.
L-spine comparing conventional reconstructed images with AIR™ Recon DL reconstructed images, demonstrating a 14% reduction in scan time, a 52% improvement in spatial resolution and 70% increase in SNR. (A) Conventional contrast-enhanced sagittal T1w, 1.0 x 1.6 x 4 mm, 2:28 min. and (B) AIR™ Recon DL contrast-enhanced sagittal T1w, 0.9 x 1.2 x 4 mm, 2:08 min.
A-2
Figure 3.
L-spine comparing conventional reconstructed images with AIR™ Recon DL reconstructed images, demonstrating a 14% reduction in scan time, a 52% improvement in spatial resolution and 70% increase in SNR. (A) Conventional contrast-enhanced sagittal T1w, 1.0 x 1.6 x 4 mm, 2:28 min. and (B) AIR™ Recon DL contrast-enhanced sagittal T1w, 0.9 x 1.2 x 4 mm, 2:08 min.
B-1
Figure 3.
L-spine comparing conventional reconstructed images with AIR™ Recon DL reconstructed images, demonstrating a 14% reduction in scan time, a 52% improvement in spatial resolution and 70% increase in SNR. (A) Conventional contrast-enhanced sagittal T1w, 1.0 x 1.6 x 4 mm, 2:28 min. and (B) AIR™ Recon DL contrast-enhanced sagittal T1w, 0.9 x 1.2 x 4 mm, 2:08 min.
B-2
Figure 3.
L-spine comparing conventional reconstructed images with AIR™ Recon DL reconstructed images, demonstrating a 14% reduction in scan time, a 52% improvement in spatial resolution and 70% increase in SNR. (A) Conventional contrast-enhanced sagittal T1w, 1.0 x 1.6 x 4 mm, 2:28 min. and (B) AIR™ Recon DL contrast-enhanced sagittal T1w, 0.9 x 1.2 x 4 mm, 2:08 min.
A
Figure 4.
(A-C) Conventional FIESTA-C, 0.8 x 0.5 x 1 mm, 3:30 min. compared to (D-F) FIESTA-C with HyperSense, 0.8 x 0.5 x 1 mm, 2:55 min. showing nearly- identical image quality.
B
Figure 4.
(A-C) Conventional FIESTA-C, 0.8 x 0.5 x 1 mm, 3:30 min. compared to (D-F) FIESTA-C with HyperSense, 0.8 x 0.5 x 1 mm, 2:55 min. showing nearly- identical image quality.
C
Figure 4.
(A-C) Conventional FIESTA-C, 0.8 x 0.5 x 1 mm, 3:30 min. compared to (D-F) FIESTA-C with HyperSense, 0.8 x 0.5 x 1 mm, 2:55 min. showing nearly- identical image quality.
D
Figure 4.
(A-C) Conventional FIESTA-C, 0.8 x 0.5 x 1 mm, 3:30 min. compared to (D-F) FIESTA-C with HyperSense, 0.8 x 0.5 x 1 mm, 2:55 min. showing nearly- identical image quality.
E
Figure 4.
(A-C) Conventional FIESTA-C, 0.8 x 0.5 x 1 mm, 3:30 min. compared to (D-F) FIESTA-C with HyperSense, 0.8 x 0.5 x 1 mm, 2:55 min. showing nearly- identical image quality.
F
Figure 4.
(A-C) Conventional FIESTA-C, 0.8 x 0.5 x 1 mm, 3:30 min. compared to (D-F) FIESTA-C with HyperSense, 0.8 x 0.5 x 1 mm, 2:55 min. showing nearly- identical image quality.
A
Figure 5.
Thinner slices and smaller voxel size are acquired using MP- RAGE with PROMO. (A) Axial DWI, 1.6 x 1.4 x 5 mm, 50 sec.; (B) axial T2w, 0.5 x 0.8 x 5 mm, 1:15 min.; (C) axial T1w spin echo, 0.7 x 1.0 x 5 mm, 2:16 min.; (D) axial T2 FLAIR, 0.7 x 1 x 5 mm, 2:05 min.; and (E) contrast-enhanced sagittal T1w MP-RAGE with PROMO, 0.8 x 0.8 x 2 mm, 3:57 min.
B
Figure 5.
Thinner slices and smaller voxel size are acquired using MP- RAGE with PROMO. (A) Axial DWI, 1.6 x 1.4 x 5 mm, 50 sec.; (B) axial T2w, 0.5 x 0.8 x 5 mm, 1:15 min.; (C) axial T1w spin echo, 0.7 x 1.0 x 5 mm, 2:16 min.; (D) axial T2 FLAIR, 0.7 x 1 x 5 mm, 2:05 min.; and (E) contrast-enhanced sagittal T1w MP-RAGE with PROMO, 0.8 x 0.8 x 2 mm, 3:57 min.
C
Figure 5.
Thinner slices and smaller voxel size are acquired using MP- RAGE with PROMO. (A) Axial DWI, 1.6 x 1.4 x 5 mm, 50 sec.; (B) axial T2w, 0.5 x 0.8 x 5 mm, 1:15 min.; (C) axial T1w spin echo, 0.7 x 1.0 x 5 mm, 2:16 min.; (D) axial T2 FLAIR, 0.7 x 1 x 5 mm, 2:05 min.; and (E) contrast-enhanced sagittal T1w MP-RAGE with PROMO, 0.8 x 0.8 x 2 mm, 3:57 min.
D
Figure 5.
Thinner slices and smaller voxel size are acquired using MP- RAGE with PROMO. (A) Axial DWI, 1.6 x 1.4 x 5 mm, 50 sec.; (B) axial T2w, 0.5 x 0.8 x 5 mm, 1:15 min.; (C) axial T1w spin echo, 0.7 x 1.0 x 5 mm, 2:16 min.; (D) axial T2 FLAIR, 0.7 x 1 x 5 mm, 2:05 min.; and (E) contrast-enhanced sagittal T1w MP-RAGE with PROMO, 0.8 x 0.8 x 2 mm, 3:57 min.
E
Figure 5.
Thinner slices and smaller voxel size are acquired using MP- RAGE with PROMO. (A) Axial DWI, 1.6 x 1.4 x 5 mm, 50 sec.; (B) axial T2w, 0.5 x 0.8 x 5 mm, 1:15 min.; (C) axial T1w spin echo, 0.7 x 1.0 x 5 mm, 2:16 min.; (D) axial T2 FLAIR, 0.7 x 1 x 5 mm, 2:05 min.; and (E) contrast-enhanced sagittal T1w MP-RAGE with PROMO, 0.8 x 0.8 x 2 mm, 3:57 min.
previous arrow_PURPLE.svgnext arrow_PURPLE.svg
GE-HLTHCR_Standard_RGB-White.png
SIGNA_Pulse_Title.svg
arrow_forward_WHITE.svg
Subscribe Now
Manage Subscription
FOLLOW US
facebook.svglinkedin.svgtwitter X.svg
Contact Us • Cookie Preferences • Privacy Policy • California Privacy Policy
Do Not Sell or Share My Personal Information • Terms & Conditions • Security
© 2024 GE HealthCare. GE is a trademark of General Electric Company. Used under trademark license.
Hirokazu Fujiwara_c.jpg
Hirokazu Fujiwara, MD
Keio University Hospital Tokyo, Japan


SPOTLIGHT

Deep-learning technology delivers shorter exam times and better image quality

Hirokazu Fujiwara_c.jpg
Hirokazu Fujiwara, MD
Keio University Hospital Tokyo, Japan

As a regional referral center, Keio University Hospital in Tokyo, Japan, provides high-quality imaging services with some of the latest MR imaging technologies in the country. Recently, the radiology department implemented AIR™ Recon DL, GE Healthcare’s innovative reconstruction technology, to improve image quality and diagnostic confidence in neurological studies while reducing scan times.

Every day, Keio University Hospital welcomes around 3,000 outpatients from across the country, in addition to approximately 800 inpatients. The hospital sees more than 16,000 emergency patients annually, and doctors perform nearly 15,000 surgeries per year. In addition to being designated as an Advanced Treatment Hospital in Japan, the hospital is also responsible for contributing to the development of regional healthcare by coordinating personnel exchange with over 100 affiliated hospitals throughout Japan.
"We have a lot of patients in our hospital, therefore imaging exam times must be shortened. However, as a university hospital, we have many patients with complex or serious diseases requiring high-quality MR examinations," says Hirokazu Fujiwara, MD, PhD, Assistant Professor of Diagnostic Radiology at Keio University Hospital. This presents Dr. Fujiwara and his colleagues with a dilemma: they need the highest quality MR exam in the shortest amount of time. In some difficult or rare patient cases, they need to shorten a particular sequence in order to add sequences for a more confident diagnosis.
Faced with these pressures, the hospital installed the newest SIGNA™Works AIR™ IQ Edition software (PX29) with AIR™ Recon DL on its SIGNA™ Pioneer 3.0T MR system in November 2020. AIR™ Recon DL is an innovative reconstruction technology based on deep learning that delivers both high-quality images and shortened scan times. The resulting TrueFidelity™ MR images elevate the science of image reconstruction for clinical excellence without conventional compromises.
Shortened neuro scan times
Dr. Fujiwara says AIR™ Recon DL is helping clinicians at Keio shorten scan times for neurological studies while delivering the high-resolution images necessary in these critical cases. In the past, the hospital’s routine brain protocol used 5 mm slices for wholebrain imaging, but AIR™ Recon DL has enabled whole-brain imaging with just 3 mm slices (Figure 1).
A
B
C
D
E
F
G
H
Figure 1. Routine brain protocol comparing (top row) conventional reconstruction with 5 mm, 22 slices and (bottom row) AIR™ Recon DL with 3 mm, 42 slices. Conventional reconstructed images: (A) Axial T2w, 0.6 x 0.8 x 5 mm, 1:15 min.; (B) axial T2 FLAIR, 0.7 x 0.8 x 5 mm, 2:05 min; (C) axial T1w, 0.6 x 0.6 x 5 mm, 2:16 min; and (D) axial T2*w, 0.6 x 1 x 5 mm, 1:43 min. AIR™ Recon DL reconstructed images: (E) Axial T2w, 0.6 x 0.8 x 3 mm, 1:12 min; (F) axial T2 FLAIR, 0.7 x 0.8 x 3 mm, 2:33 min (G)axial T1w, 0.6 x 0.6 x 3 mm, 1:58 min; and (H) axial T2*w, 0.6 x 1 x 3 mm, 1:40 min.
"The thin slice of brain scans is more effective than high in-plane spatial resolution imaging because it reduces the partial volume effect. Thin-slice scans definitely improve detection of small lesions. There’s high clinical value to be able to obtain those images without extending the scan time," he says.
Other types of cases, such as stroke, require a reduced exam time. Traditionally, shortening the scan time required sacrificing SNR or spatial resolution. "By using AIR™ Recon DL, we shorten the scan times and still obtain excellent clinical images without degrading the image quality," Dr. Fujiwara adds.

He explains that he can choose any of the AIR™ Recon DL SNR improvement levels – low, medium or high – to achieve high-quality images.
"No matter what level is selected, the image resolution is improved. It’s not just an improvement in SNR. In the case of the conventional denoising algorithm, there is a concern about a decrease in sharpness. But AIR™ Recon DL brings a better sharpness to images, making it easier for us to visualize lesions."
Dr. Hirokazu Fujiwara
AIR™ Recon DL is suitable for all contrasts such as T2-weighted, T1-weighted, FLAIR and contrast-enhanced T1-weighted. In addition, Dr. Fujiwara expects the reduction of truncation, or Gibbs ringing, artifacts will help improve the diagnosis of spinal cord disease.
Balancing efficiency and quality
AIR™ Recon DL has helped the department provide MR exams for a variety of patients, as it is no longer necessary to choose between reducing scan time or higher resolution image quality. "Using AIR™ Recon DL, it’s possible to shorten the imaging time with the same image quality for diseases that require a rapid exam, such as in patients with a suspected stroke. On the other hand, in rare diseases, the shorter scan time for routine sequences can make it possible to facilitate additional imaging. In addition, we can make choices that improve spatial resolution in the same scan time," he explains.
"The benefit of noise reduction by AIR™ Recon DL is great. The improvement in spatial resolution can be clearly seen in the images, even without changing the parameters," says Dr. Fujiwara.
A
B
C
Figure 2. Intracerebral hemorrhage with AIR™ Recon DL. (A) Axial T2 FLAIR, 0.5 x 0.6 x 5 mm, 2:33 min; (B) axial T2w, 0.5 x 0.6 x 5 mm 1:05 min; and (C) axial T1w, 0.8 x 0.9 x 5 mm, 52 sec.
Figure 3. L-spine comparing conventional reconstructed images with AIR™ Recon DL reconstructed images, demonstrating a 14% reduction in scan time, a 52% improvement in spatial resolution and 70% increase in SNR. (A) Conventional contrast-enhanced sagittal T1w, 1.0 x 1.6 x 4 mm, 2:28 min. and (B) AIR™ Recon DL contrast-enhanced sagittal T1w, 0.9 x 1.2 x 4 mm, 2:08 min.
Figure 4. (A-C) Conventional FIESTA-C, 0.8 x 0.5 x 1 mm, 3:30 min. compared to (D-F) FIESTA-C with HyperSense, 0.8 x 0.5 x 1 mm, 2:55 min. showing nearly- identical image quality.
Figure 5. Thinner slices and smaller voxel size are acquired using MP- RAGE with PROMO. (A) Axial DWI, 1.6 x 1.4 x 5 mm, 50 sec.; (B) axial T2w, 0.5 x 0.8 x 5 mm, 1:15 min.; (C) axial T1w spin echo, 0.7 x 1.0 x 5 mm, 2:16 min.; (D) axial T2 FLAIR, 0.7 x 1 x 5 mm, 2:05 min.; and (E) contrast-enhanced sagittal T1w MP-RAGE with PROMO, 0.8 x 0.8 x 2 mm, 3:57 min.
During a spinal study with AIR™ Recon DL, the patient position was lateral decubitus using a contrast media and a fixed posterior array (PA). "We acquired an excellent spine image using AIR™ Recon DL," he says. "Normally, it’s not possible to obtain sufficient image quality due to the distance from the PA, but AIR™ Recon DL makes it possible to make a sufficient diagnosis."
HyperSense helps improve diagnostic confidence
The facility also implemented HyperSense, an acceleration technique based on sparse data compressibility allowing scan time reduction while maintaining SNR efficiency. Facial nerve images were obtained with FIESTA-C and HyperSense.
"Although HyperSense can use a shorter scan time than the conventional scan, there is no reduced SNR. Using it, we were able to acquire a detailed image in a short scan time of about 2 minutes."
Dr. Hirokazu Fujiwara
The radiology department has added PROMO, an advanced 3D motion correction technology for neuroimaging that helps eliminate the need for patients to lie motionless throughout the scan, making scanning less stressful for patients.
"PROMO has high clinical value because it can accommodate patients with unexpected movements and it does not extend imaging time in patients who don’t move. Therefore, I would like to incorporate this application into our routine examinations."
Dr. Hirokazu Fujiwara
In Dr. Fujiwara’s experience, AIR™ Recon DL has improved the depiction of anatomy and lesions to improve the quality of the study and the diagnosis, and boost diagnostic confidence.