Machine Learning-based Iterative Image Reconstruction Algorithm Allows Significant Reduction in Brain MRI Scan Times - Medicvision

ABSTRACTS & ARTICLES

Machine Learning-based Iterative Image Reconstruction Algorithm Allows Significant Reduction in Brain MRI Scan Times

L Tanenbaum , W Gibbs , B Johnson , I Varaganov , J Gomori , A Pais , T Aharoni , R Shreter
RadNet, Inc., new york, NY, Keck School of Medicine, University of South California, Los Angeles, CA, Center for
Diagnostic Imaging, St. Louis Park, MN, Rambam Health Care Campus,

Haifa, Israel, Hadassah Hebrew
University Medcial Center, Jerusalem, Israel, Medic Vision Imaging Solutions Ltd., Tirat-Carmel, Israel, Hillel Yaffe
Medical Center, Hadera, Israel. ASNR Annual Meeting, Vancouver B.C, June 2018.

 Authors:
L Tanenbaum , W Gibbs , B Johnson , I Varaganov , J Gomori , A Pais , T Aharoni , R Shreter

Institutions:
RadNet, Inc., new york, NY, Keck School of Medicine, University of South California, Los Angeles, CA, Center for
Diagnostic Imaging, St. Louis Park, MN, Rambam Health Care Campus, Haifa, Israel, Hadassah Hebrew
University Medcial Center, Jerusalem, Israel, Medic Vision Imaging Solutions Ltd., Tirat-Carmel, Israel, Hillel Yaffe
Medical Center, Hadera, Israel

Presenting Author:
- Lawrence N. Tanenbaum, MD, FACR
RadNet, Inc.

Purpose:
To evaluate image quality, grey-white (GW) matter differentiation and overall diagnostic sufficiency of brain MRI
images acquired with reduced scan time protocols, processed with a novel 3D image enhancement algorithm
("iQMR" by Medic Vision Ltd.) compared with images from sites' routine protocols.

Materials and Methods:
Under IRB-approval, thirty-seven subjects (mean age 48+/-15 years) were scanned on three 1.5T scanners (Ingenia-
Philips, n=18; Aera-Siemens, n=9; Signa-HDx-General Electric, n=10), at three different sites using the site's routine
clinical brain protocols as well as time-reduced variants (30% shorter, in average, than the sites' routine brain exam).
The short scans were prescribed by changing conventional acquisition parameters, such that the revised setting
traded scan time reductions for decreased signal to noise ratio. After offline processing with iQMR, images were
compared with the original-unprocessed images (155 scans) and with the corresponding routine protocols (153
scans). Independent, blinded, side-by-side comparisons of diagnostic quality, visual image quality, GW-matter
differentiation and presence of artifacts were performed by six neuroradiologists using a 5-point Likert-scale (3=
equal, >3 processed image is superior).


Results:
Reviewer assessments exhibited superiority of the processed images over the original-unprocessed images for
diagnostic quality (median=3, mode=3, mean=3.29+/-0.56), visual image quality (median=4, mode=4,
mean=3.7+/-0.63), GW-matter differentiation (median=3, mode=3, mean 3.26+/-0.61) and artifacts' appearance
(median=3, mode=3, mean=3.29+/-0.59) (n=850 reads).
Additionally, the processed time-reduced images demonstrated equality for diagnostic quality (median=3, mode=3,
mean=2.94+/-0.39), visual image quality (median=3, mode=3, mean=2.83+/-0.69), GW-matter differentiation
(median=3, mode=3, mean=2.91+/-0.46) and artifacts' appearance (median=3, mode=3, mean=2.91+/-0.54) (n=614
reads). Figure 1 shows comparable T1- and T2- weighted imaging of conventional and processed-short protocols
from a normal brain.


Conclusions:
iQMR processing can reduce routine brain exam scan time by 30% without adversely affecting image quality. These
findings may have substantial applications in the routine clinical practice by potentially enabling reduced scan time,
decreased motion artifacts and a lesser need for repeat scans.


Categories:
ADULT BRAIN, New Techniques/Post-processing
(https://files.aievolution.com/asn1801/abstracts/abs_2009/Figure1_asnr.jpg)


Reference One:
Bar-Aviv E, Devir Z, Dahan E et al. Denoising medical images. U.S. Patent 8605970, 2013

Reference Two:
Bar-Aviv E, Dahan E, Devir Z et al. Noise reduction of images. U.S. Patent 8582916, 2013

 

No result.

© 2017 Medic Vision. All Rights Reserved.

facebook Linkedin facebook