Abstract Details

Title Deep learning enables thalamic atrophy measurement on clinical quality T2 FLAIR images

Topic Multiple Sclerosis

Presentation(s) P5 - Poster Session 5 (5:30 PM-6:30 PM)

Poster/Presentation
Number 15-057

Objective To develop and validate a deep-learning approach to provide reliable thalamic measurement using only clinical routine T2-FLAIR MRI.

Background The thalamus is involved in many neurological functions and is a critical structure for examining neurodegeneration in MS and other diseases. However, a recent 33-center study confirmed that the type of 3D-T1 MR images required for high-resolution thalamic atrophy analysis are frequently unavailable in clinical routine.

Design/Methods Using a semantic segmentation convolutional neural network architecture, we developed a method called DeepGRAI (Deep Gray Rating via Artificial Intelligence). De-identified data from 160 subjects scanned on 3T MRI was used, including both high-resolution 3D T1 images and clinical-routine T2-FLAIR images. FSL’s FIRST was used to generate reference thalamic maps on the 3D T1 images. Then, these maps were co-registered to T2-FLAIR images and used to train the network with reference only to the T2-FLAIR images. The system was trained for 100 epochs on an NVidia Titan XP GPU. To assess the overall accuracy, we compared predicted segmentations to similarly prepared test data. We also evaluated our measure’s association to cognition using the symbol digit modalities test (SDMT), and compared it with both conventional SIENAX brain volumetry and standard FIRST thalamic measurement on 3D T1.

Results Overall agreement with FIRST was very good (r=0.824, p<0.001), and the approach was also able to handle both low and high atrophy cases. Association with SDMT was stronger for both FIRST-derived thalamic volume (r=0.578, p<0.001) and our DeepGRAI-derived thalamic volume (r=0.506, p<0.001) than whole brain volume (r=0.460,p<0.001).

Conclusions Thalamic volumetry on clinical-quality T2-FLAIR MRI is feasible via deep learning. Future work should focus on improvements to overall accuracy and generalizability, but these results demonstrate that large-scale latent databases can be leveraged to better understand thalamic dynamics in disease, and also remove a potential barrier to clinical translation.

Authors/Disclosures
Michael G. Dwyer III, MD, PhD (Buffalo Neurological Analysis Center) PRESENTER	Dr. Dwyer has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Bristol Myers Squibb. Dr. Dwyer has received personal compensation in the range of $10,000-$49,999 for serving as a Consultant for Keystone Heart, Ltd. Dr. Dwyer has received personal compensation in the range of $10,000-$49,999 for serving as a Consultant for Novartis. The institution of Dr. Dwyer has received research support from Novartis. The institution of Dr. Dwyer has received research support from Keystone Heart, Ltd. The institution of Dr. Dwyer has received research support from Bristol Myers Squibb. The institution of Dr. Dwyer has received research support from Roche.
Keith Carolus	No disclosure on file
Niels Bergsland (Buffalo Neuroimaging Analysis Center / State University of New York At Buffalo)	Prof. Bergsland has nothing to disclose.
Dejan Jakimovski, MD, PhD (Buffalo Neuroimaging Analysis Center, University at Buffalo)	Dr. Jakimovski has nothing to disclose.
Bianca Weinstock-Guttman, MD (Department of Neurology, University At Buffalo)	Dr. Weinstock-Guttman has received personal compensation in the range of $10,000-$49,999 for serving as a Consultant for Biogen. Dr. Weinstock-Guttman has received personal compensation in the range of $5,000-$9,999 for serving as a Consultant for Genentech. Dr. Weinstock-Guttman has received personal compensation in the range of $5,000-$9,999 for serving as a Consultant for Novartis . Dr. Weinstock-Guttman has received personal compensation in the range of $500-$4,999 for serving as a Consultant for EMD Serono. Dr. Weinstock-Guttman has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Abbvie. Dr. Weinstock-Guttman has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Genzyme &Sanofi. Dr. Weinstock-Guttman has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Janssen . Dr. Weinstock-Guttman has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Bayer. Dr. Weinstock-Guttman has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Horizon. Dr. Weinstock-Guttman has received personal compensation in the range of $500-$4,999 for serving on a Scientific Advisory or Data Safety Monitoring board for Genentech. Dr. Weinstock-Guttman has received personal compensation in the range of $0-$499 for serving on a Scientific Advisory or Data Safety Monitoring board for Novartis. Dr. Weinstock-Guttman has received personal compensation in the range of $5,000-$9,999 for serving on a Speakers Bureau for Biogen. Dr. Weinstock-Guttman has received personal compensation in the range of $500-$4,999 for serving on a Speakers Bureau for Janssen. Dr. Weinstock-Guttman has received personal compensation in the range of $0-$499 for serving as a Reviewer with NIH.
Ferdinand Schweser, PhD (SUNY University At Buffalo)	Dr. Schweser has received personal compensation in the range of $500-$4,999 for serving as an Editor, Associate Editor, or Editorial Advisory Board Member for Elsevier.
Ralph H. Benedict, PhD (University At Buffalo)	Dr. Benedict has received personal compensation in the range of $10,000-$49,999 for serving as a Consultant for Roche. Dr. Benedict has received personal compensation in the range of $5,000-$9,999 for serving as a Consultant for Sanofi. Dr. Benedict has received personal compensation in the range of $10,000-$49,999 for serving as a Consultant for Bristol Meyers Squibb. Dr. Benedict has received personal compensation in the range of $5,000-$9,999 for serving as a Consultant for Immunic Therapeutics. Dr. Benedict has received intellectual property interests from a discovery or technology relating to health care.
Robert Zivadinov, MD, PhD, FAAN (Buffalo Neuroimaging Analysis Center)	The institution of Dr. Zivadinov has received personal compensation in the range of $500-$4,999 for serving as a Consultant for BMS. The institution of Dr. Zivadinov has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Omnicuris. The institution of Dr. Zivadinov has received personal compensation in the range of $500-$4,999 for serving on a Scientific Advisory or Data Safety Monitoring board for Myrobalan. Dr. Zivadinov has received personal compensation in the range of $500-$4,999 for serving on a Speakers Bureau for Sanofi. Dr. Zivadinov has received personal compensation in the range of $10,000-$49,999 for serving on a Speakers Bureau for EMD Serono. Dr. Zivadinov has received personal compensation in the range of $5,000-$9,999 for serving on a Speakers Bureau for Bristol Myers Squibb. The institution of Dr. Zivadinov has received personal compensation in the range of $5,000-$9,999 for serving on a Speakers Bureau for Biogen.

��ɫ��

��ɫ��