Meeting Banner
Abstract #1721

Data Augmentation with Generative Deep Learning for Automatic Bone Segmentation from Fat Fraction MRI

Nicholas Dwork1, Poonguzhali Elangovan2, Daniel O Connor3, Alex McManus4, Roland Krug5, Galateia J Kazakia5, Catherine M Jankowski6, and Julio Carballido-Gamio2
1Biomedical Informatics, University of Colorado Anschutz, Aurora, CO, United States, 2Radiology, University of Colorado Anschutz, Aurora, CO, United States, 3Mathematics and Statistics, University of San Francisco, San Francisco, CA, United States, 4Applied Mathematics, University of Colorado Boulder, Boulder, CA, United States, 5Radiology and Biomedical Imaging, University of California San Francisco, San Francisco, CA, United States, 6College of Nursing, University of Colorado Anschutz, Aurora, CO, United States

Synopsis

Keywords: Analysis/Processing, Bone, Segmentation, GAN, Proton Density Fat Fraction

Motivation: Measurements of bone marrow adiposity (BMA) from proton density fat fraction (PDFF) maps could improve our understanding of bone fragility. Automatic analysis requires segmentation of the bone(s) from these images.

Goal(s): To develop an accurate deep learning model for automatically segmenting bone from PDFF maps.

Approach: We use a cycleGAN to create an image translation network that converts CT images into synthetic fat fraction images. We augment the training set of the segmentation model with this synthetic data.

Results: We show that we are able to achieve a lower binary cross entropy loss when segmentation is used with the synthetic data augmentation.

Impact: With an automatic bone segmentation algorithm from fat fraction MR images, future work will conduct a thorough investigation into how bone marrow adiposity affects bone fragility.

How to access this content:

For one year after publication, abstracts and videos are only open to registrants of this annual meeting. Registrants should use their existing login information. Non-registrant access can be purchased via the ISMRM E-Library.

After one year, current ISMRM & ISMRT members get free access to both the abstracts and videos. Non-members and non-registrants must purchase access via the ISMRM E-Library.

After two years, the meeting proceedings (abstracts) are opened to the public and require no login information. Videos remain behind password for access by members, registrants and E-Library customers.

Click here for more information on becoming a member.

Keywords