DTI-DeformIt: Generating ground-truth validation data for diffusion tensor image analysis tasks

Brian G. Booth, Ghassan Hamarneh

Abstract

We propose DTI-DeformIt: a framework to generate realistic synthetic datasets from a smaller number of, or even one, annotated image(s). Our approach extends the DeformIt technique of Hamarneh et al. [1] to handle the deformations and noise conditions of diffusion tensor images. An implementation of our proposed framework is also provided as a free download. We further show that DTI-DeformIt generates images that, according to eigenvector distance, are no different from real images than other real images, making them suitable for machine learning and validation.
BibTeX
@inproceedings{booth2014dti,
  author = {Brian G. Booth and Ghassan Hamarneh},
  title = {{DTI-DeformIt: Generating ground-truth validation data for diffusion tensor image analysis tasks}},
  booktitle = {ISBI},
  year = {2014},
  doi = {10.1109/isbi.2014.6867974},
}

Loading related content...

Figure