ANTs Ecosystem Overview: Toward Practical, Integrative Pattern Theory

Overall goal of ANTs Ecosystem: Enable interpretable, visualizable mapping of high-dimensional spaces, starting with images but extending to modalities such as psychometric, genetics, clinical measurements, etc.

• we will build towards this goal by demonstrating examples that build incrementally on each other

Schedule

Brief introductory material showcasing R and Python wrapping of ANTs functionality.

9:15 - 10: Special Intro to ANTs (for ICERM grads and postdocs)

• background (discussed in parallel with installation)

  • ITK and ANTs

  • ANTs history and current development strategy ( google scholar )

  • Information resources (GitHub, Sourceforge)

    • ANTsR core documentation here - see the "Articles" link
    • ANTsPy core documentation here
    • Example 2D neurodegenerative condition brain slices here and the associated study script here
    • Example 3D pediatric data here

• installation:

  • ANTsR

  • ANTsPy

  • python docker/binder:

    • https://mybinder.org/v2/gh/stnava/ANTsPyDocker/master
    • https://hub.docker.com/r/stnava/antspy

  • R docker/binder:

    • https://mybinder.org/v2/gh/stnava/ANTsRDocker/master
    • https://hub.docker.com/r/stnava/antsr

  • Joint R and python binder:

    • launch and go to "New" -> Rstudio -> File -> open -> ANTsX_R_Python.Rmd -> knit to html ( might need to allow popup windows )
    • docker image: https://hub.docker.com/r/stnava/antsrpy

  • to run containers:

    • docker pull containername
    • docker run -p 8888:8888 containername e.g. docker run -p 8888:8888 stnava/antsrpy:latest
    • follow the instructions to open the html file

  • for core ANTs: on linux and osx

these will evolve as our tutorial material matures.

10:30 - 12: Intro to ANTs for Medical Imaging

• Theoretical framework: "Integrative pattern theory"

• Discussion: Definition of images (physical space, transformation groups, pairwise mapping, groupwise mapping, etc.)

• Discussion and working examples:

  • mapping biomedical images for statistical analysis and quantification: image registration

  • labeling biomedical images for statistical analysis and quantification: image segmentation

  • template construction: toward statistical representations of image populations

  • joint label fusion for anatomical labeling

  • functional MRI quantification - time permitting https://github.com/stnava/structuralFunctionalJointRegistration

12 - 1:30: Lunch

1:30 - 3: Deep Learning and Statistical Approaches in Medical Imaging

• Discussion: Overview of ANTsRNet a collection of deep learning tools for biomedical image quantification

  • Tensorflow + Keras

  • pre-trained networks

• Discussion and working examples:

  • U-net segmentation with template-based augmentation

    • brain extraction

    • brain segmentation (whole image, patch-based)

    • tumor segmentation

  • Deep learning-based regression

    • super-resolution

    • Res-nets and other architectures

    • integrative brain mapping with deep learning as a tool

• Other possible topics (time permitting)

  • clustering

  • activation maps

  • deep feature maps

3:30 - 4:30: "Ask us about…” audience-motivated discussion of other topics

• Possible topics:

  • neuroimaging modalities: DTI, PET, ASL, BOLD fMRI, microscopy

  • non-human primate and other animal studies

  • other organs: lungs, heart

  • population variability: baby brains, neurodegeneration, stroke/lesions, etc.

  • integration of imaging, genetics, psychometrics, socioeconomic status etc.

  • role of these tools in industry as applications, scientific tools in treatment of disease, etc.