Automated MRI Segmentation and Analysis Framework (AMSAF)

Further documentation contained here: https://www.gitbook.com/book/iancmcdonald/amsaf/details

Meta:

Goal: Automate the generation and testing of MRI segmentations

**High-level implementation: **Grid search over the Elastix parameter space, constrained by our prior intuition and experience.

Tools: Python, Elastix, SimpleElastix

Thoughts to keep in mind:

• Time constraints

• Each registration + segmentation takes a few minutes to complete

• Space constraints:

• Each image takes up a significant amount of space, sometimes up to 500MB

• Modularity:

• As this project evolves, we may replace components of this framework to fit new insights or tools. For instance, we should not write this in such a way that we could easily replace Elastix for another registration framework like ANTs

• Each piece is sufficiently complex so as to warrant component-level unit testing

• Abstraction and ease of use:

• People years from now might want to use this, so write with them in mind

• Abstract away low-level implementation details without sacrificing results

High-Level Components:

Segmenter

• Purpose: Automatically Segment MRI images
• Implementation: Registration-based segmentation parameterized by target and manually segmented reference images. This should be flexible enough to accommodate switching out registration tools and frameworks while keeping the code altering to a minimum.

Segmentation Evaluator

• Purpose: Evaluate segmentation results
• Implementation: Image subtraction or DICE similarity

Linker

• Purpose: Link individual components

Executor

• Set Linker values and execute segmentation

Component Tester

• Purpose: Test other framework components