NA-MIC Project Weeks

Back to Projects List

Brain Tumor segmentation with Missing Data

Key Investigators

• Reuben Dorent (BWH, USA)
• Tina Kapur (BWH, USA)
• Sarah Frisken (BWH, USA)

Presenter location: In-person

Project Description

This project aims to create a Slicer extension that can automatically segment brain tumors in brain multi-parametric MRI, even in the presence of missing data.

This project will focus on two use cases where:

• all MR sequences (T1, contrast-enhanced T1, T2, FLAIR) are available
• only pre-contrast T1 and contrast-enhanced T1

The algorithm will not only segment the scans but also perform the required pre-processing steps (co-registration and skull-stripping).

Objective

1. Develop a Slicer module that can automatically perform brain tumor segmentation
2. Create a module that has the flexibility to handle two potential sets of input data
3. Integrate pre-processing steps for end-to-end inference
4. Validate the module with a subset of BraTS and clinical data

Approach and Plan

1. Train two combinations of nnUnet using the BraTS dataset.
2. Integrate the pre-trained nnUnet frameworks into Slicer using the TotalSegmentator Slicer plugin as a template
3. Leverage Slicer tools to perform the BraTS preprocessing steps
4. Collect clinical data for validation

Progress and Next Steps

• nnUnet have been trained for two combinations of input (ceT1 + T1 and ceT1 + T1 + T2 + FLAIR).
• Preprocessing has been implemented using base Slicer extensions.
• The extension has been tested on Windows (GPU and CPU) and MacOS (CPU).
• Next steps: clean the GitHub repository.

Illustrations

Background and References

No response