Brain_Mets_Classification

Table of Contents

1. About The Project
   • Built With
2. Roadmap
3. License

About The Project

The aim of this project is to write an ai that classifies brain metastases based on their primary cancers.

Built with

The following dependencies, libraries and ressources were used: HD-BET Tensorflow ...

Road Map

Work in Progress

• Train AIs (2D)

To-do

• Explore results

Done

• Acquire patient data
• Find correct sequences for each patient
• Preprocessing
  • Extract dicom metadata
  • Convert dicom to nifti
  • Extract brain
    • extract patients brain using HD-BET
    • compare HD-BET images with synthstrip images (chose HD-BET)
  • Fill holes
  • Binary Segment
  • Cropy images
  • Bias correction
  • Coregister images
  • Resample images
  • Z-score normalization
  • Merge images
• redo preprocessing using the brats-toolkit
• redo preprocessing with n4 bias correction (last time, I swear)
• segmentation
  • redo segmentation on n4 bias corrected files
  • manually adjust segmentation
• Build AIs
  • 3D CNN (entire brain) -> unfortunately unsuccessfull :/
    • Transfer ResNeXt architecture to 3D
    • custom scheduler
    • custom ai architecture (input: images, age, sex)
  • 2D CNNs
    • 2D CNN (only metastasis cutout)
    • 2D CNN* (transfer learning)
    • Vision Transformer (pretrained)