/afids-CNN

Automatic localization and identification of salient brain landmarks in healthy and Parkinson's disease brains using U-Net architecture

Primary LanguageJupyter NotebookMIT LicenseMIT

afids-NN

Utilizing the anatomical fiducals framework to identify other salient brain regions and automatic localization of anatomical fiducials using neural networks

Processing data for training

Convert3D

Anatomical landmark data (AFIDs)

Convert3D:

  1. .fcsv -> threshold image -> landmark distance map (could be considered probability map)
  2. distance map used for training

Structural T1w imaging

Convert3D:

  1. brainmask.nii -> 3D patches sampled at x voxels
  2. matching of distance maps and anatomical imaging patches is crucial for proper training