Skip to content

Latest commit

 

History

History
43 lines (27 loc) · 1.57 KB

readme.md

File metadata and controls

43 lines (27 loc) · 1.57 KB

Generating 128 dimensional UDIPs (also referred to as ENDOs)

Running the UDIP pipeline

Change the directory to UDIP

python udip_pipeline.py -i [input] -m [modality] -c [ckpt] -d [device] -o [output]

Example

python udip_pipeline.py -i input.csv -m image_paths -c ckpts/T1.ckpt -d cuda:0 -o output

Arguments

  • -i or --input: (required) The input CSV file containing T1 / T2 MRI paths for linearly registered (MNI152 space) brain extracted MRI.
  • -m or --modality: (required) The column name in the CSV file containing the image paths.
  • -c or --ckpt: (required) The path to the checkpoint file.
  • -d or --device: (optional) The device to run on (with a default value of "cuda:0" if not specified).
  • -o or --output: (required) The output directory. Output contains csv containing 128 dimensional UDIPs, losses, and correlation heatmap.

Code walkthrough pipeline files

udip_dataset.py: Defines custom pytorch dataset

udip_model.py: Defining the model

udip_pipeline.py: Main file

ckpts: Download checkpoints here

Download ckpts in the ckpts directory

Download the checkpoint file from the provided link and place it in the ckpt directory: Checkpoints Download

Dummy data

Dummy data of random nifti images generated of size 182,218,182 (similar to MNI registered brain MRI) are provided to help understand how dataloader expects the data to be presented https://drive.google.com/drive/folders/1A_wKs5yhRs_c0BL0PEKWkQ2IrvRQncbE?usp=sharing