Tyger

Appearance

Gadgeton examples ​

Gadgetron is an open source project for medical image reconstruction and can be used with Tyger. These examples assume that you have ismrmrd installed.

Basic example ​

Generate some test data with:

bash
ismrmrd_generate_cartesian_shepp_logan

This will generate a testdata.h5 file. You can then run a reconstruction with:

bash
ismrmrd_hdf5_to_stream -i testdata.h5 --use-stdout \
| tyger run exec -f  basic_gadgetron.yml \
| ismrmrd_stream_to_hdf5 --use-stdin -o out_basic.h5

basic_gadgetron.yml looks like this:

yaml
job:
  codespec:
    image: ghcr.io/gadgetron/gadgetron/gadgetron_ubuntu_rt_nocuda:latest
    buffers:
      inputs:
        - input
      outputs:
        - output
    args:
      - "-c"
      - "default.xml"
      - "--from_stream"
      - "-i"
      - "$(INPUT_PIPE)"
      - "-o"
      - "$(OUTPUT_PIPE)"
    resources:
      requests:
        cpu: 1

Distributed reconstruction ​

Next is an example that uses a distributed run. First, download the raw data:

bash
curl -OL https://aka.ms/tyger/docs/samples/binning.h5

Then run:

bash
ismrmrd_hdf5_to_stream -i binning.h5 --use-stdout \
| tyger run exec -f  distributed_gadgetron.yml --logs \
| ismrmrd_stream_to_hdf5 --use-stdin -o out_binning.h5

distributed_gadgetron.yml looks like this:

yaml
job:
  codespec:
    image: ghcr.io/gadgetron/gadgetron/gadgetron_ubuntu_rt_nocuda:latest
    buffers:
      inputs:
        - input
      outputs:
        - output
    args:
      - "-c"
      - "CMR_2DT_RTCine_KspaceBinning_Cloud.xml"
      - "--from_stream"
      - "-i"
      - "$(INPUT_PIPE)"
      - "-o"
      - "$(OUTPUT_PIPE)"
    env:
      GADGETRON_REMOTE_WORKER_COMMAND: printenv TYGER_GADGETRON_WORKER_ENDPOINT_ADDRESSES
    resources:
      requests:
        cpu: 1

worker:
  codespec:
    image: ghcr.io/gadgetron/gadgetron/gadgetron_ubuntu_rt_nocuda:latest
    args: []
    endpoints:
      gadgetron: 9002
    resources:
      requests:
        cpu: 3000m
        memory: 4G
      limits:
        memory: 4G
  replicas: 2

This run is made up of a job with two worker replicas.