Compare Fourier Model and T2* Model for Stack of Spirals trajectory

This examples walks through the elementary components of SNAKE.

Here we proceed step by step and use the Python interface. A more integrated alternative is to use the CLI snake-main

# Imports
from snake.core.simulation import SimConfig, default_hardware, GreConfig
from snake.core.phantom import Phantom
from snake.core.smaps import get_smaps
from snake.core.sampling import StackOfSpiralSampler

from mrinufft import get_operator


# For faster computation, try to use the GPU

NUFFT_BACKEND = "stacked-gpunufft"
COMPUTE_BACKEND = "cupy"

try:
    import cupy as cp

    if not cp.cupy.cuda.runtime.getDeviceCount():
        raise ValueError("No CUDA Device found")

    get_operator("stacked-gpunufft")
except Exception:
    try:
        get_operator("stacked-finufft")
    except ValueError as e:
        raise ValueError("No NUFFT backend available") from e

    NUFFT_BACKEND = "stacked-finufft"
    COMPUTE_BACKEND = "numpy"

print(
    f"Using NUFFT backend: {NUFFT_BACKEND}", f"Using Compute backend: {COMPUTE_BACKEND}"
)

Using NUFFT backend: stacked-finufft Using Compute backend: numpy

sim_conf = SimConfig(
    max_sim_time=3,
    seq=GreConfig(TR=50, TE=22, FA=12),
    hardware=default_hardware,
    fov_mm=(181, 217, 181),
    shape=(60, 72, 60),
)
sim_conf.hardware.n_coils = 1  # Update to get multi coil results.
sim_conf.hardware.field_strength = 7
phantom = Phantom.from_brainweb(sub_id=4, sim_conf=sim_conf, tissue_file="tissue_7T")

Traceback (most recent call last):
  File "/home/runner/work/snake-fmri/snake-fmri/examples/anatomical/example_gpu_anat_spirals.py", line 59, in <module>
    phantom = Phantom.from_brainweb(sub_id=4, sim_conf=sim_conf, tissue_file="tissue_7T")
  File "/home/runner/work/snake-fmri/snake-fmri/src/snake/core/phantom/static.py", line 112, in from_brainweb
    tissues_mask = get_mri(sub_id, contrast="fuzzy").astype(np.float32)
  File "/opt/hostedtoolcache/Python/3.10.16/x64/lib/python3.10/site-packages/brainweb_dl/mri.py", line 173, in get_mri
    data = _get_mri_sub20(
  File "/opt/hostedtoolcache/Python/3.10.16/x64/lib/python3.10/site-packages/brainweb_dl/mri.py", line 91, in _get_mri_sub20
    filename = get_brainweb20(
  File "/opt/hostedtoolcache/Python/3.10.16/x64/lib/python3.10/site-packages/brainweb_dl/_brainweb.py", line 329, in get_brainweb20
    Parallel(n_jobs=-1, backend="threading")(
  File "/opt/hostedtoolcache/Python/3.10.16/x64/lib/python3.10/site-packages/joblib/parallel.py", line 2007, in __call__
    return output if self.return_generator else list(output)
  File "/opt/hostedtoolcache/Python/3.10.16/x64/lib/python3.10/site-packages/joblib/parallel.py", line 1650, in _get_outputs
    yield from self._retrieve()
  File "/opt/hostedtoolcache/Python/3.10.16/x64/lib/python3.10/site-packages/joblib/parallel.py", line 1754, in _retrieve
    self._raise_error_fast()
  File "/opt/hostedtoolcache/Python/3.10.16/x64/lib/python3.10/site-packages/joblib/parallel.py", line 1789, in _raise_error_fast
    error_job.get_result(self.timeout)
  File "/opt/hostedtoolcache/Python/3.10.16/x64/lib/python3.10/site-packages/joblib/parallel.py", line 745, in get_result
    return self._return_or_raise()
  File "/opt/hostedtoolcache/Python/3.10.16/x64/lib/python3.10/site-packages/joblib/parallel.py", line 763, in _return_or_raise
    raise self._result
requests.exceptions.ConnectTimeout: HTTPConnectionPool(host='brainweb.bic.mni.mcgill.ca', port=80): Max retries exceeded with url: /cgi/brainweb1/?do_download_alias=subject04_bck&format_value=raw_short&zip_value=gnuzip&download_for_real=%5BStart+download%21%5D (Caused by ConnectTimeoutError(<urllib3.connection.HTTPConnection object at 0x7f2a349ac520>, 'Connection to brainweb.bic.mni.mcgill.ca timed out. (connect timeout=None)'))

Setting up Acquisition Pattern and Initializing Result file.

# The next piece of simulation is the acquisition trajectory.
# Here nothing fancy, we are using a stack of spiral, that samples a 3D
# k-space, with an acceleration factor AF=4 on the z-axis.

sampler = StackOfSpiralSampler(
    accelz=2,
    acsz=0.1,
    orderz="top-down",
    nb_revolutions=12,
    obs_time_ms=30,
    constant=True,
)

smaps = None
if sim_conf.hardware.n_coils > 1:
    smaps = get_smaps(sim_conf.shape, n_coils=sim_conf.hardware.n_coils)

The acquisition trajectory looks like this

traj = sampler.get_next_frame(sim_conf)
from mrinufft.trajectories.display import display_3D_trajectory

display_3D_trajectory(traj)

traj.shape

Adding noise in Image

from snake.core.handlers.noise import NoiseHandler

noise_handler = NoiseHandler(variance=0.01)

Acquisition with Cartesian Engine

The generated file example_EPI.mrd does not contains any k-space data for now, only the sampling trajectory. let’s put some in. In order to do so, we need to setup the acquisition engine that models the MR physics, and get sampled at the specified k-space trajectory.

SNAKE comes with two models for the MR Physics:

• model=”simple” :: Each k-space shot acquires a constant signal, which is the image contrast at TE.

• model=”T2s” :: Each k-space shot is degraded by the T2* decay induced by each tissue.

# Here we will use the "simple" model, which is faster.
#
# SNAKE's Engine are capable of simulating the data in parallel, by distributing
# the shots to be acquired to a set of processes. To do so , we need to specify
# the number of jobs that will run in parallel, as well as the size of a job.
# Setting the job size and the number of jobs can have a great impact on total
# runtime and memory consumption.
#
# Here, we have a single frame to acquire with 60 frames (one EPI per slice), so
# a single worker will do.

from snake.core.engine import NufftAcquisitionEngine

engine = NufftAcquisitionEngine(model="simple", snr=30000)

engine(
    "example_spiral.mrd",
    sampler,
    phantom,
    sim_conf,
    handlers=[noise_handler],
    smaps=smaps,
    worker_chunk_size=60,
    n_workers=1,
    nufft_backend=NUFFT_BACKEND,
)
engine_t2s = NufftAcquisitionEngine(model="T2s", snr=30000)

engine_t2s(
    "example_spiral_t2s.mrd",
    sampler,
    phantom,
    sim_conf,
    handlers=[noise_handler],
    worker_chunk_size=60,
    n_workers=1,
    nufft_backend=NUFFT_BACKEND,
)

Simple reconstruction

Getting k-space data is nice, but SNAKE also provides rudimentary reconstruction tools to get images (and check that we didn’t mess up the acquisition process). This is available in the companion package snake.toolkit.

Loading the .mrd file to retrieve all information can be done using the ismrmd python package, but SNAKE provides convenient dataloaders, which are more efficient, and take cares of managing underlying files access. As we are showcasing the API, we will do things manually here, and use only core SNAKE.

from snake.mrd_utils import NonCartesianFrameDataLoader
from snake.toolkit.reconstructors import (
    SequentialReconstructor,
    ZeroFilledReconstructor,
)

zer_rec = ZeroFilledReconstructor(
    nufft_backend=NUFFT_BACKEND, density_compensation=None
)
seq_rec = SequentialReconstructor(
    nufft_backend=NUFFT_BACKEND,
    density_compensation=None,
    max_iter_per_frame=30,
    threshold=2e-6,
    optimizer="fista",
    compute_backend=COMPUTE_BACKEND,
)
with NonCartesianFrameDataLoader("example_spiral.mrd") as data_loader:
    adjoint_spiral = abs(zer_rec.reconstruct(data_loader)[0])
    cs_spiral = abs(seq_rec.reconstruct(data_loader)[0])
with NonCartesianFrameDataLoader("example_spiral_t2s.mrd") as data_loader:
    adjoint_spiral_T2s = abs(zer_rec.reconstruct(data_loader,sim_conf)[0])
    cs_spiral_T2s = abs(seq_rec.reconstruct(data_loader)[0])

with NonCartesianFrameDataLoader("example_spiral.mrd") as data_loader:
    traj,data = data_loader.get_kspace_frame(0)

data.shape

Plotting the result

import matplotlib.pyplot as plt
from snake.toolkit.plotting import axis3dcut

fig, axs = plt.subplots(
    2,
    3,
    figsize=(19, 10),
    gridspec_kw=dict(wspace=0, hspace=0),
)


for ax, img, title in zip(
    axs[0],
    (adjoint_spiral, adjoint_spiral_T2s, abs(adjoint_spiral - adjoint_spiral_T2s)),
    ("simple", "T2s", "diff"),
):
    axis3dcut(img.T, None, None, cbar=True, cuts=(40, 40, 40), ax=ax,width_inches=4)
    ax.set_title(title)


for ax, img, title in zip(
    axs[1],
    (cs_spiral, cs_spiral_T2s, abs(cs_spiral - cs_spiral_T2s)),
    ("simple", "T2s", "diff"),
):
    axis3dcut(img.T, None, None, cbar=True, cuts=(40, 40, 40), ax=ax,width_inches=4)
    ax.set_title(title + " CS")


plt.show()