def slice_wise_fft(in_file, ftmask=None, spike_thres=3.0, out_prefix=None):
"""Search for spikes in slices using the 2D FFT"""
import os.path as op
import nibabel as nb
import numpy as np
from mriqc.workflows.utils import spectrum_mask
from scipy.ndimage import binary_erosion, generate_binary_structure
from scipy.ndimage.filters import median_filter
from statsmodels.robust.scale import mad
if out_prefix is None:
fname, ext = op.splitext(op.basename(in_file))
if ext == ".gz":
fname, _ = op.splitext(fname)
out_prefix = op.abspath(fname)
func_data = nb.load(in_file).get_data()
if ftmask is None:
ftmask = spectrum_mask(tuple(func_data.shape[:2]))
fft_data = []
for t in range(func_data.shape[-1]):
func_frame = func_data[..., t]
fft_slices = []
for z in range(func_frame.shape[2]):
sl = func_frame[..., z]
fftsl = (median_filter(
np.real(np.fft.fft2(sl)).astype(np.float32),
size=(5, 5),
mode="constant",
) * ftmask)
fft_slices.append(fftsl)
fft_data.append(np.stack(fft_slices, axis=-1))
# Recompose the 4D FFT timeseries
fft_data = np.stack(fft_data, -1)
# Z-score across t, using robust statistics
mu = np.median(fft_data, axis=3)
sigma = np.stack([mad(fft_data, axis=3)] * fft_data.shape[-1], -1)
idxs = np.where(np.abs(sigma) > 1e-4)
fft_zscored = fft_data - mu[..., np.newaxis]
fft_zscored[idxs] /= sigma[idxs]
# save fft z-scored
out_fft = op.abspath(out_prefix + "_zsfft.nii.gz")
nii = nb.Nifti1Image(fft_zscored.astype(np.float32), np.eye(4), None)
nii.to_filename(out_fft)
# Find peaks
spikes_list = []
for t in range(fft_zscored.shape[-1]):
fft_frame = fft_zscored[..., t]
for z in range(fft_frame.shape[-1]):
sl = fft_frame[..., z]
if np.all(sl < spike_thres):
continue
# Any zscore over spike_thres will be called a spike
sl[sl <= spike_thres] = 0
sl[sl > 0] = 1
# Erode peaks and see how many survive
struc = generate_binary_structure(2, 2)
sl = binary_erosion(sl.astype(np.uint8),
structure=struc).astype(np.uint8)
if sl.sum() > 10:
spikes_list.append((t, z))
out_spikes = op.abspath(out_prefix + "_spikes.tsv")
np.savetxt(out_spikes,
spikes_list,
fmt=b"%d",
delimiter=b"\t",
header="TR\tZ")
return len(spikes_list), out_spikes, out_fft
def slice_wise_fft(in_file, ftmask=None, spike_thres=3., out_prefix=None):
"""Search for spikes in slices using the 2D FFT"""
import os.path as op
import numpy as np
import nibabel as nb
from mriqc.workflows.utils import spectrum_mask
from scipy.ndimage.filters import median_filter
from scipy.ndimage import generate_binary_structure, binary_erosion
from statsmodels.robust.scale import mad
if out_prefix is None:
fname, ext = op.splitext(op.basename(in_file))
if ext == '.gz':
fname, _ = op.splitext(fname)
out_prefix = op.abspath(fname)
func_data = nb.load(in_file).get_data()
if ftmask is None:
ftmask = spectrum_mask(tuple(func_data.shape[:2]))
fft_data = []
for t in range(func_data.shape[-1]):
func_frame = func_data[..., t]
fft_slices = []
for z in range(func_frame.shape[2]):
sl = func_frame[..., z]
fftsl = median_filter(np.real(np.fft.fft2(sl)).astype(np.float32),
size=(5, 5), mode='constant') * ftmask
fft_slices.append(fftsl)
fft_data.append(np.stack(fft_slices, axis=-1))
# Recompose the 4D FFT timeseries
fft_data = np.stack(fft_data, -1)
# Z-score across t, using robust statistics
mu = np.median(fft_data, axis=3)
sigma = np.stack([mad(fft_data, axis=3)] * fft_data.shape[-1], -1)
idxs = np.where(np.abs(sigma) > 1e-4)
fft_zscored = fft_data - mu[..., np.newaxis]
fft_zscored[idxs] /= sigma[idxs]
# save fft z-scored
out_fft = op.abspath(out_prefix + '_zsfft.nii.gz')
nii = nb.Nifti1Image(fft_zscored.astype(np.float32), np.eye(4), None)
nii.to_filename(out_fft)
# Find peaks
spikes_list = []
for t in range(fft_zscored.shape[-1]):
fft_frame = fft_zscored[..., t]
for z in range(fft_frame.shape[-1]):
sl = fft_frame[..., z]
if np.all(sl < spike_thres):
continue
# Any zscore over spike_thres will be called a spike
sl[sl <= spike_thres] = 0
sl[sl > 0] = 1
# Erode peaks and see how many survive
struc = generate_binary_structure(2, 2)
sl = binary_erosion(sl.astype(np.uint8), structure=struc).astype(np.uint8)
if sl.sum() > 10:
spikes_list.append((t, z))
out_spikes = op.abspath(out_prefix + '_spikes.tsv')
np.savetxt(out_spikes, spikes_list, fmt=b'%d', delimiter=b'\t', header='TR\tZ')
return len(spikes_list), out_spikes, out_fft