def get_maps(prior,
prior_list,
std_list,
CR_std_maps,
VE_list,
tdof_list,
mask_file,
smoothing=False):
volume_indices = np.asarray(nb.load(mask_file).dataobj).astype(bool)
Y = np.array(prior_list)
average = Y.mean(axis=0)
# compute MAD to be resistant to outliers
mad = np.median(np.abs(Y - np.median(Y, axis=0)), axis=0)
#network_var=Y.std(axis=0)
network_var = mad
X = np.array(std_list)
corr_map_std = elementwise_spearman(X, Y)
X = np.array(CR_std_maps)
corr_map_CR_std = elementwise_spearman(X, Y)
X = np.array(VE_list)
corr_map_VE = elementwise_spearman(X, Y)
# tdof effect; if there's no variability don't compute
if np.array(tdof_list).std() == 0:
corr_map_tdof = None
else:
tdof = np.array(tdof_list).reshape(-1, 1)
corr_map_tdof = elementwise_spearman(tdof, Y)
if smoothing:
prior = np.array(
nilearn.image.smooth_img(recover_3D(mask_file, prior),
0.3).dataobj)[volume_indices]
average = np.array(
nilearn.image.smooth_img(recover_3D(mask_file, average),
0.3).dataobj)[volume_indices]
corr_map_std = np.array(
nilearn.image.smooth_img(recover_3D(mask_file, corr_map_std),
0.3).dataobj)[volume_indices]
corr_map_CR_std = np.array(
nilearn.image.smooth_img(recover_3D(mask_file, corr_map_CR_std),
0.3).dataobj)[volume_indices]
corr_map_VE = np.array(
nilearn.image.smooth_img(recover_3D(mask_file, corr_map_VE),
0.3).dataobj)[volume_indices]
if np.array(tdof_list).std() == 0:
corr_map_tdof = None
else:
corr_map_tdof = np.array(
nilearn.image.smooth_img(recover_3D(mask_file, corr_map_tdof),
0.3).dataobj)[volume_indices]
return prior, average, network_var, corr_map_std, corr_map_CR_std, corr_map_VE, corr_map_tdof
def spatial_external_formating(spatial_info, file_dict):
import os
import pathlib # Better path manipulation
from rabies.analysis_pkg import analysis_functions
mask_file = file_dict['brain_mask_file']
bold_file = file_dict['bold_file']
filename_split = pathlib.Path(bold_file).name.rsplit(".nii")
# calculate STD and tSNR map on preprocessed timeseries
std_filename = os.path.abspath(filename_split[0] + '_tSTD.nii.gz')
analysis_functions.recover_3D(
mask_file, spatial_info['temporal_std']).to_filename(std_filename)
GS_corr_filename = os.path.abspath(filename_split[0] + '_GS_corr.nii.gz')
analysis_functions.recover_3D(
mask_file, spatial_info['GS_corr']).to_filename(GS_corr_filename)
DVARS_corr_filename = os.path.abspath(filename_split[0] +
'_DVARS_corr.nii.gz')
analysis_functions.recover_3D(
mask_file, spatial_info['DVARS_corr']).to_filename(DVARS_corr_filename)
FD_corr_filename = os.path.abspath(filename_split[0] + '_FD_corr.nii.gz')
analysis_functions.recover_3D(
mask_file, spatial_info['FD_corr']).to_filename(FD_corr_filename)
DR_maps_filename = os.path.abspath(filename_split[0] + '_DR_maps.nii.gz')
analysis_functions.recover_3D_multiple(
mask_file, spatial_info['DR_all']).to_filename(DR_maps_filename)
if len(spatial_info['prior_modeling_maps']) > 0:
import numpy as np
prior_modeling_filename = os.path.abspath(filename_split[0] +
'_prior_modeling.nii.gz')
analysis_functions.recover_3D_multiple(
mask_file,
np.array(spatial_info['prior_modeling_maps'])).to_filename(
prior_modeling_filename)
else:
prior_modeling_filename = None
return std_filename, GS_corr_filename, DVARS_corr_filename, FD_corr_filename, DR_maps_filename, prior_modeling_filename
def scan_diagnosis(bold_file, mask_file_dict, temporal_info, spatial_info, CR_data_dict, regional_grayplot=False):
template_file = mask_file_dict['template_file']
fig = plt.figure(figsize=(6, 18))
#fig.suptitle(name, fontsize=30, color='white')
ax0 = fig.add_subplot(3,1,1)
ax1 = fig.add_subplot(12,1,5)
ax1_ = fig.add_subplot(12,1,6)
ax2 = fig.add_subplot(6,1,4)
ax3 = fig.add_subplot(6,1,5)
ax4 = fig.add_subplot(6,1,6)
# disable function
regional_grayplot=False
if regional_grayplot:
from mpl_toolkits.axes_grid1 import make_axes_locatable
divider = make_axes_locatable(ax0)
im, slice_alt, region_mask_label = grayplot_regional(
bold_file, mask_file_dict, fig, ax0)
ax0.yaxis.labelpad = 40
ax_slice = divider.append_axes('left', size='5%', pad=0.0)
ax_label = divider.append_axes('left', size='5%', pad=0.0)
ax_slice.imshow(slice_alt.reshape(-1, 1), cmap='gray',
vmin=0, vmax=1.1, aspect='auto')
ax_label.imshow(region_mask_label.reshape(-1, 1),
cmap='Spectral', aspect='auto')
ax_slice.axis('off')
ax_label.axis('off')
else:
im = grayplot(bold_file, mask_file_dict, fig, ax0)
ax0.set_ylabel('Voxels', fontsize=20)
ax0.spines['right'].set_visible(False)
ax0.spines['top'].set_visible(False)
ax0.spines['bottom'].set_visible(False)
ax0.spines['left'].set_visible(False)
ax0.axes.get_yaxis().set_ticks([])
plt.setp(ax1.get_xticklabels(), visible=False)
y = temporal_info['FD_trace'].to_numpy()
x = range(len(y))
ax0.set_xlim([0, len(y)-1])
ax1.set_xlim([0, len(y)-1])
ax1_.set_xlim([0, len(y)-1])
ax2.set_xlim([0, len(y)-1])
ax3.set_xlim([0, len(y)-1])
ax4.set_xlim([0, len(y)-1])
# plot the motion timecourses
confounds_csv = CR_data_dict['confounds_csv']
time_range = CR_data_dict['time_range']
frame_mask = CR_data_dict['frame_mask']
df = pd.read_csv(confounds_csv)
# take proper subset of timepoints
ax1.plot(x,df['mov1'].to_numpy()[time_range][frame_mask])
ax1.plot(x,df['mov2'].to_numpy()[time_range][frame_mask])
ax1.plot(x,df['mov3'].to_numpy()[time_range][frame_mask])
ax1.legend(['translation 1', 'translation 2', 'translation 3'],
loc='center left', bbox_to_anchor=(1.15, 0.5))
ax1.spines['right'].set_visible(False)
ax1.spines['top'].set_visible(False)
plt.setp(ax1.get_xticklabels(), visible=False)
ax1_.plot(x,df['rot1'].to_numpy()[time_range][frame_mask])
ax1_.plot(x,df['rot2'].to_numpy()[time_range][frame_mask])
ax1_.plot(x,df['rot3'].to_numpy()[time_range][frame_mask])
ax1_.legend(['rotation 1', 'rotation 2', 'rotation 3'],
loc='center left', bbox_to_anchor=(1.15, 0.5))
plt.setp(ax1_.get_xticklabels(), visible=False)
ax1_.spines['right'].set_visible(False)
ax1_.spines['top'].set_visible(False)
y = temporal_info['FD_trace'].to_numpy()
ax2.plot(x,y, 'r')
ax2.set_ylabel('FD in mm', fontsize=15)
DVARS = temporal_info['DVARS']
DVARS[0] = None
ax2_ = ax2.twinx()
y2 = DVARS
ax2_.plot(x,y2, 'b')
ax2_.set_ylabel('DVARS', fontsize=15)
ax2.spines['right'].set_visible(False)
ax2.spines['top'].set_visible(False)
ax2_.spines['top'].set_visible(False)
plt.setp(ax2.get_xticklabels(), visible=False)
plt.setp(ax2_.get_xticklabels(), visible=False)
ax2.legend(['Framewise \nDisplacement (FD)'
], loc='center left', bbox_to_anchor=(1.15, 0.6))
ax2_.legend(['DVARS'
], loc='center left', bbox_to_anchor=(1.15, 0.4))
ax3.plot(x,temporal_info['edge_trace'])
ax3.plot(x,temporal_info['WM_trace'])
ax3.plot(x,temporal_info['CSF_trace'])
ax3.plot(x,temporal_info['predicted_time'])
ax3.set_ylabel('Mask L2-norm', fontsize=15)
ax3_ = ax3.twinx()
ax3_.plot(x,temporal_info['VE_temporal'], 'darkviolet')
ax3_.set_ylabel('CR R^2', fontsize=15)
ax3_.spines['right'].set_visible(False)
ax3_.spines['top'].set_visible(False)
plt.setp(ax3_.get_xticklabels(), visible=False)
ax3.legend(['Edge Mask', 'WM Mask', 'CSF Mask', 'CR prediction'
], loc='center left', bbox_to_anchor=(1.15, 0.7))
ax3_.legend(['CR R^2'
], loc='center left', bbox_to_anchor=(1.15, 0.3))
ax3_.set_ylim([0,1])
y = temporal_info['signal_trace']
ax4.plot(x,y)
ax4.plot(x,temporal_info['noise_trace'])
ax4.legend(['BOLD components', 'Confound components'
], loc='center left', bbox_to_anchor=(1.15, 0.5))
ax4.spines['right'].set_visible(False)
ax4.spines['top'].set_visible(False)
ax4.set_xlabel('Timepoint', fontsize=15)
ax4.set_ylabel('Abs. Beta \ncoefficients (Avg.)', fontsize=15)
dr_maps = spatial_info['DR_BOLD']
mask_file = mask_file_dict['brain_mask']
nrows = 6+dr_maps.shape[0]
fig2, axes2 = plt.subplots(nrows=nrows, ncols=3, figsize=(12*3, 2*nrows))
plt.tight_layout()
from rabies.visualization import otsu_scaling, plot_3d
axes = axes2[0, :]
scaled = otsu_scaling(template_file)
plot_3d(axes, scaled, fig2, vmin=0, vmax=1,
cmap='gray', alpha=1, cbar=False, num_slices=6)
temporal_std = spatial_info['temporal_std']
analysis_functions.recover_3D(
mask_file, temporal_std).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
# select vmax at 95th percentile value
vector = temporal_std.flatten()
vector.sort()
vmax = vector[int(len(vector)*0.95)]
cbar_list = plot_3d(axes, sitk_img, fig2, vmin=0, vmax=vmax,
cmap='inferno', alpha=1, cbar=True, num_slices=6)
for cbar in cbar_list:
cbar.ax.get_yaxis().labelpad = 35
cbar.set_label('Standard \n Deviation', fontsize=15, rotation=270, color='white')
for ax in axes:
ax.set_title('BOLD-Temporal s.d.', fontsize=30, color='white')
axes = axes2[1, :]
scaled = otsu_scaling(template_file)
plot_3d(axes, scaled, fig2, vmin=0, vmax=1,
cmap='gray', alpha=1, cbar=False, num_slices=6)
predicted_std = spatial_info['predicted_std']
analysis_functions.recover_3D(
mask_file, predicted_std).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
# select vmax at 95th percentile value
vector = predicted_std.flatten()
vector.sort()
vmax = vector[int(len(vector)*0.95)]
cbar_list = plot_3d(axes, sitk_img, fig2, vmin=0, vmax=vmax,
cmap='inferno', alpha=1, cbar=True, num_slices=6)
for cbar in cbar_list:
cbar.ax.get_yaxis().labelpad = 35
cbar.set_label('Standard \n Deviation', fontsize=15, rotation=270, color='white')
for ax in axes:
ax.set_title('CR-Temporal s.d.', fontsize=30, color='white')
axes = axes2[2, :]
plot_3d(axes, scaled, fig2, vmin=0, vmax=1,
cmap='gray', alpha=1, cbar=False, num_slices=6)
analysis_functions.recover_3D(
mask_file, spatial_info['VE_spatial']).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
cbar_list = plot_3d(axes, sitk_img, fig2, vmin=0, vmax=1, cmap='inferno',
alpha=1, cbar=True, threshold=0.1, num_slices=6)
for cbar in cbar_list:
cbar.ax.get_yaxis().labelpad = 20
cbar.set_label('CR R^2', fontsize=15, rotation=270, color='white')
for ax in axes:
ax.set_title('CR R^2', fontsize=30, color='white')
axes = axes2[3, :]
plot_3d(axes, scaled, fig2, vmin=0, vmax=1,
cmap='gray', alpha=1, cbar=False, num_slices=6)
analysis_functions.recover_3D(
mask_file, spatial_info['GS_corr']).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
cbar_list = plot_3d(axes, sitk_img, fig2, vmin=-1, vmax=1, cmap='cold_hot',
alpha=1, cbar=True, threshold=0.1, num_slices=6)
for cbar in cbar_list:
cbar.ax.get_yaxis().labelpad = 20
cbar.set_label("Pearson's' r", fontsize=15, rotation=270, color='white')
for ax in axes:
ax.set_title('Global Signal Correlation', fontsize=30, color='white')
axes = axes2[4, :]
plot_3d(axes, scaled, fig2, vmin=0, vmax=1,
cmap='gray', alpha=1, cbar=False, num_slices=6)
analysis_functions.recover_3D(
mask_file, spatial_info['DVARS_corr']).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
cbar_list = plot_3d(axes, sitk_img, fig2, vmin=-1, vmax=1, cmap='cold_hot',
alpha=1, cbar=True, threshold=0.1, num_slices=6)
for cbar in cbar_list:
cbar.ax.get_yaxis().labelpad = 20
cbar.set_label("Pearson's' r", fontsize=15, rotation=270, color='white')
for ax in axes:
ax.set_title('DVARS Correlation', fontsize=30, color='white')
axes = axes2[5, :]
plot_3d(axes, scaled, fig2, vmin=0, vmax=1,
cmap='gray', alpha=1, cbar=False, num_slices=6)
analysis_functions.recover_3D(
mask_file, spatial_info['FD_corr']).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
cbar_list = plot_3d(axes, sitk_img, fig2, vmin=-1, vmax=1, cmap='cold_hot',
alpha=1, cbar=True, threshold=0.1, num_slices=6)
for cbar in cbar_list:
cbar.ax.get_yaxis().labelpad = 20
cbar.set_label("Pearson's' r", fontsize=15, rotation=270, color='white')
for ax in axes:
ax.set_title('FD Correlation', fontsize=30, color='white')
for i in range(dr_maps.shape[0]):
axes = axes2[i+6, :]
analysis_functions.recover_3D(
mask_file, dr_maps[i, :]).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
cbar_list = masked_plot(fig2,axes, sitk_img, scaled, percentile=0.015, vmax=None)
for cbar in cbar_list:
cbar.ax.get_yaxis().labelpad = 35
cbar.set_label("Beta \nCoefficient", fontsize=15, rotation=270, color='white')
for ax in axes:
ax.set_title(f'BOLD component {i}', fontsize=30, color='white')
return fig, fig2
def spatial_external_formating(spatial_info, file_dict):
import os
import pathlib # Better path manipulation
from rabies.analysis_pkg import analysis_functions
mask_file = file_dict['mask_file']
bold_file = file_dict['bold_file']
filename_split = pathlib.Path(
bold_file).name.rsplit(".nii")
VE_filename = os.path.abspath(filename_split[0]+'_VE.nii.gz')
analysis_functions.recover_3D(
mask_file, spatial_info['VE_spatial']).to_filename(VE_filename)
std_filename = os.path.abspath(filename_split[0]+'_tSTD.nii.gz')
analysis_functions.recover_3D(
mask_file, spatial_info['temporal_std']).to_filename(std_filename)
predicted_std_filename = os.path.abspath(filename_split[0]+'_predicted_std.nii.gz')
analysis_functions.recover_3D(
mask_file, spatial_info['predicted_std']).to_filename(predicted_std_filename)
GS_corr_filename = os.path.abspath(filename_split[0]+'_GS_corr.nii.gz')
analysis_functions.recover_3D(
mask_file, spatial_info['GS_corr']).to_filename(GS_corr_filename)
DVARS_corr_filename = os.path.abspath(
filename_split[0]+'_DVARS_corr.nii.gz')
analysis_functions.recover_3D(
mask_file, spatial_info['DVARS_corr']).to_filename(DVARS_corr_filename)
FD_corr_filename = os.path.abspath(filename_split[0]+'_FD_corr.nii.gz')
analysis_functions.recover_3D(
mask_file, spatial_info['FD_corr']).to_filename(FD_corr_filename)
DR_maps_filename = os.path.abspath(filename_split[0]+'_DR_maps.nii.gz')
analysis_functions.recover_3D_multiple(
mask_file, spatial_info['DR_all']).to_filename(DR_maps_filename)
if len(spatial_info['dual_ICA_maps']) > 0:
import numpy as np
dual_ICA_filename = os.path.abspath(
filename_split[0]+'_dual_ICA.nii.gz')
analysis_functions.recover_3D_multiple(mask_file, np.array(
spatial_info['dual_ICA_maps'])).to_filename(dual_ICA_filename)
else:
dual_ICA_filename = None
return VE_filename, std_filename, predicted_std_filename, GS_corr_filename, DVARS_corr_filename, FD_corr_filename, DR_maps_filename, dual_ICA_filename
def scan_diagnosis(bold_file,
mask_file_dict,
temporal_info,
spatial_info,
confounds_csv,
regional_grayplot=False):
template_file = mask_file_dict['template_file']
from mpl_toolkits.axes_grid1 import make_axes_locatable
fig, axCenter = plt.subplots(figsize=(6, 18))
fig.subplots_adjust(.2, .1, .8, .95)
divider = make_axes_locatable(axCenter)
ax1 = divider.append_axes('bottom', size='25%', pad=0.5)
ax1_ = divider.append_axes('bottom', size='25%', pad=0.1)
ax2 = divider.append_axes('bottom', size='50%', pad=0.5)
ax3 = divider.append_axes('bottom', size='50%', pad=0.5)
ax4 = divider.append_axes('bottom', size='50%', pad=0.5)
if regional_grayplot:
im, slice_alt, region_mask_label = grayplot_regional(
bold_file, mask_file_dict, fig, axCenter)
axCenter.yaxis.labelpad = 40
ax_slice = divider.append_axes('left', size='5%', pad=0.0)
ax_label = divider.append_axes('left', size='5%', pad=0.0)
ax_slice.imshow(slice_alt.reshape(-1, 1),
cmap='gray',
vmin=0,
vmax=1.1,
aspect='auto')
ax_label.imshow(region_mask_label.reshape(-1, 1),
cmap='Spectral',
aspect='auto')
ax_slice.axis('off')
ax_label.axis('off')
else:
im = grayplot(bold_file, mask_file_dict, fig, axCenter)
axCenter.set_ylabel('Voxels', fontsize=20)
axCenter.spines['right'].set_visible(False)
axCenter.spines['top'].set_visible(False)
axCenter.spines['bottom'].set_visible(False)
axCenter.spines['left'].set_visible(False)
axCenter.axes.get_yaxis().set_ticks([])
plt.setp(axCenter.get_xticklabels(), visible=False)
# plot the motion timecourses
import pandas as pd
df = pd.read_csv(confounds_csv)
ax1.plot(df['mov1'])
ax1.plot(df['mov2'])
ax1.plot(df['mov3'])
ax1.legend(['translation 1', 'translation 2', 'translation 3'],
loc='center left',
bbox_to_anchor=(1, 0.5))
ax1.spines['right'].set_visible(False)
ax1.spines['top'].set_visible(False)
plt.setp(ax1.get_xticklabels(), visible=False)
ax1_.plot(df['rot1'])
ax1_.plot(df['rot2'])
ax1_.plot(df['rot3'])
ax1_.legend(['rotation 1', 'rotation 2', 'rotation 3'],
loc='center left',
bbox_to_anchor=(1, 0.5))
plt.setp(ax1_.get_xticklabels(), visible=False)
ax1_.spines['right'].set_visible(False)
ax1_.spines['top'].set_visible(False)
#ax1.set_title(name, fontsize=15)
y = temporal_info['FD_trace']
ax2.plot(y, 'r')
ax2.set_xlim([0, len(y)])
ax2.legend(['Framewise Displacement (FD)'], loc='upper right')
ax2.spines['right'].set_visible(False)
ax2.spines['top'].set_visible(False)
ax2.set_ylim([0.0, 0.1])
plt.setp(ax2.get_xticklabels(), visible=False)
DVARS = temporal_info['DVARS']
DVARS[0] = None
y = DVARS
ax3.plot(y)
ax3.set_xlim([0, len(y)])
ax3.plot(temporal_info['edge_trace'])
ax3.plot(temporal_info['WM_trace'])
ax3.plot(temporal_info['CSF_trace'])
ax3.plot(temporal_info['VE_temporal'])
ax3.legend(['DVARS', 'Edge Mask', 'WM Mask', 'CSF Mask', 'CR R^2'],
loc='center left',
bbox_to_anchor=(1, 0.5))
ax3.spines['right'].set_visible(False)
ax3.spines['top'].set_visible(False)
ax3.set_ylim([0.0, 1.5])
plt.setp(ax3.get_xticklabels(), visible=False)
y = temporal_info['signal_trace']
ax4.plot(y)
ax4.set_xlim([0, len(y)])
ax4.plot(temporal_info['noise_trace'])
ax4.legend(['BOLD components', 'Confound components'], loc='upper right')
ax4.spines['right'].set_visible(False)
ax4.spines['top'].set_visible(False)
ax4.set_ylim([0.0, 4.0])
ax4.set_xlabel('Timepoint', fontsize=15)
dr_maps = spatial_info['DR_BOLD']
mask_file = mask_file_dict['brain_mask']
nrows = 5 + dr_maps.shape[0]
fig2, axes2 = plt.subplots(nrows=nrows,
ncols=3,
figsize=(12 * 3, 2 * nrows))
plt.tight_layout()
from rabies.preprocess_pkg.preprocess_visual_QC import plot_3d, otsu_scaling
axes = axes2[0, :]
scaled = otsu_scaling(template_file)
plot_3d(axes,
scaled,
fig2,
vmin=0,
vmax=1,
cmap='gray',
alpha=1,
cbar=False,
num_slices=6)
analysis_functions.recover_3D(
mask_file, spatial_info['temporal_std']).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
plot_3d(axes,
sitk_img,
fig2,
vmin=0,
vmax=1,
cmap='inferno',
alpha=1,
cbar=True,
num_slices=6)
for ax in axes:
ax.set_title('Temporal STD', fontsize=25)
axes = axes2[1, :]
plot_3d(axes,
scaled,
fig2,
vmin=0,
vmax=1,
cmap='gray',
alpha=1,
cbar=False,
num_slices=6)
analysis_functions.recover_3D(
mask_file, spatial_info['VE_spatial']).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
plot_3d(axes,
sitk_img,
fig2,
vmin=0,
vmax=1,
cmap='inferno',
alpha=1,
cbar=True,
threshold=0.1,
num_slices=6)
for ax in axes:
ax.set_title('CR R^2', fontsize=25)
axes = axes2[2, :]
plot_3d(axes,
scaled,
fig2,
vmin=0,
vmax=1,
cmap='gray',
alpha=1,
cbar=False,
num_slices=6)
analysis_functions.recover_3D(
mask_file, spatial_info['GS_corr']).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
plot_3d(axes,
sitk_img,
fig2,
vmin=-1,
vmax=1,
cmap='cold_hot',
alpha=1,
cbar=True,
threshold=0.1,
num_slices=6)
for ax in axes:
ax.set_title('Global Signal Correlation', fontsize=25)
axes = axes2[3, :]
plot_3d(axes,
scaled,
fig2,
vmin=0,
vmax=1,
cmap='gray',
alpha=1,
cbar=False,
num_slices=6)
analysis_functions.recover_3D(
mask_file, spatial_info['DVARS_corr']).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
plot_3d(axes,
sitk_img,
fig2,
vmin=-1,
vmax=1,
cmap='cold_hot',
alpha=1,
cbar=True,
threshold=0.1,
num_slices=6)
for ax in axes:
ax.set_title('DVARS Correlation', fontsize=25)
axes = axes2[4, :]
plot_3d(axes,
scaled,
fig2,
vmin=0,
vmax=1,
cmap='gray',
alpha=1,
cbar=False,
num_slices=6)
analysis_functions.recover_3D(
mask_file, spatial_info['FD_corr']).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
plot_3d(axes,
sitk_img,
fig2,
vmin=-1,
vmax=1,
cmap='cold_hot',
alpha=1,
cbar=True,
threshold=0.1,
num_slices=6)
for ax in axes:
ax.set_title('FD Correlation', fontsize=25)
for i in range(dr_maps.shape[0]):
axes = axes2[i + 5, :]
plot_3d(axes,
scaled,
fig2,
vmin=0,
vmax=1,
cmap='gray',
alpha=1,
cbar=False,
num_slices=6)
analysis_functions.recover_3D(
mask_file, dr_maps[i, :]).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
plot_3d(axes,
sitk_img,
fig2,
vmin=-1,
vmax=1,
cmap='cold_hot',
alpha=1,
cbar=True,
threshold=0.1,
num_slices=6)
for ax in axes:
ax.set_title('BOLD component %s' % (i), fontsize=25)
return fig, fig2
def _run_interface(self, runtime):
from rabies.preprocess_pkg.utils import flatten_list
merged = flatten_list(list(self.inputs.spatial_info_list))
if len(merged) < 3:
import logging
log = logging.getLogger('root')
log.warning(
"Cannot run statistics on a sample size smaller than 3, so an empty figure is generated."
)
fig, axes = plt.subplots()
fig.savefig(os.path.abspath('empty_dataset_diagnosis.png'),
bbox_inches='tight')
setattr(self, 'figure_dataset_diagnosis',
os.path.abspath('empty_dataset_diagnosis.png'))
return runtime
dict_keys = [
'temporal_std', 'VE_spatial', 'GS_corr', 'DVARS_corr', 'FD_corr',
'DR_BOLD', 'prior_modeling_maps'
]
voxelwise_list = []
for spatial_info in merged:
sub_list = [spatial_info[key] for key in dict_keys]
voxelwise_sub = np.array(sub_list[:5])
if len(sub_list[6]) > 0:
voxelwise_sub = np.concatenate(
(voxelwise_sub, np.array(sub_list[5]), np.array(
sub_list[6])),
axis=0)
else:
voxelwise_sub = np.concatenate(
(voxelwise_sub, np.array(sub_list[5])), axis=0)
voxelwise_list.append(voxelwise_sub)
num_DR_maps = len(sub_list[5])
num_prior_maps = len(sub_list[6])
voxelwise_array = np.array(voxelwise_list)
label_name = [
'temporal_std', 'VE_spatial', 'GS_corr', 'DVARS_corr', 'FD_corr'
]
label_name += [
'BOLD Dual Regression map %s' % (i) for i in range(num_DR_maps)
]
label_name += [
'BOLD Dual Convergence map %s' % (i) for i in range(num_prior_maps)
]
template_file = self.inputs.mask_file_dict['template_file']
mask_file = self.inputs.mask_file_dict['brain_mask']
from rabies.preprocess_pkg.preprocess_visual_QC import plot_3d, otsu_scaling
scaled = otsu_scaling(template_file)
ncols = 5
fig, axes = plt.subplots(nrows=voxelwise_array.shape[1],
ncols=ncols,
figsize=(12 * ncols,
2 * voxelwise_array.shape[1]))
for i, x_label in zip(range(voxelwise_array.shape[1]), label_name):
for j, y_label in zip(range(ncols), label_name[:ncols]):
ax = axes[i, j]
if i <= j:
ax.axis('off')
continue
X = voxelwise_array[:, i, :]
Y = voxelwise_array[:, j, :]
corr = elementwise_corrcoef(X, Y)
plot_3d([ax],
scaled,
fig,
vmin=0,
vmax=1,
cmap='gray',
alpha=1,
cbar=False,
num_slices=6,
planes=('coronal'))
analysis_functions.recover_3D(
mask_file, corr).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
plot_3d([ax],
sitk_img,
fig,
vmin=-0.7,
vmax=0.7,
cmap='cold_hot',
alpha=1,
cbar=True,
threshold=0.1,
num_slices=6,
planes=('coronal'))
ax.set_title('Cross-correlation for %s and %s' %
(x_label, y_label),
fontsize=15)
fig.savefig(os.path.abspath('dataset_diagnosis.png'),
bbox_inches='tight')
setattr(self, 'figure_dataset_diagnosis',
os.path.abspath('dataset_diagnosis.png'))
return runtime
def eval_relationships(maps, mask_file, percentile=0.01):
prior, average, network_var, corr_map_std, corr_map_CR_std, corr_map_VE, corr_map_tdof = maps
# 1) the correlation between the prior and the fitted average
#prior_cor = np.corrcoef(prior,average)[0,1]
# 2) the weighted average of the temporal s.d. correlations
#weights=prior/prior.sum()
#weighted_std_corr=(weights*corr_map_std)[np.isnan(corr_map_std)==0].sum()
#std_corr = np.corrcoef(prior[np.isnan(corr_map_std)==0], corr_map_std[np.isnan(corr_map_std)==0])[0,1]
# 3) the correlation between the prior and the fitted average
#VE_corr = np.corrcoef(prior[np.isnan(corr_map_VE)==0], corr_map_VE[np.isnan(corr_map_VE)==0])[0,1]
#weighted_VE_corr=(weights*corr_map_VE)[np.isnan(corr_map_VE)==0].sum()
tmppath = tempfile.mkdtemp()
map_masks = []
for map in maps:
if map is None:
map_masks.append(None)
tdof_avg_corr = None
continue
recover_3D(mask_file, map).to_filename(f'{tmppath}/temp_img.nii.gz')
img = sitk.ReadImage(f'{tmppath}/temp_img.nii.gz')
mask = percent_masking(img, percentile=percentile)
map_masks.append(mask)
# get also estimates of effect sizes for power estimation
# defined by average correlation within the temporal s.d. map for temporal s.d./CR R^2
if map is corr_map_std:
std_avg_corr = sitk.GetArrayFromImage(img)[map_masks[0]].mean()
if map is corr_map_CR_std:
CR_std_avg_corr = sitk.GetArrayFromImage(img)[map_masks[0]].mean()
if map is corr_map_VE:
VE_avg_corr = sitk.GetArrayFromImage(img)[map_masks[0]].mean()
if map is corr_map_tdof:
tdof_avg_corr = sitk.GetArrayFromImage(img)[map_masks[0]].mean()
prior_mask, average_mask, std_mask, corr_map_std_mask, corr_map_CR_std_mask, corr_map_VE_mask, tdof_mask = map_masks
if tdof_mask is None:
tdof_spec = None
else:
tdof_spec = dice_coefficient(prior_mask, tdof_mask)
return {
'Overlap: Prior - Dataset avg.':
dice_coefficient(prior_mask, average_mask),
'Overlap: Prior - Dataset MAD':
dice_coefficient(prior_mask, std_mask),
'Overlap: Prior - BOLD-Temporal s.d.':
dice_coefficient(prior_mask, corr_map_std_mask),
'Overlap: Prior - CR-Temporal s.d.':
dice_coefficient(prior_mask, corr_map_CR_std_mask),
'Overlap: Prior - tDOF':
tdof_spec,
'Avg.: BOLD-Temporal s.d.':
std_avg_corr,
'Avg.: CR-Temporal s.d.':
CR_std_avg_corr,
'Avg.: tDOF':
tdof_avg_corr,
}
def spatial_crosscorrelations(merged, scaled, mask_file, fig_path):
dict_keys = [
'temporal_std', 'predicted_std', 'VE_spatial', 'GS_corr', 'DVARS_corr',
'FD_corr', 'DR_BOLD', 'dual_ICA_maps'
]
voxelwise_list = []
for scan_data in merged:
sub_list = [scan_data[key] for key in dict_keys]
voxelwise_sub = np.array(sub_list[:6])
if len(sub_list[7]) > 0:
voxelwise_sub = np.concatenate(
(voxelwise_sub, np.array(sub_list[6]), np.array(sub_list[7])),
axis=0)
else:
voxelwise_sub = np.concatenate(
(voxelwise_sub, np.array(sub_list[6])), axis=0)
voxelwise_list.append(voxelwise_sub)
num_prior_maps = len(sub_list[6])
voxelwise_array = np.array(voxelwise_list)
label_name = [
'BOLD-Temporal s.d.', 'CR-Temporal s.d.', 'CR R^2', 'GS corr',
'DVARS corr', 'FD corr'
]
label_name += [
f'BOLD Dual Regression map {i}' for i in range(num_prior_maps)
]
label_name += [f'BOLD Dual ICA map {i}' for i in range(num_prior_maps)]
ncols = 6
fig, axes = plt.subplots(nrows=voxelwise_array.shape[1],
ncols=ncols,
figsize=(12 * ncols,
2 * voxelwise_array.shape[1]))
for i, x_label in zip(range(voxelwise_array.shape[1]), label_name):
for j, y_label in zip(range(ncols), label_name[:ncols]):
ax = axes[i, j]
if i <= j:
ax.axis('off')
continue
X = voxelwise_array[:, i, :]
Y = voxelwise_array[:, j, :]
corr = elementwise_spearman(X, Y)
plot_3d([ax],
scaled,
fig,
vmin=0,
vmax=1,
cmap='gray',
alpha=1,
cbar=False,
num_slices=6,
planes=('coronal'))
recover_3D(mask_file, corr).to_filename('temp_img.nii.gz')
sitk_img = sitk.ReadImage('temp_img.nii.gz')
cbar_list = plot_3d([ax],
sitk_img,
fig,
vmin=-1.0,
vmax=1.0,
cmap='cold_hot',
alpha=1,
cbar=True,
threshold=0.1,
num_slices=6,
planes=('coronal'))
ax.set_title(f'Cross-correlation for \n{x_label} and {y_label}',
fontsize=20,
color='white')
for cbar in cbar_list:
cbar.ax.get_yaxis().labelpad = 20
cbar.set_label("Spearman rho",
fontsize=12,
rotation=270,
color='white')
fig.savefig(fig_path, bbox_inches='tight')
def plot_relationships(mask_file, scaled, maps, percentile=0.01):
prior, average, network_var, corr_map_std, corr_map_CR_std, corr_map_VE, corr_map_tdof = maps
fig, axes = plt.subplots(nrows=6, ncols=1, figsize=(12, 2 * 6))
tmppath = tempfile.mkdtemp()
recover_3D(mask_file, prior).to_filename(f'{tmppath}/temp_img.nii.gz')
img = sitk.ReadImage(f'{tmppath}/temp_img.nii.gz')
ax = axes[0]
cbar_list = masked_plot(fig,
ax,
img,
scaled,
vmax=None,
percentile=percentile)
ax.set_title('Prior network', fontsize=25, color='white')
for cbar in cbar_list:
cbar.ax.get_yaxis().labelpad = 20
cbar.set_label("Prior measure",
fontsize=12,
rotation=270,
color='white')
tmppath = tempfile.mkdtemp()
recover_3D(mask_file, average).to_filename(f'{tmppath}/temp_img.nii.gz')
img = sitk.ReadImage(f'{tmppath}/temp_img.nii.gz')
ax = axes[1]
cbar_list = masked_plot(fig,
ax,
img,
scaled,
vmax=None,
percentile=percentile)
ax.set_title('Dataset average', fontsize=25, color='white')
for cbar in cbar_list:
cbar.ax.get_yaxis().labelpad = 20
cbar.set_label("Mean", fontsize=12, rotation=270, color='white')
tmppath = tempfile.mkdtemp()
recover_3D(mask_file,
network_var).to_filename(f'{tmppath}/temp_img.nii.gz')
img = sitk.ReadImage(f'{tmppath}/temp_img.nii.gz')
ax = axes[2]
cbar_list = masked_plot(fig,
ax,
img,
scaled,
vmax=None,
percentile=percentile)
ax.set_title('Dataset MAD', fontsize=25, color='white')
for cbar in cbar_list:
cbar.ax.get_yaxis().labelpad = 20
cbar.set_label("Median Absolute \nDeviation",
fontsize=12,
rotation=270,
color='white')
tmppath = tempfile.mkdtemp()
recover_3D(mask_file,
corr_map_std).to_filename(f'{tmppath}/temp_img.nii.gz')
img = sitk.ReadImage(f'{tmppath}/temp_img.nii.gz')
ax = axes[3]
cbar_list = masked_plot(fig,
ax,
img,
scaled,
vmax=1.0,
percentile=percentile)
ax.set_title('BOLD-Temporal s.d.', fontsize=25, color='white')
for cbar in cbar_list:
cbar.ax.get_yaxis().labelpad = 20
cbar.set_label("Spearman rho",
fontsize=12,
rotation=270,
color='white')
tmppath = tempfile.mkdtemp()
recover_3D(mask_file,
corr_map_CR_std).to_filename(f'{tmppath}/temp_img.nii.gz')
img = sitk.ReadImage(f'{tmppath}/temp_img.nii.gz')
ax = axes[4]
cbar_list = masked_plot(fig,
ax,
img,
scaled,
vmax=1.0,
percentile=percentile)
ax.set_title('CR-Temporal s.d.', fontsize=25, color='white')
for cbar in cbar_list:
cbar.ax.get_yaxis().labelpad = 20
cbar.set_label("Spearman rho",
fontsize=12,
rotation=270,
color='white')
ax = axes[5]
if corr_map_tdof is None:
ax.axis('off')
else:
tmppath = tempfile.mkdtemp()
recover_3D(mask_file,
corr_map_tdof).to_filename(f'{tmppath}/temp_img.nii.gz')
img = sitk.ReadImage(f'{tmppath}/temp_img.nii.gz')
cbar_list = masked_plot(fig,
ax,
img,
scaled,
vmax=1.0,
percentile=percentile)
ax.set_title('tDOF correlation', fontsize=25, color='white')
for cbar in cbar_list:
cbar.ax.get_yaxis().labelpad = 20
cbar.set_label("Spearman rho",
fontsize=12,
rotation=270,
color='white')
plt.tight_layout()
return fig
