def load_bids_events(layout, subject):
'''Create a design_matrix instance from BIDS event file'''
tr = layout.get_tr()
n_tr = nib.load(layout.get(subject=subject, scope='raw', suffix='bold')[0].path).shape[-1]
onsets = pd.read_csv(layout.get(subject=subject, suffix='events')[0].path, sep='\t')
onsets.columns = ['Onset', 'Duration', 'Stim']
return onsets_to_dm(onsets, sampling_freq=1/tr, run_length=n_tr)
def get_trialtype_pain_regressors(self,nifti_data,onset_file):
print("importing nifti")
#import the nifti
if (os.path.isfile(nifti_data + "nltoolstandard.nii.gz")):
msmrl1 = Brain_Data(
nifti_data + "nltoolstandard.nii.gz")
else:
msmrl1 = Brain_Data(
nifti_data + ".nii.gz")
msmrl1.write(nifti_data + "nltoolstandard.nii.gz")
#preprocess the nifti?
print("importing onsets")
#import the onset
onsets = onsets_to_dm(
onset_file,
TR=2,
runLength=msmrl1.shape()[0]
)
#process the onset files
#
onsets.sampling_rate=2
onsets_convolved=onsets.convolve()
for c in onsets_convolved.columns:
if sum(onsets_convolved.ix[:, c]) <= 0:
print('deleting '+ str(c))
del onsets_convolved[c]
onsets_convolved['linearterm']=range(1,361)
onsets_convolved['quadraticterm']=[pow(x,2) for x in onsets_convolved['linearterm']]
onsets_convolved['cubicterm']=[pow(x,3) for x in onsets_convolved['linearterm']]
onsets_convolved['ones']=[1]*360
msmrl1.X=onsets_convolved
print("convolved onsets; regressing...")
#regress
regression=msmrl1.regress()
print("Regressing; calculating similarity...")
msm_predicted_pain = regression['beta'].similarity(self.stats['weight_map'], 'dot_product')
onset_colnames = onsets_convolved.columns.tolist()
msm_predicted_pain_dict={}
for i, b in enumerate(msm_predicted_pain):
msm_predicted_pain_dict[onset_colnames[i]] = b
return msm_predicted_pain_dict
def get_trialtype_pain_regressors(self, nifti_data, onset_file):
print("importing nifti")
#import the nifti
#load the nltools prepped file if it's available.
if (os.path.isfile(nifti_data + "nltoolstandard.nii.gz")):
msmrl1 = Brain_Data(nifti_data + "nltoolstandard.nii.gz")
else: #but if it's not; no worries; just load the original one.
msmrl1 = Brain_Data(nifti_data + ".nii.gz")
msmrl1.write(nifti_data + "nltoolstandard.nii.gz")
#I want to standardize globally; this will preserve the relative strengths of each time point
#and preserve the relative activity at each voxel.
#and let's use the mean standard deviation across all the images.
#msmrl1.data = msmrl1.data - np.tile(msmrl1.mean().mean(),msmrl1.data.shape)
#msmrl1.data = msmrl1.data / np.tile(np.std(msmrl1.data,axis=1).mean(),msmrl1.data.shape)
# OR we could apply the standardization to the OUTPUT.
#grand_mean=msmrl1.mean().mean()
#grand_sd=np.std(msmrl1.data,axis=1).mean()
#preprocess the nifti?
print("importing onsets")
#import the onset
onsets = onsets_to_dm(onset_file, TR=2, runLength=msmrl1.shape()[0])
#process the onset files
#
onsets.sampling_rate = 2
onsets_convolved = onsets.convolve()
#delete columns with no information in them.
for c in onsets_convolved.columns:
if sum(onsets_convolved.ix[:, c]) <= 0:
print('deleting ' + str(c))
del onsets_convolved[c]
rowcount = onsets_convolved.__len__()
if rowcount != 360:
warnings.warn(
"Just a friendly FYI: expected number of rows is 360 but this subject had "
+ str(rowcount) +
". Probably this subject got cut off the task half-way through."
)
onsets_convolved['linearterm'] = range(1, rowcount + 1)
onsets_convolved['quadraticterm'] = [
pow(x, 2) for x in onsets_convolved['linearterm']
]
onsets_convolved['cubicterm'] = [
pow(x, 3) for x in onsets_convolved['linearterm']
]
onsets_convolved['ones'] = [1] * rowcount
msmrl1.X = onsets_convolved
print("convolved onsets; regressing...")
#regress the file on each of the onsets. So then, when we compare similarity to the regression, we'll be getting the
#regression to the each event, not to each TR.
regression = msmrl1.regress()
print("Regressing; calculating similarity to the pain map from " +
self.decoder_origin + "...")
msm_predicted_pain = regression['beta'].similarity(
self.decoder, 'dot_product')
msm_predicted_pain_scaled = msm_predicted_pain - msmrl1.data.std()
onset_colnames = onsets_convolved.columns.tolist()
msm_predicted_pain_dict = {}
for i, b in enumerate(msm_predicted_pain_scaled):
msm_predicted_pain_dict[onset_colnames[i]] = b
return msm_predicted_pain_dict
######################################################################### # Load and Manipulate an Onsets File # ------------------------------------ # # Nltools provides basic file-reading support for 2 or 3 column formatted onset files. # Users can look at the onsets_to_dm function() as a template to build more complex file readers if desired or to see additional features. # Here we simply point to an onsetfile where each event lasted exactly 1 TR, provide some basic experiment metadata, add an intercept, and get back a basic design matrix. from nltools.utils import get_resource_path from nltools.file_reader import onsets_to_dm from nltools.data import Design_Matrix import os onsetsFile = os.path.join(get_resource_path(),'onsets_example.txt') dm = onsets_to_dm(onsetsFile, TR=2.0, runLength=160, sort=True, addIntercept=True) ######################################################################### # The class stores basic meta data including convolution functions (default is glover HRF) and whether convolution has been performed, or whether the model contains a constant term. print(dm.info()) ######################################################################### # We can easily visualize the design matrix too dm.heatmap() ######################################################################### # We can also add nth order polynomial terms. In this case we'll add a linear term to capture linear trends. # By default the class will add all lower-order polynomials, but is smart enough to realize we already have a constant so it won't be duplicated.
# Creating a Design Matrix from an onsets file
# ********************************************
#
# Nltools provides basic file-reading support for 2 or 3 column formatted onset files. Users can look at the onsets_to_dm function as a template to build more complex file readers if desired or to see additional features. Nltools includes an example onsets file where each event lasted exactly 1 TR and TR = 2s. Lets use that to create a design matrix with an intercept and linear trend
from nltools.utils import get_resource_path
from nltools.file_reader import onsets_to_dm
from nltools.data import Design_Matrix
import os
TR = 2.0
sampling_freq = 1. / TR
onsetsFile = os.path.join(get_resource_path(), 'onsets_example.txt')
dm = onsets_to_dm(onsetsFile,
sampling_freq=sampling_freq,
run_length=160,
sort=True,
add_poly=1)
dm.heatmap()
#########################################################################
# Creating a Design Matrix from a generic csv file
# ************************************************
#
# Alternatively you can read a generic csv file and transform it into a Design Matrix using pandas file reading capability. Here we'll read in an example covariates file that contains the output of motion realignment estimated during a fMRI preprocessing pipeline.
import pandas as pd
covariatesFile = os.path.join(get_resource_path(), 'covariates_example.csv')
cov = pd.read_csv(covariatesFile)
cov = Design_Matrix(cov, sampling_freq=sampling_freq)
def import_and_convolve(self,
nifti_data,
onset_file,
data_mask=None,
motion_regressors=None):
print("importing nifti")
# import the nifti
# load the nltools prepped file if it's available.
if (os.path.isfile(nifti_data + self.data_fmri_space + ".nii.gz")):
msmrl1 = Brain_Data(nifti_data + self.data_fmri_space + ".nii.gz",
mask=data_mask)
else: # but if it's not; no worries; just load the original one.
msmrl1 = Brain_Data(nifti_data + ".nii.gz", mask=data_mask)
msmrl1.write(nifti_data + self.data_fmri_space + ".nii.gz")
# I want to standardize globally; this will preserve the relative strengths of each time point
# and preserve the relative activity at each voxel.
# and let's use the mean standard deviation across all the images.
# msmrl1.data = msmrl1.data - np.tile(msmrl1.mean().mean(),msmrl1.data.shape)
# msmrl1.data = msmrl1.data / np.tile(np.std(msmrl1.data,axis=1).mean(),msmrl1.data.shape)
# OR we could apply the standardization to the OUTPUT.
# grand_mean=msmrl1.mean().mean()
# grand_sd=np.std(msmrl1.data,axis=1).mean()
# preprocess the nifti?
print("importing onsets")
# import the onset
onsets = onsets_to_dm(onset_file, TR=2, runLength=msmrl1.shape()[0])
# process the onset files
#
onsets.sampling_rate = 2
onsets_convolved = onsets.convolve()
# delete columns with no information in them.
for c in onsets_convolved.columns:
if sum(onsets_convolved.ix[:, c]) <= 0:
print('deleting ' + str(c))
del onsets_convolved[c]
rowcount = onsets_convolved.__len__()
if rowcount != 360:
warnings.warn(
"Just a friendly FYI: expected number of rows is 360 but this subject had "
+ str(rowcount) +
". Probably this subject got cut off the task half-way through."
)
onsets_convolved['linearterm'] = range(1, rowcount + 1)
onsets_convolved['quadraticterm'] = [
pow(x, 2) for x in onsets_convolved['linearterm']
]
onsets_convolved['cubicterm'] = [
pow(x, 3) for x in onsets_convolved['linearterm']
]
onsets_convolved['ones'] = [1] * rowcount
if (motion_regressors is not None):
onsets_convolved = pandas.concat(
[onsets_convolved, motion_regressors], axis=1)
msmrl1.X = onsets_convolved
return msmrl1
# delim_whitespace=True, header=None, comment='#',
# names=["Index","SegId","NVoxels","Volume_mm3","StructName","Mean","StdDev","Min","Max","Range"])
# great and now add the labels to the summary file.
# sub_r_m_roi_sum_dt=sub_r_m_roi_sum_dt.merge(freesurfer_LUT_df[["RegionId","Label"]],how='left',left_on='SegId',right_on='RegionId')
colnames = sub_r_m_roi_dt.columns
#onset_file='/Users/benjaminsmith/Dropbox/joint-modeling/data/runfiledetail20171020T012224_s214_reward_r1.txt'
if (os.path.isfile(onset_file)):
print('we have a match; ' + onset_file)
print("importing onsets")
# import the onset
onsets = onsets_to_dm(
onset_file,
TR=2,
runLength=sub_r_m_roi_dt.shape[
0] #we don't know this for sure. But in this instance it doesn't matter.
# import the run_length data from the data.
)
onsets.sampling_rate = 2
onsets_convolved = onsets.convolve()
#removing columns with nothing in them.
for c in onsets_convolved.columns:
if sum(onsets_convolved.loc[:, c]) <= 0:
print('deleting ' + str(c))
del onsets_convolved[c]
rowcount = onsets_convolved.__len__()
def test_onsets_to_dm():
fpath = os.path.join(get_resource_path(), "onsets_example.txt")
data = pd.read_csv(os.path.join(get_resource_path(), "onsets_example.txt"))
sampling_freq = 0.5
run_length = 1364
Duration = 10
TR = 1 / sampling_freq
# Two-column
# Test loading from a file
dm = onsets_to_dm(fpath, sampling_freq, run_length)
assert isinstance(dm, Design_Matrix)
# Check it has run_length rows and nStim columns
assert dm.shape == (run_length, data.Stim.nunique())
# Get the unique number of presentations of each Stim from the original file
stim_counts = data.Stim.value_counts(sort=False)[dm.columns]
# Check there are only as many onsets as occurences of each Stim
np.allclose(stim_counts.values, dm.sum().values)
# Three-column with loading from dataframe
data["Duration"] = Duration
dm = onsets_to_dm(data, sampling_freq, run_length)
# Check it has run_length rows and nStim columns
assert dm.shape == (run_length, data.Stim.nunique())
# Because timing varies in seconds and isn't TR-locked each stimulus should last at Duration/TR number of TRs and at most Duration/TR + 1 TRs
# Check that the total number of TRs for each stimulus >= 1 + (Duration/TR) and <= 1 + (Duration/TR + 1)
onsets = dm.sum().values
durations = data.groupby("Stim").Duration.mean().values
for o, c, d in zip(onsets, stim_counts, durations):
assert c * (d / TR) <= o <= c * ((d / TR) + 1)
# Multiple onsets
dm = onsets_to_dm([data, data], sampling_freq, run_length)
# Check it has run_length rows and nStim columns
assert dm.shape == (run_length * 2, data.Stim.nunique())
# Multiple onsets with polynomials auto-added
dm = onsets_to_dm([data, data], sampling_freq, run_length, add_poly=2)
assert dm.shape == (run_length * 2, data.Stim.nunique() + (3 * 2))
dm = onsets_to_dm([data, data],
sampling_freq,
run_length,
add_poly=2,
keep_separate=False)
assert dm.shape == (run_length * 2, data.Stim.nunique() + 3)
# Three-column from file with variable durations
data = pd.read_csv(
os.path.join(get_resource_path(), "onsets_example_with_dur.txt"))
run_length = 472
dm = onsets_to_dm(data, sampling_freq, run_length)
assert dm.shape == (run_length, data.Stim.nunique())
onsets = dm.sum().values
stim_counts = data.Stim.value_counts().values
durations = data.groupby("Stim").Duration.mean().values
for o, c, d in zip(onsets, stim_counts, durations):
assert c * (d / TR) <= o <= c * ((d / TR) + 1)
msm_predicted_pain=msmrl1_detrended.similarity(stats['weight_map'],'correlation')# + stats['intercept']
msm_dat = pd.DataFrame(data=
{'PredictedPain':msm_predicted_pain,
'Timepoint':msmrl1_detrended.X['TimePoint']})
with sns.plotting_context(context='paper',font_scale=2):
sns.factorplot(data=msm_dat
,x='Timepoint',y='PredictedPain')
#file should have headers: Stim,Onset,Duration
#Onset and Duration should both be set in seconds.
onsets=onsets_to_dm(
sub113_onset_file,
TR=2,
runLength=msmrl1.shape()[0]
)
onsets.sampling_rate=2
#I believe there is a bug in the code here which means we have to set Sampling Rate twice.
onsets_convolved=onsets.convolve()
list(onsets_convolved)
#graph the pain activity against the pain regressor
msm_dat = pd.DataFrame(data=
{'PredictedPain':msm_predicted_pain,
'Timepoint':msmrl1_detrended.X['TimePoint']})
with sns.plotting_context(context='paper',font_scale=2):
sns.factorplot(data=msm_dat
,x='Timepoint',y='PredictedPain')
dm_with_cosine = dm.add_dct_basis(duration=5) print(dm_with_cosine.details()) ######################################################################### # Load and Manipulate an Onsets File # ----------------------------------- # # Nltools provides basic file-reading support for 2 or 3 column formatted onset files. Users can look at the onsets_to_dm function as a template to build more complex file readers if desired or to see additional features. Nltools includes an example onsets file where each event lasted exactly 1 TR. Lets use that to create a design matrix with an intercept and linear trend from nltools.utils import get_resource_path from nltools.file_reader import onsets_to_dm from nltools.data import Design_Matrix import os onsetsFile = os.path.join(get_resource_path(),'onsets_example.txt') dm = onsets_to_dm(onsetsFile, TR=2.0, runLength=160, sort=True,add_poly=1) dm.heatmap() ######################################################################### # Design Matrix makes it easy to perform convolution and will auto-ignore all columns that are consider polynomials. By default it will use the one-parameter glover_hrf kernel (see nipy for details). However, any kernel can be passed as an argument, including a list of different kernels for highly flexible convolution across many types of data (e.g. SCR). dm = dm.convolve() print(dm.details()) dm.heatmap() ######################################################################### # Load and Z-score a Covariates File # ---------------------------------- # # Now we're going to handle a covariates file that's been generated by a preprocessing routine. First we'll read in the text file using pandas and convert it to a design matrix.
