def threshold_rmatrix(corr_matrix, option,
threshold, scans):
"""
Method to threshold the correaltion matrix based on
any of the two threshold options- sparsity, probability
or by simply providing correlation threshold. it is two
step process, first calculate the threshold and then apply
the threshold to the correlation matrix.
Parameters
----------
corr_matrix : string (numpy npy file)
patht o file containing correlation matrix
option : string (int)
list of threshold option: 0 for pvalue, 1 for sparsity,
any other for simply correlation threshold
threshold : string (float)
pvalue/sparsity_threshold/correaltion_threshold
scans : string (int)
Total number of scans in input data
Returns
-------
threshold_file : string (numpy npy file)
file containing threshold correlation matrix
Raises
------
Exception
"""
import numpy as np
import os
from CPAC.network_centrality import load_mat,\
convert_pvalue_to_r,\
convert_sparsity_to_r
try:
r_matrix = load_mat(corr_matrix)
print "threshold_option -->", option
try:
if option == 0:
r_value = convert_pvalue_to_r(scans, threshold)
if option == 1:
r_value = convert_sparsity_to_r(r_matrix, threshold)
else:
r_value = threshold
except:
print "Exception in calculating thresold value"
raise
print "correlation threshold value -> ", r_value
print "thresholding the correlation matrix...."
threshold_matrix = r_matrix > r_value
threshold_file = os.path.join(os.getcwd(), 'threshold_matrix.npy')
np.save(threshold_file, threshold_matrix.astype(np.float))
except Exception:
print "Exception while thresholding correlation matrix"
raise
return threshold_file
def calc_centrality(in_file, template, method_option, threshold_option,
threshold, allocated_memory):
'''
Function to calculate centrality and map them to a nifti file
Parameters
----------
in_file : string (nifti file)
path to subject data file
template : string (nifti file)
path to mask/parcellation unit
method_option : string
accepted values are 'degree centrality', 'eigenvector centrality', and
'lfcd'
threshold_option : string
accepted values are: 'significance', 'sparsity', and 'correlation'
threshold : float
pvalue/sparsity_threshold/threshold value
allocated_memory : string
amount of memory allocated to degree centrality
Returns
-------
out_list : list
list containing out mapped centrality images
'''
# Import packages
from CPAC.network_centrality import load,\
get_centrality_by_rvalue,\
get_centrality_by_sparsity,\
get_centrality_fast,\
map_centrality_matrix,\
calc_blocksize,\
convert_pvalue_to_r
from CPAC.network_centrality.utils import check_centrality_params
from CPAC.cwas.subdist import norm_cols
# First check input parameters and get proper formatted method/thr options
method_option, threshold_option = \
check_centrality_params(method_option, threshold_option, threshold)
# Init variables
out_list = []
ts, aff, mask, t_type, scans = load(in_file, template)
# If we're doing degree sparsity
if method_option == 'degree' and threshold_option == 'sparsity':
block_size = calc_blocksize(ts, memory_allocated=allocated_memory,
sparsity_thresh=threshold)
# Otherwise
elif method_option == 'eigenvector':
block_size = calc_blocksize(ts, memory_allocated=allocated_memory,
include_full_matrix=True)
# Otherwise, compute blocksize with regards to available memory
else:
block_size = calc_blocksize(ts, memory_allocated=allocated_memory,
include_full_matrix=False)
# Normalize the timeseries for easy dot-product correlation calc.
ts_normd = norm_cols(ts.T)
# P-value threshold centrality
if threshold_option == 'significance':
r_value = convert_pvalue_to_r(in_file, threshold, two_tailed=False)
centrality_matrix = get_centrality_by_rvalue(ts_normd,
mask,
method_option,
r_value,
block_size)
# Sparsity threshold
elif threshold_option == 'sparsity':
centrality_matrix = get_centrality_by_sparsity(ts_normd,
method_option,
threshold,
block_size)
# R-value threshold centrality
elif threshold_option == 'correlation':
centrality_matrix = get_centrality_by_rvalue(ts_normd,
mask,
method_option,
threshold,
block_size)
# For fast approach (no thresholding)
elif threshold_option == 3:
centrality_matrix = get_centrality_fast(ts, method_option)
# Otherwise, incorrect input for threshold_option
else:
err_msg = 'Threshold option: %s not supported for network centrality '\
'measure: %s; fix this in the pipeline config'\
% (str(threshold_option), str(method_option))
raise Exception(err_msg)
# Map the arrays back to images
for mat in centrality_matrix:
centrality_image = map_centrality_matrix(mat, aff, mask, t_type)
out_list.append(centrality_image)
# Finally return
return out_list
def calc_centrality(datafile,
template,
method_option,
threshold_option,
threshold,
weight_options,
allocated_memory):
'''
Method to calculate centrality and map them to a nifti file
Parameters
----------
datafile : string (nifti file)
path to subject data file
template : string (nifti file)
path to mask/parcellation unit
method_option : integer
0 - degree centrality calculation, 1 - eigenvector centrality calculation, 2 - lFCD calculation
threshold_option : an integer
0 for probability p_value, 1 for sparsity threshold,
2 for actual threshold value, and 3 for no threshold and fast approach
threshold : a float
pvalue/sparsity_threshold/threshold value
weight_options : list (boolean)
list of booleans, where, weight_options[0] corresponds to binary counting
and weight_options[1] corresponds to weighted counting (e.g. [True,False])
allocated_memory : string
amount of memory allocated to degree centrality
Returns
-------
out_list : list
list containing out mapped centrality images
'''
# Import packages
from CPAC.network_centrality import load,\
get_centrality_by_rvalue,\
get_centrality_by_sparsity,\
get_centrality_fast,\
map_centrality_matrix,\
calc_blocksize,\
convert_pvalue_to_r
from CPAC.cwas.subdist import norm_cols
# Check for input errors
if weight_options.count(True) == 0:
raise Exception("Invalid values in weight options" \
"At least one True value is required")
# If it's sparsity thresholding, check for (0,1]
if threshold_option == 1:
if threshold <= 0 or threshold > 1:
raise Exception('Threshold value must be a positive number'\
'greater than 0 and less than or equal to 1.'\
'\nCurrently it is set at %d' % threshold)
if method_option == 2 and threshold_option != 2:
raise Exception('lFCD must use correlation-type thresholding.'\
'Check the pipline configuration has this setting')
import time
start = time.clock()
# Init variables
out_list = []
ts, aff, mask, t_type, scans = load(datafile, template)
# If we're doing eigenvectory centrality, need entire correlation matrix
if method_option == 0 and threshold_option == 1:
block_size = calc_blocksize(ts, memory_allocated=allocated_memory,
sparsity_thresh=threshold)
elif method_option == 1:
block_size = calc_blocksize(ts, memory_allocated=allocated_memory,
include_full_matrix=True)
# Otherwise, compute blocksize with regards to available memory
else:
block_size = calc_blocksize(ts, memory_allocated=allocated_memory,
include_full_matrix=False)
# Normalize the timeseries for easy dot-product correlation calc.
ts_normd = norm_cols(ts.T)
# P-value threshold centrality
if threshold_option == 0:
r_value = convert_pvalue_to_r(scans, threshold)
centrality_matrix = get_centrality_by_rvalue(ts_normd,
mask,
method_option,
weight_options,
r_value,
block_size)
# Sparsity threshold
elif threshold_option == 1:
centrality_matrix = get_centrality_by_sparsity(ts_normd,
method_option,
weight_options,
threshold,
block_size)
# R-value threshold centrality
elif threshold_option == 2:
centrality_matrix = get_centrality_by_rvalue(ts_normd,
mask,
method_option,
weight_options,
threshold,
block_size)
# For fast approach (no thresholding)
elif threshold_option == 3:
centrality_matrix = get_centrality_fast(ts, method_option)
# Otherwise, incorrect input for threshold_option
else:
raise Exception('Option must be between 0-3 and not %s, check your '\
'pipeline config file' % str(threshold_option))
# Print timing info
print 'Timing:', time.clock() - start
# Map the arrays back to images
for mat in centrality_matrix:
centrality_image = map_centrality_matrix(mat, aff, mask, t_type)
out_list.append(centrality_image)
# Finally return
return out_list
def get_centrality_by_thresh(timeseries,
template,
method_option,
weight_options,
threshold,
r_value,
memory_allocated):
"""
Method to calculate degree and eigen vector centrality.
This method takes into consideration the amount of memory
allocated by the user to calculate degree centrality.
Parameters
----------
timeseries_data : numpy array
timeseries of the input subject
template : numpy array
Mask/ROI template for timeseries of subject
method_option : integer
0 - degree centrality calculation, 1 - eigenvector centrality calculation, 2 - lFCD calculation
weight_options : string (list of boolean)
list of two booleans for binarize and weighted options respectively
threshold : float
p-value threshold for the correlation values (ignored if the r_value option is specified)
r_value : float
threshold value in terms of the correlation (this will override the threshold option)
memory_allocated : a string
amount of memory allocated to degree centrality
Returns
-------
out_list : string (list of tuples)
list of tuple containing output name to be used to store nifti image
for centrality and centrality matrix
Raises
------
Exception
"""
import numpy as np
import os
from CPAC.network_centrality import calc_blocksize,\
cluster_data,\
convert_pvalue_to_r,\
degree_centrality,\
eigenvector_centrality
from CPAC.cwas.subdist import norm_cols
try:
# Init variables for use
out_list = []
nvoxs = timeseries.shape[0]
ntpts = timeseries.shape[1]
r_matrix = None # init correlation matrix
calc_degree = False # init degree measure flag to false
calc_eigen = False # init eigen measure flag to false
calc_lfcd= False # init lFCD measure flag to false
# Select which method we're going to perform
if method_option == 0:
calc_degree = True
elif method_option == 1:
calc_eigen = True
elif method_option == 2:
calc_lfcd = True
# Set weighting parameters
out_binarize = weight_options[0]
out_weighted = weight_options[1]
# Calculate the block size (i.e., number of voxels) to compute part of the
# connectivity matrix at once.
if calc_eigen:
# We still use a block size to calculate the whole correlation matrix
# because of issues in numpy that lead to extra memory usage when
# computing the dot product.
# See https://cmi.hackpad.com/Numpy-Memory-Issues-BlV9Pg5nRDM.
block_size = calc_blocksize(timeseries, memory_allocated, include_full_matrix=True)
else:
block_size = calc_blocksize(timeseries, memory_allocated)
if r_value == None:
print "Calculating threshold"
r_value = convert_pvalue_to_r(ntpts, threshold)
print "...%s -> %s" % (threshold, r_value)
print "Setup Intermediates/Outputs"
# Degree matrix init
if calc_degree:
print "...degree"
if out_binarize:
degree_binarize = np.zeros(nvoxs, dtype=timeseries.dtype)
out_list.append(('degree_centrality_binarize', degree_binarize))
if out_weighted:
degree_weighted = np.zeros(nvoxs, dtype=timeseries.dtype)
out_list.append(('degree_centrality_weighted', degree_weighted))
# Eigen matrix init
if calc_eigen:
print "...eigen"
r_matrix = np.zeros((nvoxs, nvoxs), dtype=timeseries.dtype)
if out_binarize:
eigen_binarize = np.zeros(nvoxs, dtype=timeseries.dtype)
out_list.append(('eigenvector_centrality_binarize', eigen_binarize))
if out_weighted:
eigen_weighted = np.zeros(nvoxs, dtype=timeseries.dtype)
out_list.append(('eigenvector_centrality_weighted', eigen_weighted))
# lFCD matrix init
if calc_lfcd:
print "...degree"
if out_binarize:
lfcd_binarize = np.zeros(nvoxs, dtype=timeseries.dtype)
out_list.append(('lFCD_binarize', lfcd_binarize))
if out_weighted:
lfcd_weighted = np.zeros(nvoxs, dtype=timeseries.dtype)
out_list.append(('lFCD_weighted', lfcd_weighted))
# Normalize the timeseries columns for simple correlation calc via dot product later..
print "Normalize TimeSeries"
timeseries = norm_cols(timeseries.T)
# Init blocking indices for correlation matrix calculation
print "Computing centrality across %i voxels" % nvoxs
i = block_size
j = 0
# Calculate correlation matrix in blocks while loop
while i <= nvoxs:
print "running block ->", i, j
try:
print "...correlating"
corr_matrix = np.dot(timeseries[:,j:i].T, timeseries)
except:
raise Exception("Error in calcuating block wise correlation for the block %i,%i"%(j,i))
if calc_eigen:
print "...storing correlation matrix"
r_matrix[j:i] = corr_matrix
if calc_degree:
if out_binarize:
print "...calculating binarize degree"
degree_centrality(corr_matrix, r_value, method="binarize", out=degree_binarize[j:i])
if out_weighted:
print "...calculating weighted degree"
degree_centrality(corr_matrix, r_value, method="weighted", out=degree_weighted[j:i])
if calc_lfcd:
xyz_a = np.argwhere(template)
krange = corr_matrix.shape[0]
print "...iterating through seeds in block - lfcd"
for k in range (0,krange):
corr_seed = corr_matrix[k,:]
labels = cluster_data(corr_seed,r_value,xyz_a)
seed_label = labels[j+k]
if out_binarize:
if seed_label > 0:
lfcd = np.sum(labels==seed_label)
else:
lfcd = 1
lfcd_binarize[j+k] = lfcd
if out_weighted:
if seed_label > 0:
lfcd = np.sum(corr_seed*(labels==seed_label))
else:
lfcd = 1
lfcd_weighted[j+k] = lfcd
print "...removing temporary correlation matrix"
del corr_matrix
j = i
if i == nvoxs:
break
elif (i+block_size) > nvoxs:
i = nvoxs
else:
i += block_size
# In case there are any zeros in lfcd matrix, set them to 1
if calc_lfcd:
if out_binarize:
lfcd_binarize[np.argwhere(lfcd_binarize == 0)] = 1
if out_weighted:
lfcd_weighted[np.argwhere(lfcd_weighted == 0)] = 1
# Perform eigenvector measures if necessary
try:
if calc_eigen:
if out_binarize:
print "...calculating binarize eigenvector"
eigen_binarize[:] = eigenvector_centrality(r_matrix, r_value, method="binarize").squeeze()
if out_weighted:
print "...calculating weighted eigenvector"
eigen_weighted[:] = eigenvector_centrality(r_matrix, r_value, method="weighted").squeeze()
except Exception:
print "Error in calcuating eigen vector centrality"
raise
if calc_degree:
print "...removing effect of auto-correlation on degree"
degree_binarize[degree_binarize!=0] = degree_binarize[degree_binarize!=0] - 1
degree_weighted[degree_weighted!=0] = degree_weighted[degree_weighted!=0] - 1
return out_list
except Exception:
print "Error in calcuating Centrality"
raise