def run(opts):
if opts.voxel:
img, img_data = loadnifti(opts.input[0])
if opts.vertex:
img, img_data = loadmgh(opts.input[0])
numMerge = len(opts.input)
outname = opts.output[0]
if opts.mask:
if opts.voxel:
mask, data_mask = loadnifti(opts.mask[0])
if opts.vertex:
mask, data_mask = loadmgh(opts.mask[0])
mask_index = data_mask > 0.99
else:
mask_index = np.zeros(
(img_data.shape[0], img_data.shape[1], img_data.shape[2]))
mask_index = (mask_index == 0)
img_data_trunc = img_data[mask_index].astype(np.float32)
if opts.scale:
img_data_trunc = zscaler(img_data_trunc.T).T
for i in xrange(numMerge):
print "merging image %s" % opts.input[i]
if i > 0:
if opts.voxel:
_, tempimgdata = loadnifti(opts.input[i])
if opts.vertex:
_, tempimgdata = loadmgh(opts.input[i])
tempimgdata = tempimgdata[mask_index].astype(np.float32)
if opts.scale:
tempimgdata = zscaler(tempimgdata.T).T
img_data_trunc = np.column_stack((img_data_trunc, tempimgdata))
# remove nan if any
img_data_trunc[np.isnan(img_data_trunc)] = 0
if opts.fastica:
ica = FastICA(n_components=int(opts.fastica[0]),
max_iter=5000,
tol=0.0001)
S_ = ica.fit_transform(img_data_trunc).T
components = ica.components_.T
#scaling
fitcomps = np.copy(S_)
fitcomps = zscaler(fitcomps)
img_data_trunc = np.copy(
fitcomps.T) # ram shouldn't be an issue here...
np.savetxt("%s.ICA_fit.csv" % sys.argv[4],
zscaler(components),
fmt='%10.8f',
delimiter=',')
#save outputs and ica functions for potential ica removal
if os.path.exists('ICA_temp'):
print 'ICA_temp directory exists'
exit()
else:
os.makedirs('ICA_temp')
np.save('ICA_temp/signals.npy', S_)
pickle.dump(ica, open("ICA_temp/icasave.p", "wb"))
if opts.bothhemi:
lh_tempmask = lh_mask_index * 1
rh_tempmask = rh_mask_index * 1
savemgh(lh_tempmask[lh_tempmask == 1], lh_img, lh_mask_index,
'ICA_temp/lh_mask.mgh')
savemgh(rh_tempmask[rh_tempmask == 1], rh_img, rh_mask_index,
'ICA_temp/rh_mask.mgh')
else:
tempmask = mask_index * 1
if opts.voxel:
savenifti(tempmask[tempmask == 1], img, mask_index,
'ICA_temp/mask.nii.gz')
else:
savemgh(tempmask[tempmask == 1], img, mask_index,
'ICA_temp/mask.mgh')
if opts.voxel:
savenifti(img_data_trunc.astype(np.float32), img, mask_index, outname)
if opts.vertex:
savemgh(img_data_trunc.astype(np.float32), img, mask_index, outname)
def run(opts):
element, image_array, masking_array, maskname, affine_array, _, _, surfname, _, _, _ = read_tm_filetype(
opts.tmifile[0])
img_data_trunc = image_array[0]
del image_array # reduce ram usage
img_data_trunc = img_data_trunc.astype(np.float32)
img_data_trunc[np.isnan(img_data_trunc)] = 0
if opts.covar:
covar = np.genfromtxt('dmy_Blocks.csv', delimiter=',', dtype=np.float)
img_data_trunc = lm_residuals(img_data_trunc.T, covar).T
if opts.scale:
img_data_trunc = zscaler(img_data_trunc)
if opts.maxnpcacomponents:
numpcacomps = opts.maxnpcacomponents = [0]
else:
numpcacomps = img_data_trunc.shape[1]
if opts.pca:
pca = PCA(n_components=numpcacomps)
S_ = pca.fit_transform(img_data_trunc).T
for i in range(len(pca.explained_variance_ratio_)):
if (pca.explained_variance_ratio_[i] < 0.01):
start_comp_number = i
print("Component %d explains %1.4f of the variance." %
(i, pca.explained_variance_ratio_[0:i].sum()))
if pca.explained_variance_ratio_[0:i].sum() < 0.80:
pass
else:
break
print(
"%d number of components, explaining %1.2f of the variance." %
(start_comp_number,
(pca.explained_variance_ratio_[0:start_comp_number].sum() * 100)))
rsquare_scores = []
std_err = []
w_rsquare_scores = []
range_comp = np.arange(0, numpcacomps - 2, 1)
for comp_number in range_comp:
x = np.array(
list(range(len(pca.explained_variance_ratio_)))[comp_number:])
y = pca.explained_variance_ratio_[comp_number:]
slope, intercept, r_value, p_value, se = stats.linregress(x, y)
rsquare_scores.append((r_value**2))
std_err.append((se))
w_rsquare_scores.append(
(r_value**2 *
((range_comp[-1] - comp_number + 1) / range_comp[-1])))
best_comp = np.argmax(rsquare_scores)
best_comp2 = np.argmin(std_err)
best_comp3 = np.argmax(w_rsquare_scores)
print(
"Best Component %d; R-square residual score %1.4f; variance explained %1.4f"
% (best_comp, rsquare_scores[best_comp],
pca.explained_variance_ratio_[:best_comp].sum()))
x = np.array(
list(range(len(pca.explained_variance_ratio_)))[best_comp:])
y = pca.explained_variance_ratio_[best_comp:]
m, b = np.polyfit(x, y, 1)
# %matplotlib
xaxis = np.arange(pca.explained_variance_ratio_.shape[0]) + 1
plt.plot(xaxis, pca.explained_variance_ratio_, 'ro-', linewidth=2)
plt.axvline(best_comp, color='r', linestyle='dashed', linewidth=2)
plt.text(int(best_comp - 10),
pca.explained_variance_ratio_.max() * .99,
'Noise; Comp=%d, Sum V(e)=%1.2f' %
(best_comp, pca.explained_variance_ratio_[:best_comp].sum()),
rotation=90)
plt.axvline(start_comp_number,
color='g',
linestyle='dashed',
linewidth=2)
plt.text(int(start_comp_number - 10),
pca.explained_variance_ratio_.max() * .99,
'Threshold; Comp=%d, Sum V(e)=%1.2f' %
(start_comp_number,
pca.explained_variance_ratio_[:start_comp_number].sum()),
rotation=90)
plt.axvline(best_comp3, color='b', linestyle='dashed', linewidth=2)
plt.text(
int(best_comp3 - 10),
pca.explained_variance_ratio_.max() * .99,
'Weight Noise; Comp=%d, Sum V(e)=%1.2f' %
(best_comp3, pca.explained_variance_ratio_[:best_comp3].sum()),
rotation=90)
plt.plot(xaxis, m * xaxis + b, '--')
plt.title('Scree Plot')
plt.xlabel('Principal Component')
plt.ylabel('Explained Variance Ratio')
plt.show()
###### TEST ##########
# xaxis = np.arange(pca.explained_variance_ratio_.shape[0]) + 1
# plt.plot(xaxis, pca.explained_variance_ratio_, 'ro-', linewidth=2)
# plt.axvline(start_comp_number, color='b', linestyle='dashed', linewidth=2)
# plt.title('Scree Plot')
# plt.xlabel('Principal Component')
# plt.ylabel('Explained Variance Ratio')
# plt.show()
if opts.fastica:
if opts.pca:
num_comp = start_comp_number
elif opts.numicacomponents:
num_comp = int(opts.numicacomponents[0])
else:
print("unknown number of compenents")
exit()
print(num_comp)
img_data_trunc[np.isnan(img_data_trunc)] = 0
ica, sort_mask, _ = tmi_run_ica(img_data_trunc,
num_comp,
variance_threshold=.5,
masking_array=masking_array,
affine_array=affine_array,
filetype='mgh',
outname='ica.mgh')
components = ica.components_.T
if opts.timeplot:
# generate graphs
analysis_name = opts.timeplot[0]
# components = np.copy(fitcomps)
components = zscaler(components[:, sort_mask].T).T
subs = np.array(list(range(components.shape[0]))) + 1
time_step = 1 / 100
if os.path.exists(analysis_name):
print('%s directory exists' % analysis_name)
exit()
else:
os.makedirs(analysis_name)
plt.figure(figsize=(10, 5))
for i in range(components.shape[1]):
plt.plot(subs, components[:, i], 'ro-', linewidth=2)
plt.title('Component %d Plot' % (i + 1))
plt.xlabel('Time or Subject (units)')
plt.savefig('%s/%s_timeplot_comp%d.jpg' %
(analysis_name, analysis_name, (i + 1)))
plt.clf()
ps = np.abs(np.fft.fft(components[:, i]))**2
freqs = np.fft.fftfreq(components[:, i].size, time_step)
idx = np.argsort(freqs)
plt.plot(np.abs(freqs[idx]), ps[idx])
plt.title('Component %d Powerspectrum' % (i + 1))
plt.xlabel('Unit Frequency (Hz / 100)')
plt.savefig('%s/%s_power_comp%d.jpg' %
(analysis_name, analysis_name, (i + 1)))
plt.clf()
def tmi_run_ica(img_data_trunc,
num_comp,
masking_array,
affine_array,
variance_threshold=0.9,
timeplot=False,
timeplot_name=None,
filetype='nii.gz',
outname='ica.nii.gz'):
ica = FastICA(n_components=int(num_comp), max_iter=1000, tol=0.00001)
S_ = ica.fit_transform(img_data_trunc).T
components = ica.components_.T
#scaling
fitcomps = np.zeros_like(S_)
fitcomps[:] = np.copy(S_)
fitcomps = zscaler(fitcomps)
img_data_trunc = np.copy(fitcomps.T) # ram shouldn't be an issue here...
np.savetxt("ICA_fit.csv", zscaler(components), fmt='%10.8f', delimiter=',')
# variance explained.
explained_total_var = np.zeros((int(num_comp)))
explained_var_ratio = np.zeros((int(num_comp)))
# total variance
back_projection = ica.inverse_transform(S_.T)
total_var = back_projection.var()
for i in range(int(num_comp)):
tempcomps = np.copy(S_)
tempcomps[i, :] = 0
temp_back_proj = ica.inverse_transform(tempcomps.T)
temp_var = temp_back_proj.var()
explained_var_ratio[i] = total_var - temp_var
explained_total_var[i] = (total_var - temp_var) / total_var
print("ICA # %d; Percent of Total Variance %1.3f" %
((i + 1), explained_total_var[i] * 100))
explained_var_ratio = explained_var_ratio / explained_var_ratio.sum()
sum_total_variance_explained = explained_total_var.sum()
print("Total variance explained by all components = %1.3f" %
sum_total_variance_explained)
print("Re-ordering components")
sort_mask = (-1 * explained_total_var).argsort()
if sum_total_variance_explained > variance_threshold:
#sort data
sort_mask = (-1 * explained_total_var).argsort()
np.savetxt("ICA_total_var.csv",
explained_total_var[sort_mask],
fmt='%1.5f',
delimiter=',')
np.savetxt("ICA_explained_var_ratio.csv",
explained_var_ratio[sort_mask],
fmt='%1.5f',
delimiter=',')
img_data_trunc = img_data_trunc[:, sort_mask]
if filetype == 'nii.gz':
savenifti_v2(img_data_trunc, masking_array[0], outname,
affine_array[0])
else:
pointer = 0
position_array = [0]
for i in range(len(masking_array)):
pointer += len(masking_array[i][masking_array[i] == True])
position_array.append(pointer)
del pointer
for i in range(len(masking_array)):
start = position_array[i]
end = position_array[i + 1]
savemgh_v2(img_data_trunc[start:end], masking_array[i],
"%d_%s" % (i, outname), affine_array[i])
# save outputs and ica functions for potential ica removal
if os.path.exists('ICA_temp'):
print('ICA_temp directory exists')
exit()
else:
os.makedirs('ICA_temp')
np.save('ICA_temp/signals.npy', S_)
pickle.dump(ica, open("ICA_temp/icasave.p", "wb"))
return ica, sort_mask, sum_total_variance_explained
def run(opts):
image_array = []
affine_array = []
masking_array = []
maskname = []
vertex_array = []
face_array = []
surfname = []
adjacency_array = []
tmi_history = []
if opts.outputname:
outname = opts.outputname[0]
if not outname.endswith('tmi'):
if opts.outputtype == 'binary':
if not outname.endswith('tmi'):
outname += '.tmi'
else:
outname += '.ascii.tmi'
if opts.append:
outname = opts.append[0]
_, image_array, masking_array, maskname, affine_array, vertex_array, face_array, surfname, adjacency_array, tmi_history, subjectids = read_tm_filetype(
outname)
if opts.inputimages:
for i in range(len(opts.inputimages)):
basename, file_ext = os.path.splitext(opts.inputimages[i])
if file_ext == '.gz':
os.system("zcat %s > %s" % (opts.inputimages[i], basename))
img = nib.load('%s' % basename)
else:
img = nib.load(opts.inputimages[i])
img_data = img.get_data()
if opts.inputmasks:
mask = nib.load(opts.inputmasks[i])
mask_data = mask.get_data()
if not np.array_equal(img_data.shape[:3], mask_data.shape[:3]):
print "Error mask data dimension do not fit image dimension"
exit()
mask_data = mask_data == 1
img_data = img_data[mask_data]
else:
img_data, mask_data = maskdata(img_data)
masking_array.append(np.array(mask_data))
image_array.append(np.array(img_data))
affine_array.append(img.affine)
maskname.append(np.array(os.path.basename(opts.inputimages[i])))
if file_ext == '.gz':
os.system("rm %s" % basename)
if opts.concatenateimages:
img = nib.load(opts.concatenateimages[0])
img_data = img.get_data()
numMerge = len(opts.concatenateimages)
if opts.inputmasks:
if not len(opts.inputmasks) == 1:
print "Only one mask can be added using concatenate. See help (hint: rerun using append option for multiple modalities/surfaces)"
exit()
mask = nib.load(opts.inputmasks[0])
mask_data = mask.get_data()
if not np.array_equal(img_data.shape[:3], mask_data.shape[:3]):
print "Error mask data dimension do not fit image dimension"
exit()
mask_data = mask_data == 1
img_data = img_data[mask_data].astype(np.float32)
else:
print "Creating mask from first image."
img_data, mask_data = maskdata(img_data)
for i in xrange(numMerge):
print "merging image %s" % opts.concatenateimages[i]
if i > 0:
tempdata = nib.load(opts.concatenateimages[i]).get_data()
tempdata = tempdata[mask_data].astype(np.float32)
if opts.scale:
tempdata = zscaler(tempdata.T).T
img_data = np.column_stack((img_data, tempdata))
else:
if opts.scale:
img_data = zscaler(img_data.T).T
masking_array.append(np.array(mask_data))
image_array.append(np.array(img_data))
affine_array.append(img.affine)
if opts.concatenatename:
maskname.append(np.array(os.path.basename(
opts.concatenatename[0])))
if opts.inputtext:
for i in range(len(opts.inputtext)):
#img_data = np.genfromtxt(opts.inputtext[i], delimiter=',') # slower, more ram usage
img_data = []
with open(opts.inputtext[i]) as data_file:
for line in data_file:
img_data.append(line.strip().split(','))
img_data = np.array(img_data).astype(np.float32)
img_data, mask_data = maskdata(img_data)
masking_array.append(np.array(mask_data))
image_array.append(np.array(img_data))
if opts.concatenatetext:
firstimg_data = np.genfromtxt(opts.concatenatetext[0], delimiter=',')
numMerge = len(opts.concatenatetext)
for i in xrange(numMerge):
print "merging text file %s" % opts.concatenatetext[i]
if i > 0:
tempdata = np.genfromtxt(opts.concatenatetext[i],
delimiter=',')
img_data = np.column_stack((img_data, tempdata))
img_data, mask_data = maskdata(img_data)
masking_array.append(np.array(mask_data))
image_array.append(np.array(img_data))
if opts.concatenatebinary:
firstimg_data = np.fromfile(opts.concatenatebinary[0], dtype='f')
numMerge = len(opts.concatenatebinary)
for i in xrange(numMerge):
print "merging simple float binary file %s" % opts.concatenatebinary[
i]
if i > 0:
tempdata = np.fromfile(opts.concatenatebinary[i], dtype='f')
img_data = np.column_stack((img_data, tempdata))
img_data, mask_data = maskdata(img_data)
masking_array.append(np.array(mask_data))
image_array.append(np.array(img_data))
if opts.inputfreesurfer:
for i in range(len(opts.inputfreesurfer)):
v, f = convert_fs(str(opts.inputfreesurfer[i]))
vertex_array.append(v)
face_array.append(f)
surfname.append(np.array(os.path.basename(
opts.inputfreesurfer[i])))
if opts.inputgifti:
for i in range(len(opts.inputgifti)):
v, f = convert_gifti(str(opts.inputgifti[i]))
vertex_array.append(v)
face_array.append(f)
surfname.append(np.array(os.path.basename(opts.inputgifti[i])))
if opts.inputmniobj:
for i in range(len(opts.inputmniobj)):
v, f = convert_mni_object(str(opts.inputmniobj[i]))
vertex_array.append(v)
face_array.append(f)
surfname.append(np.array(os.path.basename(opts.inputmniobj[i])))
if opts.inputply:
for i in range(len(opts.inputply)):
v, f = convert_ply(str(opts.inputply[i]))
vertex_array.append(v)
face_array.append(f)
surfname.append(np.array(os.path.basename(opts.inputply[i])))
if opts.inputadjacencyobject:
for i in range(len(opts.inputadjacencyobject)):
adjacency_array.append(np.load(str(opts.inputadjacencyobject[i])))
if not np.equal(len(adjacency_array), len(masking_array)):
if not len(adjacency_array) % len(masking_array) == 0:
print "Number of adjacency objects does not match number of images."
else:
print "Error number of adjacency objects is not divisable by the number of masking arrays."
exit()
# Write tmi file
if not image_array == []:
write_tm_filetype(outname,
output_binary=opts.outputtype == 'binary',
image_array=np.vstack(image_array),
masking_array=masking_array,
maskname=maskname,
affine_array=affine_array,
vertex_array=vertex_array,
face_array=face_array,
surfname=surfname,
adjacency_array=adjacency_array,
checkname=False,
tmi_history=tmi_history)
else:
write_tm_filetype(outname,
output_binary=opts.outputtype == 'binary',
masking_array=masking_array,
maskname=maskname,
affine_array=affine_array,
vertex_array=vertex_array,
face_array=face_array,
surfname=surfname,
adjacency_array=adjacency_array,
checkname=False,
tmi_history=tmi_history)