def bcg_filt_average_nh_fit(x, ts, m=1024, k=20, Fs=1000.0):
"""Filter BCG-artifacts from simultaneous recordings of EEG and fMRI using AAS
(average artifact subtraction), but in local neighborhood. Additionally perform a fit
of the template to each artifact."""
def residuals(p, y, x):
"""Residuals used for leastsq-fitting"""
factor = p[0]
#err = (y*factor-offset)-x
err = (y * factor) - x
return abs(err**2).sum()
assert len(x.shape) in [1, 2], "x must be a 1d- or 2d-array"
if len(x.shape) == 1:
x = x.reshape((x.shape[0], 1))
x_hp = _highpass(x, Fs=Fs)
window = n.hanning(m)
#Other possibilities for window:
#n.ones((m),"d")
#n.hamming(m)
#n.bartlett(m)
ts_bcg = ts
for ch_num in range(x.shape[1]):
if debug:
print "Get all BCG-artifacts of channel %i in one array" % ch_num
bcgs = n.zeros((m, len(ts_bcg)), "d")
for i in range(bcgs.shape[1]):
#print eeg[ts_bcg[i]-m/2:ts_bcg[i]+m/2,14].shape
try:
bcgs[:, i] = x_hp[ts_bcg[i] - m / 2:ts_bcg[i] + m / 2, ch_num]
except ValueError, e:
print "Value Error, probably shape missmatch", e
if debug:
print "Find nearest neighbors"
neighbors = find_n_nearest_neighbors(bcgs.T, k)
for i in range(bcgs.shape[1]):
boolslice = n.zeros((bcgs.shape[1]), n.bool)
for ni, neighbor in enumerate(neighbors[i]):
boolslice[neighbor] = True
try:
template = bcgs[:, boolslice].mean(axis=1) * window
p_lsq = fmin(residuals, [1],
args=(template,
x[ts_bcg[i] - m / 2:ts_bcg[i] + m / 2,
ch_num]),
disp=False)
factor = p_lsq[0]
#print i, p_lsq
if factor > 0:
#template = template*factor-offset
template = template * factor
x[ts_bcg[i] - m / 2:ts_bcg[i] + m / 2, ch_num] -= template
except ValueError, e:
print "Value Error, probably shape missmatch", e
def bcg_filt_pca_nh(x, ts, m=1024, k=20):
"""Filter BCG-artifacts from simultaneous recordings of EEG and fMRI using PCA
(principal component analysis), but in local neighborhood"""
assert len(x.shape) in [1, 2], "x must be a 1d- or 2d-array"
if len(x.shape) == 1:
x = x.reshape((x.shape[0], 1))
x = _highpass(x, Fs=Fs)
window = n.hanning(m)
#Other possibilities for window:
#n.ones((m),"d")
#n.hamming(m)
#n.bartlett(m)
ts_bcg = ts
for ch_num in range(x.shape[1]):
if debug:
print "Get all BCG-artifacts of channel %i in one array" % ch_num
bcgs = n.zeros((m, len(ts_bcg)), "d")
for i in range(bcgs.shape[1]):
#print eeg[ts_bcg[i]-m/2:ts_bcg[i]+m/2,14].shape
try:
bcgs[:, i] = x[ts_bcg[i] - m / 2:ts_bcg[i] + m / 2, ch_num]
except ValueError, e:
print "Value Error, probably shape missmatch", e
if debug:
print "Find nearest neighbors"
neighbors = find_n_nearest_neighbors(bcgs.T, k)
for i in range(bcgs.shape[1]):
boolslice = n.zeros((bcgs.shape[1]), n.bool)
for ni, neighbor in enumerate(neighbors[i]):
boolslice[neighbor] = True
NH = bcgs[:, boolslice]
components = pcafilt.unmix(NH)
for k in range(components.shape[1]):
if components[:, k].std() > 100.0: #acc > 0.03:
if debug:
print "Channel %i, Component %i, std %f, mean(abs) %f" % (
ch_num, k, components[:, k].std(),
abs(components[:, k]).mean()), "removed"
components[:, k] = n.zeros(components[:, k].shape, "d")
else:
if k == 0:
if debug:
print "Channel %i, Component %i, std %f, mean(abs) %f" % (
ch_num, k, components[:, k].std(),
abs(components[:, k]).mean()), "not removed"
NH = pcafilt.mix(components)
try:
x[ts_bcg[i] - m / 2:ts_bcg[i] + m / 2, ch_num] = NH[:, 0]
except ValueError, e:
print "Value Error, probably shape missmatch", e
def bcg_filt_average_nh_fit(x,ts,m=1024,k=20,Fs=1000.0):
"""Filter BCG-artifacts from simultaneous recordings of EEG and fMRI using AAS
(average artifact subtraction), but in local neighborhood. Additionally perform a fit
of the template to each artifact."""
def residuals(p,y,x):
"""Residuals used for leastsq-fitting"""
factor = p[0]
#err = (y*factor-offset)-x
err = (y*factor)-x
return abs(err**2).sum()
assert len(x.shape) in [1,2], "x must be a 1d- or 2d-array"
if len(x.shape) == 1:
x=x.reshape((x.shape[0],1))
x_hp = _highpass(x,Fs=Fs)
window = n.hanning(m)
#Other possibilities for window:
#n.ones((m),"d")
#n.hamming(m)
#n.bartlett(m)
ts_bcg = ts
for ch_num in range(x.shape[1]):
if debug:
print "Get all BCG-artifacts of channel %i in one array" % ch_num
bcgs = n.zeros((m,len(ts_bcg)),"d")
for i in range(bcgs.shape[1]):
#print eeg[ts_bcg[i]-m/2:ts_bcg[i]+m/2,14].shape
try:
bcgs[:,i] = x_hp[ts_bcg[i]-m/2:ts_bcg[i]+m/2,ch_num]
except ValueError,e:
print "Value Error, probably shape missmatch", e
if debug:
print "Find nearest neighbors"
neighbors = find_n_nearest_neighbors(bcgs.T,k)
for i in range(bcgs.shape[1]):
boolslice = n.zeros((bcgs.shape[1]),n.bool)
for ni,neighbor in enumerate(neighbors[i]):
boolslice[neighbor] = True
try:
template = bcgs[:,boolslice].mean(axis=1) * window
p_lsq = fmin(residuals,[1],args=(template,x[ts_bcg[i]-m/2:ts_bcg[i]+m/2,ch_num]),disp=False)
factor = p_lsq[0]
#print i, p_lsq
if factor>0:
#template = template*factor-offset
template = template*factor
x[ts_bcg[i]-m/2:ts_bcg[i]+m/2,ch_num] -= template
except ValueError,e:
print "Value Error, probably shape missmatch", e
def bcg_filt_average_nh(x, ts, m=1024, k=20, Fs=1000.0):
"""Filter BCG-artifacts from simultaneous recordings of EEG and fMRI using AAS
(average artifact subtraction), but in local neighborhood"""
assert len(x.shape) in [1, 2], "x must be a 1d- or 2d-array"
if len(x.shape) == 1:
x = x.reshape((x.shape[0], 1))
x = _highpass(x, Fs=Fs)
window = n.hanning(m)
#Other possibilities for window:
#n.ones((m),"d")
#n.hamming(m)
#n.bartlett(m)
ts_bcg = ts
for ch_num in range(x.shape[1]):
if debug:
print "Get all BCG-artifacts of channel %i in one array" % ch_num
bcgs = n.zeros((m, len(ts_bcg)), "d")
for i in range(bcgs.shape[1]):
#print eeg[ts_bcg[i]-m/2:ts_bcg[i]+m/2,14].shape
try:
bcgs[:, i] = x[ts_bcg[i] - m / 2:ts_bcg[i] + m / 2, ch_num]
except ValueError, e:
print "Value Error, probably shape missmatch", e
if debug:
print "Find nearest neighbors"
neighbors = find_n_nearest_neighbors(bcgs.T, k)
# if debug:
# print "Testing neighborhood: Do some plots..."
# for i in range(0,neighbors.shape[0],50):
# boolslice = n.zeros((bcgs.shape[1]),n.bool)
# #neighborhood = n.zeros((m,k),"d")
# for j in range(k):
# boolslice[neighbors[i,j]] = True
# neighborhood = bcgs[:,boolslice]
#p.plot(neighborhood.mean(axis=1))
#p.plot(bcgs[:,:].mean(axis=1))
for i in range(bcgs.shape[1]):
boolslice = n.zeros((bcgs.shape[1]), n.bool)
for ni, neighbor in enumerate(neighbors[i]):
boolslice[neighbor] = True
try:
x[ts_bcg[i] - m / 2:ts_bcg[i] + m / 2,
ch_num] -= bcgs[:, boolslice].mean(axis=1) * window
except ValueError, e:
print "Value Error, probably shape missmatch", e
def bcg_filt_average_nh(x,ts,m=1024,k=20,Fs=1000.0):
"""Filter BCG-artifacts from simultaneous recordings of EEG and fMRI using AAS
(average artifact subtraction), but in local neighborhood"""
assert len(x.shape) in [1,2], "x must be a 1d- or 2d-array"
if len(x.shape) == 1:
x=x.reshape((x.shape[0],1))
x = _highpass(x,Fs=Fs)
window = n.hanning(m)
#Other possibilities for window:
#n.ones((m),"d")
#n.hamming(m)
#n.bartlett(m)
ts_bcg = ts
for ch_num in range(x.shape[1]):
if debug:
print "Get all BCG-artifacts of channel %i in one array" % ch_num
bcgs = n.zeros((m,len(ts_bcg)),"d")
for i in range(bcgs.shape[1]):
#print eeg[ts_bcg[i]-m/2:ts_bcg[i]+m/2,14].shape
try:
bcgs[:,i] = x[ts_bcg[i]-m/2:ts_bcg[i]+m/2,ch_num]
except ValueError,e:
print "Value Error, probably shape missmatch", e
if debug:
print "Find nearest neighbors"
neighbors = find_n_nearest_neighbors(bcgs.T,k)
# if debug:
# print "Testing neighborhood: Do some plots..."
# for i in range(0,neighbors.shape[0],50):
# boolslice = n.zeros((bcgs.shape[1]),n.bool)
# #neighborhood = n.zeros((m,k),"d")
# for j in range(k):
# boolslice[neighbors[i,j]] = True
# neighborhood = bcgs[:,boolslice]
#p.plot(neighborhood.mean(axis=1))
#p.plot(bcgs[:,:].mean(axis=1))
for i in range(bcgs.shape[1]):
boolslice = n.zeros((bcgs.shape[1]),n.bool)
for ni,neighbor in enumerate(neighbors[i]):
boolslice[neighbor] = True
try:
x[ts_bcg[i]-m/2:ts_bcg[i]+m/2,ch_num] -= bcgs[:,boolslice].mean(axis=1) * window
except ValueError,e:
print "Value Error, probably shape missmatch", e
def bcg_filt_pca_nh(x,ts,m=1024,k=20):
"""Filter BCG-artifacts from simultaneous recordings of EEG and fMRI using PCA
(principal component analysis), but in local neighborhood"""
assert len(x.shape) in [1,2], "x must be a 1d- or 2d-array"
if len(x.shape) == 1:
x=x.reshape((x.shape[0],1))
x = _highpass(x,Fs=Fs)
window = n.hanning(m)
#Other possibilities for window:
#n.ones((m),"d")
#n.hamming(m)
#n.bartlett(m)
ts_bcg = ts
for ch_num in range(x.shape[1]):
if debug:
print "Get all BCG-artifacts of channel %i in one array" % ch_num
bcgs = n.zeros((m,len(ts_bcg)),"d")
for i in range(bcgs.shape[1]):
#print eeg[ts_bcg[i]-m/2:ts_bcg[i]+m/2,14].shape
try:
bcgs[:,i] = x[ts_bcg[i]-m/2:ts_bcg[i]+m/2,ch_num]
except ValueError,e:
print "Value Error, probably shape missmatch", e
if debug:
print "Find nearest neighbors"
neighbors = find_n_nearest_neighbors(bcgs.T,k)
for i in range(bcgs.shape[1]):
boolslice = n.zeros((bcgs.shape[1]),n.bool)
for ni,neighbor in enumerate(neighbors[i]):
boolslice[neighbor] = True
NH = bcgs[:,boolslice]
components = pcafilt.unmix(NH)
for k in range(components.shape[1]):
if components[:,k].std()>100.0:#acc > 0.03:
if debug:
print "Channel %i, Component %i, std %f, mean(abs) %f" % (ch_num,k,components[:,k].std(), abs(components[:,k]).mean()), "removed"
components[:,k] = n.zeros(components[:,k].shape,"d")
else:
if k==0:
if debug:
print "Channel %i, Component %i, std %f, mean(abs) %f" % (ch_num,k,components[:,k].std(), abs(components[:,k]).mean()), "not removed"
NH = pcafilt.mix(components)
try:
x[ts_bcg[i]-m/2:ts_bcg[i]+m/2,ch_num] = NH[:,0]
except ValueError,e:
print "Value Error, probably shape missmatch", e
def bcg_filt_wavelet(x,
ts,
m=1024,
wave_name="coif4",
k=20,
lambda_r=1,
Fs=1000.0):
"""Filter BCG-artifacts from simultaneous recordings of EEG and fMRI"""
assert len(x.shape) in [1, 2], "x must be a 1d- or 2d-array"
if len(x.shape) == 1:
x = x.reshape((x.shape[0], 1))
x = _highpass(x, Fs=Fs)
window = n.hanning(m)
#Other possibilities for window:
#n.ones((m),"d")
#n.hamming(m)
#n.bartlett(m)
ts_bcg = ts
for ch_num in range(x.shape[1]):
if debug:
print "Get all BCG-artifacts of channel %i in one array" % ch_num
bcgs = n.zeros((m, len(ts_bcg)), "d")
for i in range(bcgs.shape[1]):
#print eeg[ts_bcg[i]-m/2:ts_bcg[i]+m/2,14].shape
try:
bcgs[:, i] = x[ts_bcg[i] - m / 2:ts_bcg[i] + m / 2,
ch_num] * window
except ValueError, e:
print "Value Error, probably shape missmatch", e
if debug:
print "Find nearest neighbors"
neighbors = find_n_nearest_neighbors(bcgs.T, k)
# if debug:
# print "Testing neighborhood: Do some plots..."
# for i in range(0,neighbors.shape[0],50):
# boolslice = n.zeros((bcgs.shape[1]),n.bool)
# #neighborhood = n.zeros((m,k),"d")
# for j in range(k):
# boolslice[neighbors[i,j]] = True
# neighborhood = bcgs[:,boolslice]
#p.plot(neighborhood.mean(axis=1))
#p.plot(bcgs[:,:].mean(axis=1))
if debug:
print "Wavelet transform"
wavelet_tr_s_lin = wavedec_lin(bcgs, wave_name)
#walk through all points
#Calculate the center-of-mass and sd for all wavelets
sqrt_kp1 = n.sqrt(k + 1)
wts_to_shrink = wavelet_tr_s_lin.copy()
#print "Shrinking for lambda_r =", lambda_r
for i in range(wts_to_shrink.shape[1]):
#numShrinked.append(0)
if debug:
if i % 100 == 0:
if debug:
print "i:", i
boolslice = n.zeros((wts_to_shrink.shape[1]), n.bool)
for ni, neighbor in enumerate(neighbors[i]):
boolslice[neighbor] = True
NH = wavelet_tr_s_lin[:, boolslice]
COM = NH.mean(axis=1)
STD = NH.std(axis=1)
#apply the shrinking condition
for idx in range(wts_to_shrink.shape[0]):
if abs(COM[idx]) < (2 * lambda_r * STD[idx] / sqrt_kp1):
wts_to_shrink[idx, i] = COM[idx]
if debug:
print "Wavelet-reconstruction"
bcgs_f = waverec_lin(wts_to_shrink, wave_name)
if debug:
print "Writing back results to eeg"
#I have to subtract the filtered bcgs from the eeg as these are now the
#"pure" artifact.
for i in range(bcgs.shape[1]):
try:
x[ts_bcg[i] - m / 2:ts_bcg[i] + m / 2, ch_num] -= bcgs_f[:, i]
except ValueError, e:
print "Value Error, probably shape missmatch", e
def bcg_filt_wavelet(x,ts,m=1024,wave_name="coif4",k=20, lambda_r=1,Fs=1000.0):
"""Filter BCG-artifacts from simultaneous recordings of EEG and fMRI"""
assert len(x.shape) in [1,2], "x must be a 1d- or 2d-array"
if len(x.shape) == 1:
x=x.reshape((x.shape[0],1))
x = _highpass(x,Fs=Fs)
window = n.hanning(m)
#Other possibilities for window:
#n.ones((m),"d")
#n.hamming(m)
#n.bartlett(m)
ts_bcg = ts
for ch_num in range(x.shape[1]):
if debug:
print "Get all BCG-artifacts of channel %i in one array" % ch_num
bcgs = n.zeros((m,len(ts_bcg)),"d")
for i in range(bcgs.shape[1]):
#print eeg[ts_bcg[i]-m/2:ts_bcg[i]+m/2,14].shape
try:
bcgs[:,i] = x[ts_bcg[i]-m/2:ts_bcg[i]+m/2,ch_num]*window
except ValueError,e:
print "Value Error, probably shape missmatch", e
if debug:
print "Find nearest neighbors"
neighbors = find_n_nearest_neighbors(bcgs.T,k)
# if debug:
# print "Testing neighborhood: Do some plots..."
# for i in range(0,neighbors.shape[0],50):
# boolslice = n.zeros((bcgs.shape[1]),n.bool)
# #neighborhood = n.zeros((m,k),"d")
# for j in range(k):
# boolslice[neighbors[i,j]] = True
# neighborhood = bcgs[:,boolslice]
#p.plot(neighborhood.mean(axis=1))
#p.plot(bcgs[:,:].mean(axis=1))
if debug:
print "Wavelet transform"
wavelet_tr_s_lin = wavedec_lin(bcgs,wave_name)
#walk through all points
#Calculate the center-of-mass and sd for all wavelets
sqrt_kp1 = n.sqrt(k+1)
wts_to_shrink = wavelet_tr_s_lin.copy()
#print "Shrinking for lambda_r =", lambda_r
for i in range(wts_to_shrink.shape[1]):
#numShrinked.append(0)
if debug:
if i%100==0:
if debug:
print "i:", i
boolslice = n.zeros((wts_to_shrink.shape[1]),n.bool)
for ni,neighbor in enumerate(neighbors[i]):
boolslice[neighbor] = True
NH = wavelet_tr_s_lin[:,boolslice]
COM = NH.mean(axis=1)
STD = NH.std(axis=1)
#apply the shrinking condition
for idx in range(wts_to_shrink.shape[0]):
if abs(COM[idx])<(2*lambda_r*STD[idx]/sqrt_kp1):
wts_to_shrink[idx,i]= COM[idx]
if debug:
print "Wavelet-reconstruction"
bcgs_f = waverec_lin(wts_to_shrink,wave_name)
if debug:
print "Writing back results to eeg"
#I have to subtract the filtered bcgs from the eeg as these are now the
#"pure" artifact.
for i in range(bcgs.shape[1]):
try:
x[ts_bcg[i]-m/2:ts_bcg[i]+m/2,ch_num] -= bcgs_f[:,i]
except ValueError,e:
print "Value Error, probably shape missmatch", e