This pipeline opens the result of ICAalamelodic.m, lets the user interactively label the components that look like neuronal activity (rather than movement artefacts or noise), sort them by label, plots a final summary for the chosen components, and save the reordered maps and time series.
In [1]:
import matplotlib
import numpy as np
import matplotlib.pyplot as plt
from scipy import stats
from scipy import io
%matplotlib inline
import pylab
Open time series
In [2]:
import scipy.io as sio
In [3]:
# from http://stackoverflow.com/questions/3579568/choosing-a-file-in-python-with-simple-dialog
from Tkinter import Tk
from tkFileDialog import askopenfilename
Tk().withdraw() # we don't want a full GUI, so keep the root window from appearing
filename = askopenfilename() # show an "Open" dialog box and return the path to the selected file
print(filename)
In [4]:
Ua=sio.loadmat(filename)
In [5]:
DT=Ua['TSmean']
In [6]:
DT.shape
Out[6]:
In [7]:
S1=DT.shape
In [8]:
DTmean=np.zeros(S1)
DTvar=np.zeros(S1)
Var=np.zeros(S1[1])
In [9]:
for i in range(S1[1]):
DTmean[:,i]=DT[:,i]-np.mean(DT[:,i],0)
In [10]:
for i in range(S1[1]):
Var[i]=np.sqrt(np.var(DTmean[:,i]))
DTvar[:,i]=DTmean[:,i]/Var[i]
In [11]:
DTvar.shape
Out[11]:
open maps
In [12]:
import nibabel as nb
In [13]:
# from http://stackoverflow.com/questions/3579568/choosing-a-file-in-python-with-simple-dialog
from Tkinter import Tk
from tkFileDialog import askopenfilename
Tk().withdraw() # we don't want a full GUI, so keep the root window from appearing
filename2 = askopenfilename() # show an "Open" dialog box and return the path to the selected file
print(filename2)
In [14]:
img1 = nb.load(filename2)
In [15]:
data = img1.get_data()
In [16]:
S=data.shape
In [17]:
S
Out[17]:
In [18]:
Demean=np.zeros(S)
Dmaps=np.zeros(S)
Dvar=np.zeros(S)
Var=np.zeros(S[3])
D2=np.zeros([S[0],S[1],5,S[3]])
Tvar=np.zeros(S[3])
Transform the maps to have zero mean
In [19]:
for i in range(S[3]):
Demean[:,:,:,i]=data[:,:,:,i]-np.mean(np.mean(np.mean(data[:,:,:,i],0),0),0)
Transform the maps to have unit variance and zscore
In [20]:
for i in range(S[3]):
Dsq=np.reshape(Demean[:,:,:,i],S[0]*S[1]*S[2])
Var[i]=np.sqrt(np.var(Dsq))
Dvar=Demean[:,:,:,i]/Var[i]
Dmaps[:,:,:,i]=Dvar-2.5
Dmaps[Dmaps<0]=0
Order ICs by variance
In [21]:
datao=data
Dmapso=Dmaps
In [22]:
plt.plot(Var)
Out[22]:
In [23]:
my_cmap=plt.cm.jet
my_cmap.set_bad(alpha=0)
Good_ICs=np.zeros(S[3])
Label_ICs=[]
pylab.rcParams['figure.figsize'] = (13, 2.5)
In [24]:
Dtemp=data[:,:,:,0]
In [25]:
%%javascript
IPython.OutputArea.auto_scroll_threshold =4000;
In [26]:
if S[2]>5:
Nstack=5
Int100=[(i+1)*100/Nstack for i in range(Nstack)]
Percs=np.percentile(range(S[2]),Int100)
Indices=np.split(range(S[2]),Percs)
D1=np.zeros([S[0],S[1],Nstack])
Dmean=Dtemp[:,:,range(Nstack)]
for i in range(Nstack):
Vmean=np.mean(Dtemp[:,:,Indices[i]],2)
#Dmean[:,:,i]=np.max(Vmean,0)
Dmean[:,:,i]=Vmean
else:
Nstack=S[2]
D1=np.zeros([S[0],S[1],S[2]])
Dmean=data[:,:,range(S[2])]
Dmean=np.squeeze(Dtemp[:,:,:])
In [27]:
DTvar.shape
Out[27]:
In [28]:
S
Out[28]:
In [32]:
# from http://stackoverflow.com/questions/3579568/choosing-a-file-in-python-with-simple-dialog
Tk().withdraw() # we don't want a full GUI, so keep the root window from appearing
filename = askopenfilename() # show an "Open" dialog box and return the path to the selected file
print(filename)
In [33]:
Ua=sio.loadmat(filename)
Xk=Ua['Xk']
#Xk[1,:]=Ua['Walk']
In [34]:
Xk.shape
Out[34]:
In [35]:
# from http://stackoverflow.com/questions/3579568/choosing-a-file-in-python-with-simple-dialog
from Tkinter import Tk
from tkFileDialog import askopenfilename
Tk().withdraw() # we don't want a full GUI, so keep the root window from appearing
filenamet = askopenfilename() # show an "Open" dialog box and return the path to the selected file
print(filenamet)
nimt=nb.load(filenamet)
Dtemp=np.squeeze(nimt.get_data())
Dtemp.shape
if S[2]>5:
Nstack=5
Int100=[(i+1)*100/Nstack for i in range(Nstack)]
Percs=np.percentile(range(S[2]),Int100)
Indices=np.split(range(S[2]),Percs)
Dmean=np.zeros([S[0],S[1],Nstack])
#Dmean=np.squeeze(data[:,:,range(Nstack),2])
for i in range(Nstack):
Vmean=np.mean(Dtemp[:,:,Indices[i]],2)
Dmean[:,:,i]=Vmean
plt.imshow(Vmean,cmap=plt.cm.gray)
Out[35]:
In [36]:
Xk=Xk.T
In [37]:
Xk.shape
Out[37]:
In [38]:
Label_ICs=[]
In [42]:
plt.plot(Xk[(range(3,6)),:].T)
Out[42]:
In [ ]:
### Label_ICs=[]
for j in range(S[3]):
if S[2]>5:
for i in range(Nstack):
V=Dmaps[:,:,Indices[i],j]
D1[:,:,i]=np.max(V,2)
D2[:,:,:,j]=D1
D1[D1==0]=np.nan
else:
for i in range(S[2]):
V=Dmaps[:,:,i,j]
D1[:,:,i]=V
print(j)
for i in range(Nstack):
plt.subplot(1,5,i+1)
plt.imshow(Dmean[:,:,i],cmap=plt.cm.gray)
plt.imshow(D1[:,:,i], cmap=my_cmap,interpolation='none')
frame1 = plt.gca()
frame1.axes.get_xaxis().set_visible(False)
frame1.axes.get_yaxis().set_visible(False)
plt.show()
# plt.plot(TS_ROI[Order[j],:])
plt.plot(DTvar[:,j])
plt.plot(Xk[3,:]/np.std(Xk[3,:])+0.5,color=(1,0,0))
plt.plot(Xk[4,:]/np.std(Xk[4,:])+0.5,color=(0,1,0))
plt.plot(Xk[5,:]/np.std(Xk[5,:])+0.5,color=(0.5,0.5,0))
#plt.plot(Xk[3,:]/np.std(Xk[1,:])+0.5,color=(0,0.5,1))
plt.show()
Label_ICs.append(raw_input())
if Label_ICs[j]=='':
Good_ICs[j]=0
else:
Good_ICs[j]=1
In [56]:
plt.plot(Xk[3,:])
Out[56]:
In [43]:
#zip(range(S[3]),Label_ICs)
In [44]:
set(Label_ICs)
Out[44]:
In [46]:
Xksmoothed=np.zeros(Xk.shape)
Xksmoothed[3,:]=np.convolve(Xk[3,999:Xk.shape[1]-1000],np.ones(2000)/2000)
Xksmoothed[4,:]=np.convolve(Xk[4,999:Xk.shape[1]-1000],np.ones(2000)/2000)
Xksmoothed[5,:]=np.convolve(Xk[5,999:Xk.shape[1]-1000],np.ones(2000)/2000)
Xkdff=Xk-Xksmoothed
In [47]:
Rsq=np.zeros((1,S[3]))
Betas=np.zeros((3,S[3]))
In [48]:
from sklearn import linear_model
In [49]:
algorithm = linear_model.LinearRegression()
In [50]:
for j in range(S[3]):
model = algorithm.fit(Xkdff[range(3,6),:].T, DT[:,j])
Betas[:,j] = model.coef_
Rsq[:,j] = model.score(Xkdff[range(3,6),:].T,DT[:,j])
In [59]:
for i in range(S[3]):
if Label_ICs[i]!='':
print(Label_ICs[i], Betas[0,i]*10000)
In [55]:
for i in range(S[3]):
if Label_ICs[i]!='':
print(Label_ICs[i], Rsq[:,i])
In [53]:
List1=[(Label_ICs[i],i) for i in range(S[3])]
Newlist=sorted(List1, key=lambda List1: List1[0])
Neworder=[Newlist[i][1] for i in range(S[3])]
NewDT=DTvar[:,Neworder[:]].T
for j in range(len(Neworder)):
A=NewDT[:,j]
V=np.sqrt(np.var(A))
NewDT[:,j]=A/V
C1=np.zeros([16,3])
C1[0][:]=(1,0,0)
C1[1][:]=(0,1,0)
C1[2][:]=(0,0,1)
C1[3][:]=(0.8,0.8,0)
C1[4][:]=(0,1,1)
C1[5][:]=(1,0,1)
C1[6][:]=(1,0.5,0)
C1[7][:]=(0,1,0.5)
C1[8][:]=(0.5,0,1)
C1[9][:]=(0.8,0.8,0.5)
C1[10][:]=(0.5,1,1)
C1[11][:]=(1,0.5,1)
C1[12]=(0.5,0.5,0.5)
C1[13]=(0.2,0.5,0.5)
C1[14]=(0.5,0.2,0.5)
C1[15]=(0.5,0.5,0.2)
h=3
Newmaps=Dmaps[:,:,:,Neworder[:]]
L=len(set([Label_ICs[Neworder[i]] for i in range(len(Neworder))]))
Regionmaps=np.zeros([S[0],S[1],L,3])
Datasort=np.zeros([S[0],S[1],S[2],L,3])
Regionname=[]
DMapsordered=Dmapso[:,:,:,Neworder[:]]
j=0
i=0
k=Label_ICs[Neworder[0]]
m=0
Regionname.append(Label_ICs[Neworder[i]])
for i in range(len(Neworder)):
#C2=C1[i%6][:]
for l in range(3):
M=np.max(np.squeeze(np.reshape(Newmaps[:,:,:,i],S[0]*S[1]*S[2])))
Regionmaps[:,:,j,l]=Regionmaps[:,:,j,l]+0.6*np.max(DMapsordered[:,:,:,i],2)*C1[i%12+1][l]/M
Datasort[:,:,:,j,l]=Datasort[:,:,:,j,l]+Dmaps[:,:,:,Neworder[i]]*C1[i%15+1][l]
i=i+1
m=m+1
if i<len(Neworder):
k1=Label_ICs[Neworder[i]]
if k1 != k:
j=j+1
k=k1
m=0
Regionname.append(Label_ICs[Neworder[i]])
pylab.rcParams['figure.figsize'] = (14, 5)
import scipy
from scipy import ndimage
j=0
m=0
L=0
k=Label_ICs[Neworder[0]]
for i in range(len(Neworder)):
m=m+1
if i<len(Neworder):
k1=Label_ICs[Neworder[i]]
if k1 != k:
k=k1
m=0
plt.show()
plt.figure(2*j+1)
Rotated_Plot = ndimage.rotate(Regionmaps[:,:,j], -90)
IM=plt.imshow(Rotated_Plot)
frame1 = plt.gca()
frame1.axes.get_xaxis().set_visible(False)
frame1.axes.get_yaxis().set_visible(False)
j=j+1
plt.figure(2*j)
plt.plot(Xk[0,:]/np.std(Xk[0,:])+0.5,color=(1,0,0))
plt.plot(Xk[1,:]/np.std(Xk[1,:])+0.5,color=(0,1,0))
plt.plot(Xk[2,:]/np.std(Xk[1,:])+0.5,color=(0.5,0.5,0))
#plt.plot(Xk[3,:]/np.std(Xk[1,:])+0.5,color=(0,0.5,1))
plt.plot(NewDT[i,:]+h*m,color=C1[i%12+1][:])
print(Neworder[i])
print(Rsq[:,Neworder[i]])
print(Betas[:,Neworder[i]])
plt.figure(2*j+1)
Rotated_Plot = ndimage.rotate(Regionmaps[:,:,j], -90)
IM=plt.imshow(Rotated_Plot)
frame1 = plt.gca()
frame1.axes.get_xaxis().set_visible(False)
frame1.axes.get_yaxis().set_visible(False)
print(Neworder)
In [ ]:
plt.plot(Xkdff.T)
In [ ]: