def create_scanner_k_space(im,
N,
P=2,
pctg=0.25,
dirData=False,
dirs=None,
radius=0.2,
cyl=[0],
style='mult',
pft=False,
ext=0.5):
'''
Read in the data, size, and the directions (if they exist) so that we can create a
retrospectively sampled set of data for testing.
'''
# Create a pdf so that we can use it to make a starting point
pdf = samp.genPDF(N[-2:],
P,
pctg,
radius=radius,
cyl=[1, N[-2], N[-1]],
style='mult',
pft=pft,
ext=0.5)
# Generate the sampling scheme, depending on whether or not
if dirData:
if dirs is None:
raise ValueError(
'If we have directional data, you need to feed this into the function'
)
k = d.dirPDFSamp([int(dirs.shape[0]), N[-2], N[-1]],
P=2,
pctg=pctg,
radius=radius,
dirs=dirs,
cyl=True,
taper=0.25)[0]
else:
k = samp.genSampling(pdf, 50, 2)[0].astype(int)
# Since our functions are built to work in 3D datasets, here we
# make sure that N and things are all in 3D
if len(N) == 2:
N = np.hstack([1, N])
k = k.reshape(N)
im = im.reshape(N)
elif len(N) == 3:
if k.ndim == 2:
k = k.reshape(np.hstack([1, N[-2:]])).repeat(N[0], 0)
k = np.fft.fftshift(k, axes=(-2, -1))
# Convert the image data into k-space
ph_ones = np.ones(N, complex)
dataFull = tf.fft2c(im, ph=ph_ones, axes=(-2, -1))
# Apply our sampling
data = k * dataFull
# Now we need to calculate the phase in order to deal with the undersampled image and the
# non perfect cancellation of terms
#filtdata = gaussian_filter(im_scan_wph.real,0,0) + 1j*gaussian_filter(im_scan_wph.imag,0,0)
#ph_scan = np.exp(1j*np.angle(filtdata.conj()))
im_scan_wph = tf.ifft2c(data, ph=ph_ones)
ph_scan = np.angle(
gaussian_filter(im_scan_wph.real, 0) +
1.j * gaussian_filter(im_scan_wph.imag, 0))
ph_scan = np.exp(1j * ph_scan)
im_scan = tf.ifft2c(data, ph=ph_scan)
pdfDiv = pdf.copy()
pdfZeros = np.where(pdf < 1e-4)
pdfDiv[pdfZeros] = 1
datadc = data / pdfDiv
return dataFull, data, datadc, pdf, k, im_scan, ph_scan
im = np.load('/home/asalerno/Documents/pyDirectionCompSense/phantom/diffusionPhantomSmoothedSNR1000.npy')
#im = np.load('/home/asalerno/Documents/pyDirectionCompSense/directionData/singleSlice_30dirs.npy')
#im=im/np.max(abs(im))
minval = np.min(abs(im))
maxval = np.max(abs(im))
N = np.array(im.shape) # image Size
szFull = im.size
P = 2
nSteps = 4
print('k for Directions')
#kDir = samp.genSamplingDir(img_sz=N, dirFile=dirFile, pctg=pctg, cyl=[1], radius=radius,
#nmins=nmins, engfile=engfile)
kDir = d.dirPDFSamp(N,P=2,pctg=0.25,radius=0.2,dirs=dirs,cyl=True,taper=0.25)
pdf = samp.genPDF(N[-2:], P, pctg, radius=radius, cyl=[1, N[-2], N[-1]], style='mult', pft=pft,ext=0.5)
k = np.zeros(N)
print('k for non')
for i in xrange(N[0]):
np.random.seed(int(i*abs(np.random.randn(1)*np.random.randn(1))))
print(i)
k[i,:,:] = samp.genSampling(pdf, 50, 2)[0].astype(int)
ph_ones = np.ones(N[-2:], complex)
dataFull = np.zeros(N,complex)
ph_scan = np.zeros(N, complex)
#pdf = samp.genPDF(N[-2:], P, pctg, radius=radius, cyl=[1, N[-2], N[-1]], style='mult', pft=pft,ext=0.5)
#k = d.dirPDFSamp([int(dirs.shape[0]), 294, 294], P=2, pctg=0.25, radius=0.2, dirs=dirs, cyl=True, taper=0.25)
#k = np.load('/hpf/largeprojects/MICe/asalerno/pyDirectionCompSense/data/comb_k_25.npy')
k = np.zeros([dirs.shape[0], N[-2], N[-1]])
pdf = samp.genPDF(N[-2:],
P,
pctg,
radius=radius,
cyl=[1, N[-2], N[-1]],
style='mult',
pft=pft,
ext=0.5)
k = d.dirPDFSamp([int(dirs.shape[0]), N[-2], N[-1]],
P=2,
pctg=0.25,
radius=0.2,
dirs=dirs,
cyl=True,
taper=0.25)[0]
if len(N) == 2:
N = np.hstack([1, N])
k = k.reshape(N)
im = im.reshape(N)
elif len(N) == 3:
if not np.all(k.shape == N):
k = k.reshape(np.hstack([1, N[-2:]]))
k = [k[0] for _ in range(im.shape[0])]
k = np.stack(k, axis=0)
#k = k.reshape(np.hstack([1,N[-2:]])).repeat(N[0],0)
#im=im/np.max(abs(im))
minval = np.min(abs(im))
maxval = np.max(abs(im))
N = np.array(im.shape) # image Size
szFull = im.size
P = 2
nSteps = 4
print('k for Directions')
#kDir = samp.genSamplingDir(img_sz=N, dirFile=dirFile, pctg=pctg, cyl=[1], radius=radius,
#nmins=nmins, engfile=engfile)
kDir = d.dirPDFSamp(N,
P=2,
pctg=0.25,
radius=0.2,
dirs=dirs,
cyl=True,
taper=0.25)
pdf = samp.genPDF(N[-2:],
P,
pctg,
radius=radius,
cyl=[1, N[-2], N[-1]],
style='mult',
pft=pft,
ext=0.5)
k = np.zeros(N)
print('k for non')
#--------------------------------------------------------------
# If our special cases have an effect here, PLEASE put the
# original code within these two bars.
#pdf = samp.genPDF(N[-2:], P, pctg, radius=radius, cyl=[1, N[-2], N[-1]], style='mult', pft=pft,ext=0.5)
#if pft:
#print('Partial Fourier sampling method used')
#k = samp.genSampling(pdf, 50, 2)[0].astype(int)
#d.dirPDFSamp(N,P=2,pctg=0.25,radius=0.2,dirs=dirs,cyl=True,taper=0.25)
#--------------------------------------------------------------
#k = np.zeros([int(np.ceil(N[0]/dirs.shape[0])*dirs.shape[0]), 294, 294])
k = np.zeros([dirs.shape[0], 294, 294])
pdf = samp.genPDF(N[-2:], P, pctg, radius=radius, cyl=[1, N[-2], N[-1]], style='mult', pft=pft,ext=0.5)
k = d.dirPDFSamp([int(dirs.shape[0]), 294, 294], P=2, pctg=0.25, radius=0.2, dirs=dirs, cyl=True, taper=0.25)
k = [k[0] for _ in range(im.shape[0])]
k = np.stack(k,axis=0)
# SPECIAL CASE END
# Since our functions are built to work in 3D datasets, here we
# make sure that N and things are all in 3D
if len(N) == 2:
N = np.hstack([1, N])
k = k.reshape(N)
im = im.reshape(N)
elif len(N) == 3:
if k.ndim == 2:
k = k.reshape(np.hstack([1,N[-2:]])).repeat(N[0],0)