def saveResults(self, savePath, worldToWorldTransformMatrix,
imageToImageTransformMatrix):
# Save the image-to-image and the world-to-world affine registration matrices
scipy.io.savemat(
os.path.join(savePath, 'template_transforms.mat'), {
'imageToImageTransformMatrix': imageToImageTransformMatrix,
'worldToWorldTransformMatrix': worldToWorldTransformMatrix
})
# Save the template transform
inputImage = fs.Volume.read(self.imageFileName)
templateImage = fs.Volume.read(self.templateFileName)
xform = fs.LinearTransform(
convertLPSTransformToRAS(worldToWorldTransformMatrix))
xform.type = fs.LinearTransform.Type.ras
xform.source = templateImage.geometry()
xform.target = inputImage.geometry()
ltaFileName = os.path.join(savePath, 'template.lta')
print('writing template transform to %s' % ltaFileName)
xform.write(ltaFileName)
# Compute and save the talairach.xfm
xform = self.computeTalairach(imageToImageTransformMatrix)
ltaFileName = os.path.join(savePath, 'samseg.talairach.lta')
print('writing talairach transform to %s' % ltaFileName)
xform.write(ltaFileName)
# Save the coregistered template
coregistered = fs.Volume(
fs.geom.resample(templateImage.data, inputImage.shape[:3],
np.linalg.inv(imageToImageTransformMatrix)))
coregistered.copy_geometry(inputImage)
coregistered.write(os.path.join(savePath, 'template_coregistered.mgz'))
def writeImage(self, data, path, saveLabels=False):
# Read source geometry
geom = fs.Volume.read(self.imageFileNames[0]).geometry()
# Account for multi-frame volumes
if data.ndim == 4:
shape = (*geom.shape, data.shape[-1])
else:
shape = geom.shape
# Uncrop image
uncropped = np.zeros(shape, dtype=data.dtype, order='F')
uncropped[self.cropping] = data
volume = fs.Volume(uncropped, affine=geom.affine, voxsize=geom.voxsize)
if saveLabels:
volume.lut = fs.LookupTable.read_default()
volume.write(path)
def compute_intensity_stats(paths,
vname_list,
method='mean',
ranges=None,
vol_list=None):
nvols = len(vname_list)
means = np.zeros(nvols)
stds = np.zeros(nvols)
for pno, path in enumerate(paths):
for vno, vname in enumerate(vname_list):
fname = os.path.join(path, vname)
if (vol_list == None):
print('%d of %d: file %s' % (pno, len(paths), fname))
mri = fs.Volume(fname)
else:
mri = vol_list[pno][vno]
im = mri.image.astype('float64')
if (method == 'mean'):
val = im.mean()
elif method == 'histo':
if ranges is not None:
ind = np.where(
np.logical_and(im > ranges[vno, 0],
im < ranges[vno, 1]))
im = im[ind]
try:
hist, edges = np.histogram(im.flatten(), bins='auto')
except ValueError:
print('exception')
hist, edges = np.histogram(im.flatten(), bins=500)
val = edges[hist.argmax()]
del edges, hist
shape = mri.image.shape
if (vol_list == None):
del mri
means[vno] += val
stds[vno] += (val * val)
for vno in range(nvols):
means[vno] /= len(paths)
stds[vno] = np.sqrt(stds[vno] / len(paths) - means[vno] * means[vno])
return means, stds, shape
import imageio, pydicom, gdcm, load_serial_cxr
bdir = '/autofs/cluster/lcnextdata1/CCDS_CXR/CXR-Serial/def_20200413'
il, sl, sn = load_serial_cxr.load_serial_cxr(bdir)
for sno, ilist in enumerate(il):
if len(ilist) >= 2:
date_list = []
time_list = []
for ino, im in enumerate(ilist):
date_list.append(int(im.StudyDate))
if hasattr(im, 'SeriesTime'):
time_list.append(int(im.SeriesTime.split('.')[0]))
else:
time_list.append(int(im.StudyTime.split('.')[0]))
ind = np.array(date_list).argsort()
date_list2 = []
time_list2 = []
ilist2 = []
for i in ind.tolist():
date_list2.append(date_list[i])
time_list2.append(time_list[i])
ilist2.append(ilist[i])
for ino, im in enumerate(ilist2):
tokens = sl[sno][ind[ino]].split('/')
fname = '/'.join(tokens[0:-2]) + '/time%2.2d.mgz' % ino
fs.Volume(im.pixel_array.astype(np.float32)).write(fname)
