def test_srtm_zhou2003_voxelwise(self):
results_img = SRTM_Zhou2003.volume_wrapper(
ti=self.ti,
refRegionMaskFile=self.refRegionMaskFile,
startActivity=self.startActivity,
weights=self.weights,
fwhm=self.fwhm)
'''
def test_srtm_zhou2003_fit(self):
self.model = SRTM_Zhou2003(self.t, self.dt, self.TAC, self.refTAC,
self.startActivity)
print('\nFitting SRTM_Zhou2003 with %s start activity' %
self.startActivity)
self.model.fit()
print('True BP = %.6f; estimated BP = %.6f' % (self.BP, self.model.BP))
print('True R1 = %.6f; estimated R1 = %.6f' % (self.R1, self.model.R1))
self.assertAlmostEqual(self.BP, self.model.BP, delta=1e-4)
self.assertAlmostEqual(self.R1, self.model.R1, delta=1e-2)
def test_srtm_zhou2003_fit(self, BP, R1, startActivity):
self.t, self.dt, self.TAC, self.refTAC = generate_fakeTAC_SRTM(BP, R1)
self.model = SRTM_Zhou2003(self.t,
self.dt,
self.TAC,
self.refTAC,
startActivity=startActivity)
print('\nFitting SRTM_Zhou2003 with %s start activity' % startActivity)
self.model.fit()
print('True BP = %.6f; estimated BP = %.6f; percent error = %.1E' % \
(BP, self.model.results['BP'],
100*abs(BP-self.model.results['BP'])/BP))
print('True R1 = %.6f; estimated R1 = %.6f; percent error = %.1E' % \
(R1, self.model.results['R1'],
100*abs(R1-self.model.results['R1'])/R1))
self.assertAlmostEqual(BP, self.model.results['BP'], delta=5e-3)
self.assertAlmostEqual(R1, self.model.results['R1'], delta=5e-2)
# In[7]:
# Initialize SRTM Lammerstma 1996 model
mdl_lammertsma = SRTM_Lammertsma1996(t,
dt,
TAC_matrix,
refTAC,
time_unit='min')
# fit model
mdl_lammertsma.fit()
# In[8]:
# Initialize SRTM Zhou 2003 model
mdl_zhou = SRTM_Zhou2003(t, dt, TAC_matrix, refTAC, time_unit='min')
mdl_zhou.fit()
# In[9]:
# we now take advantage of the spatial constraint capabilities of Zhou model
# Initialize SRTM Zhou 2003 model
mdl_zhou_spatial_constraint = SRTM_Zhou2003(t,
dt,
TAC_matrix,
refTAC,
time_unit='min')
# we first reorganize the TAC data in a 4-D matrix and apply Gaussian smoothing to
# get model results mdl_lammertsma.results # In[91]: mdl_gunn = SRTM_Gunn1997(t, dt, TAC, refTAC, time_unit='min') mdl_gunn.fit() mdl_gunn.results # In[307]: # Initialize SRTM Zhou 2003 model mdl_zhou = SRTM_Zhou2003(t, dt, TAC, refTAC, time_unit='min') mdl_zhou.fit() mdl_zhou.results # In[296]: # Generate noisy simulations by adding normal noise -- I don't think this is a good way pct_noise = np.array([0, 5, 10, 15, 20, 25, 30]) TAC_matrix = TAC + np.random.normal(0, np.outer(TAC, pct_noise / 100).T) # In[297]: fig, ax = plt.subplots()
local_install_path = 'H:\gitrepos\kineticmodel'
data_path = local_install_path + '/data/nru/'
# loading real volumetric data
inputTac_filename = os.path.join(data_path, 'input.mni305.2mm.sm6.nii.gz')
cerebMask_filename = os.path.join(data_path, 'cereb.mni305.2mm.nii.gz')
tim_filename = os.path.join(data_path, 'info_tim.csv')
inputTac = nib.load(inputTac_filename)
cerebMask = nib.load(cerebMask_filename)
# using Zhou's algorithm to run it
# Note: The srtm Lammertsmaa implementation is very slow in the volume and hence not shown here
results_img = SRTM_Zhou2003.volume_wrapper(
timeSeriesImgFile=inputTac_filename,
frameTimingCsvFile=tim_filename,
refRegionMaskFile=cerebMask_filename,
time_unit='s',
startActivity='flat',
fwhm=(2 * np.sqrt(2 * np.log(2))) * 5)
# In[2]:
results_img.keys()
BP_img = results_img['BP']
import matplotlib.pyplot as plt
get_ipython().magic('matplotlib inline')
plt.imshow(BP_img[:, :, 20])
# In[3]:
ax.set_title('Real PET data');
ax.legend();
# In[5]:
# Initialize SRTM Lammerstma 1996 model
mdl_lammertsma = SRTM_Lammertsma1996(t, dt, TAC, refTAC, time_unit='s')
# fit model
mdl_lammertsma.fit();
# get model results
mdl_lammertsma.results
# In[6]:
# Initialize SRTM Zhou 2003 model
mdl_zhou = SRTM_Zhou2003(t, dt, TAC, refTAC, time_unit='s')
mdl_zhou.fit();
mdl_zhou.results
# In[ ]: