def plot_average_and_optimized_transfers():
fig = plt.figure(facecolor="white")
ax = fig.add_subplot(1, 2, 1)
compare_transfers(ax, fmean_t1lab, opt_list_t1lab[4], "Iso-set mean", "Optimized", 1, "T1", "F[T1]")
ax = fig.add_subplot(1, 2, 2)
compare_transfers(ax, fmean_t2lab, opt_list_t2lab[4], "Iso-set mean", "Optimized", 1, "T2", "F[T2]")
# Apply the transfer
t2_lab = t2.astype(np.int32)
means_t2t1, vars_t2t1 = get_mean_transfer(t2_lab, t1)
ss_t1 = means_t2t1[t2_lab]
t1_lab = t1.astype(np.int32)
means_t1t2, vars_t1t2 = get_mean_transfer(t1_lab, t2)
ss_t2 = means_t1t2[t1_lab]
# Compute residuals of locally affine fit on T1 and T2
radius = 4
residuals_nb = ccr.compute_cc_residuals(t1, t2, radius, 1)
residuals_nb = np.array(residuals_nb)
# Compute residuals of locally affine fit on F(T1) and T2
radius = 4
residuals_t2 = ccr.compute_cc_residuals(ss_t2, t2, radius, 1)
residuals_t2 = np.array(residuals_t2)
# Compute residuals of locally affine fit on T1 and F[T2]
radius = 4
residuals_t1 = ccr.compute_cc_residuals(t1, ss_t1, radius, 1)
residuals_t1 = np.array(residuals_t1)
def demo_near_dwi(idx=-1):
# Plot local linear reconstruction error from the pair of images whose corresponding
# diffusion encoding vectors' dot product rank `idx` from lowest to highest, e.g.
# demo_near_dwi(0) shows the error for the "least coherent" pair, while
# demo_near_dwi(-1) shows the error for the "most coherent" pair
# Load data
dwi_fname = "usmcuw_dipy.nii.gz"
dwi_nib = nib.load(dwi_fname)
dwi = dwi_nib.get_data().squeeze()
B_name = "B_dipy.txt"
B = np.loadtxt(B_name)
n = B.shape[0]
pp = []
for i in range(1, n):
for j in range(i + 1, n):
p = np.abs(B[i, :3].dot(B[j, :3]))
pp.append((p, (i, j)))
pp.sort()
sel = pp[idx][1]
t1 = dwi[..., sel[0]]
t1_lab = t1.astype(np.int32)
t1 = t1.astype(np.float64)
t1 = (t1.astype(np.float64) - t1.min()) / (t1.max() - t1.min())
t2 = dwi[..., sel[1]]
t2_lab = t2.astype(np.int32)
t2 = t2.astype(np.float64)
t2 = (t2.astype(np.float64) - t2.min()) / (t2.max() - t2.min())
# Prepare interactive graphs
global_figure = None
global_map = None
sel_x = None
sel_y = None
# Compute residuals of locally affine fit on T1 and T2
radius = 4
residuals_nb = ccr.compute_cc_residuals(t1, t2, radius, 1)
residuals_nb = np.array(residuals_nb)
# Apply the transfer
means_t2t1, vars_t2t1 = get_mean_transfer(t2_lab, t1)
ss_t1 = means_t2t1[t2_lab]
means_t1t2, vars_t1t2 = get_mean_transfer(t1_lab, t2)
ss_t2 = means_t1t2[t1_lab]
# Compute residuals of locally affine fit on T1 and F[T2]
radius = 4
residuals_t1 = ccr.compute_cc_residuals(t1, ss_t1, radius, 1)
residuals_t1 = np.array(residuals_t1)
# Compute residuals of locally affine fit on F(T1) and T2
radius = 4
residuals_t2 = ccr.compute_cc_residuals(ss_t2, t2, radius, 1)
residuals_t2 = np.array(residuals_t2)
slice_type = 2
slice_index = residuals_nb.shape[slice_type] // 2
if slice_type == 0:
max_val = np.max(
[
residuals_nb[slice_index, :, :].max(),
residuals_t1[slice_index, :, :].max(),
residuals_t2[slice_index, :, :].max(),
]
)
elif slice_type == 1:
max_val = np.max(
[
residuals_nb[:, slice_index, :].max(),
residuals_t1[:, slice_index, :].max(),
residuals_t2[:, slice_index, :].max(),
]
)
else:
max_val = np.max(
[
residuals_nb[:, :, slice_index].max(),
residuals_t1[:, :, slice_index].max(),
residuals_t2[:, :, slice_index].max(),
]
)
run_interactive_pair(t1, "T1", t2, "T2", residuals_nb, slice_type, slice_index, vmin=0, vmax=max_val)
def demo_t1_t2():
# Load data
t1_name = t1_name = get_brainweb("t1", "strip")
t1_nib = nib.load(t1_name)
t1 = t1_nib.get_data().squeeze()
t1_lab = t1.astype(np.int32)
t1 = t1.astype(np.float64)
# t1 = (t1.astype(np.float64) - t1.min())/(t1.max()-t1.min())
t2_name = t2_name = get_brainweb("t2", "strip")
t2_nib = nib.load(t2_name)
t2 = t2_nib.get_data().squeeze()
t2_lab = t2.astype(np.int32)
t2 = t2.astype(np.float64)
# t2 = (t2.astype(np.float64) - t2.min())/(t2.max()-t2.min())
# Prepare interactive graphs
global_figure = None
global_map = None
sel_x = None
sel_y = None
# Compute residuals of locally affine fit on T1 and T2
radius = 4
residuals_nb = ccr.compute_cc_residuals(t1, t2, radius, 1)
residuals_nb = np.array(residuals_nb)
# Apply the transfer
means_t2t1, vars_t2t1 = get_mean_transfer(t2_lab, t1)
ss_t1 = means_t2t1[t2_lab]
means_t1t2, vars_t1t2 = get_mean_transfer(t1_lab, t2)
ss_t2 = means_t1t2[t1_lab]
# Compute residuals of locally affine fit on T1 and F[T2]
radius = 4
residuals_t1 = ccr.compute_cc_residuals(t1, ss_t1, radius, 1)
residuals_t1 = np.array(residuals_t1)
# Compute residuals of locally affine fit on F(T1) and T2
radius = 4
residuals_t2 = ccr.compute_cc_residuals(ss_t2, t2, radius, 1)
residuals_t2 = np.array(residuals_t2)
slice_type = 1
slice_index = residuals_nb.shape[slice_type] // 2
if slice_type == 0:
max_val = np.max(
[
residuals_nb[slice_index, :, :].max(),
residuals_t1[slice_index, :, :].max(),
residuals_t2[slice_index, :, :].max(),
]
)
elif slice_type == 1:
max_val = np.max(
[
residuals_nb[:, slice_index, :].max(),
residuals_t1[:, slice_index, :].max(),
residuals_t2[:, slice_index, :].max(),
]
)
else:
max_val = np.max(
[
residuals_nb[:, :, slice_index].max(),
residuals_t1[:, :, slice_index].max(),
residuals_t2[:, :, slice_index].max(),
]
)
run_interactive_pair(t1, "T1", t2, "T2", residuals_nb, slice_type, slice_index, vmin=0, vmax=max_val)
run_interactive_pair(t1, "T1", ss_t1, "F[T2]", residuals_t1, slice_type, slice_index, vmin=0, vmax=max_val)
run_interactive_pair(t2, "T2", ss_t2, "F[T1]", residuals_t2, slice_type, slice_index, vmin=0, vmax=max_val)
