# Compute error
err[2, n, r] = np.mean(Y_h[M] != Y[M])
dcc[2, n, r] = dice(Y_h[M], Y[M])
if vis:
fn_segs = fn + 'SGM_segs_p' + str(n + 1) + '_r' + str(r +
1) + '.png'
plot_segmentation(Y_h, savefn=fn_segs)
''' Unsupervised hidden Potts'''
# Initialize model
UHP = VariationalHiddenPotts(num_components=K,
num_channels=D,
max_iter=max_iter,
tol=x_tol,
init_params='kmeans',
tissue_specific=True)
# Segment image
Y_h, post, theta = UHP.segment(X, beta=beta)
# Set cluster assignments to correct tissue labels
Y_h = set_classes(Y_h, z)
# Compute error
err[3, n, r] = np.mean(Y_h[M] != Y[M])
dcc[3, n, r] = dice(Y_h[M], Y[M])
# Plot images, plus error image
if vis:
scan[scan < 0] = 0
scan[scan > 255] = 255
scan = scan / 255.
# Load segmentation
segm = np.fromfile(pdir + '.raw', count=H * W, dtype='uint8')
# Reshape binary list to image
segm = nd.rotate(segm.reshape((H, W)), 90)
# Restrict segmentation
for k in np.setdiff1d(np.unique(segm), np.arange(K)):
segm[segm == k] = 0
hPotts = VariationalHiddenPotts(num_components=K,
num_channels=D,
tissue_specific=True)
Y1 = hPotts.one_hot(segm)
# Strip skull
scan[segm == 0] = 0
for d in range(3):
X[n, :, :, d] = scan
Y[n, :, :, :] = Y1
if os.path.exists('beta_brainweb01.npy'):
# Load previously estimated smoothing parameters
beta_h = np.load('beta_brainweb01.npy')