def create_mapping((sub, hemi)):
print sub, hemi
complex_file = '/scr/ilz3/myelinconnect/struct/surf_%s/orig/mid_surface/%s_%s_mid.vtk'
simple_file = '/scr/ilz3/myelinconnect/groupavg/indv_space/%s/lowres_%s_d_def.vtk'# version d
log_file = '/scr/ilz3/myelinconnect/all_data_on_simple_surf/labels/logs/log_worker_%s.txt'%(str(os.getpid()))
seed_file = '/scr/ilz3/myelinconnect/all_data_on_simple_surf/seeds/%s_%s_highres2lowres_seeds.npy'
label_file = '/scr/ilz3/myelinconnect/all_data_on_simple_surf/labels/%s_%s_highres2lowres_labels.npy'
surf_label_file = '/scr/ilz3/myelinconnect/all_data_on_simple_surf/labels/%s_%s_highres2lowres_labels.vtk'
# load the meshes
log(log_file, 'Processing %s %s'%(sub, hemi))
log(log_file, '...loading data', logtime=False)
complex_v, complex_f, complex_d = read_vtk(complex_file%(hemi, sub, hemi))
simple_v, simple_f, simple_d = read_vtk(simple_file%(sub, hemi))
# find those points on the individuals complex mesh that correspond best
# to the simplified group mesh in subject space
log(log_file, '...finding unique voronoi seeds')
try:
voronoi_seed_idx = np.load(seed_file%(sub, hemi))
voronoi_seed_coord = complex_v[voronoi_seed_idx]
except IOError:
voronoi_seed_idx, inaccuracy = find_voronoi_seeds(simple_v, complex_v)
np.save(seed_file%(sub, hemi), voronoi_seed_idx)
# find coordinates of those points in the highres mesh
voronoi_seed_coord = complex_v[voronoi_seed_idx]
# double check differences
log(log_file, '...checking unique vs nearest mapping')
dist = np.linalg.norm((voronoi_seed_coord - simple_v), axis=1)
if ((np.mean(dist)-np.mean(inaccuracy[:,0])>0.1)):
log(log_file, 'Unique seeds very far from nearest seeds!')
return dist
sys.exit("Unique seeds very far from nearest seeds %s %s"%(sub,hemi))
# convert highres mesh into graph containing edge length
log(log_file, '...creating graph')
complex_graph = graph_from_mesh(complex_v, complex_f, edge_length=True)
# find the actual labels
log(log_file, '...competetive fast marching')
labels = competetive_fast_marching(complex_v, complex_graph, voronoi_seed_idx)
# write out labelling file and surface with labels
log(log_file, '...saving data')
np.save(label_file%(sub, hemi), labels)
write_vtk(surf_label_file%(sub, hemi), complex_v, complex_f,
data=labels[:,1, np.newaxis])
log(log_file, 'Finished %s %s'%(sub, hemi))
return log_file
nonmask = np.delete(idcs, mask)
embed = np.zeros(((lv.shape[0] + rv.shape[0]), 100))
embed[nonmask] = embed_masked
# extend mask to nodes that have a t1avg < 1500
full_t1 = np.concatenate((lh_t1, rh_t1))
fullmask = np.unique(np.concatenate((mask, np.where(full_t1 <= 1500)[0])))
fullmask = np.asarray(fullmask, dtype='int64')
nonmask_bigmask = np.delete(idcs, fullmask)
# mask embedding and t1
masked_t1 = np.delete(full_t1, fullmask)
masked_embed = np.delete(embed, fullmask, axis=0)
# estimate sigma2_res from noise in T1
Gl = graph_from_mesh(lv, lf)
Gr = graph_from_mesh(rv, rf)
left_medians = []
for li in range(lv.shape[0]):
neigh = Gl.neighbors(li)
neigh_t1 = [full_t1[n] for n in neigh if not full_t1[n] == 0]
neigh_dist = [np.abs(full_t1[li] - nt) for nt in neigh_t1]
left_medians.append(np.median(neigh_dist))
right_medians = []
for ri in range(rv.shape[0]):
neigh = [x + lv.shape[0] for x in Gr.neighbors(ri)]
neigh_t1 = [full_t1[n] for n in neigh if not full_t1[n] == 0]
neigh_dist = [
np.abs(full_t1[ri + lv.shape[0]] - nt) for nt in neigh_t1
for sub in subjects:
for hemi in hemis:
if os.path.isfile('/scr/ilz3/myelinconnect/all_data_on_simple_surf/labels/%s_%s_highres2lowres_labels.npy'%(sub, hemi)):
print 'processing '+sub+' '+hemi
# load data
complex_v,complex_f, complex_d = read_vtk('/scr/ilz3/myelinconnect/struct/surf_%s/orig/mid_surface/%s_%s_mid.vtk'%(hemi, sub, hemi))
simple_v, simple_f, simple_d = read_vtk('/scr/ilz3/myelinconnect/groupavg/indv_space/%s/lowres_%s_d_def.vtk'%(sub, hemi))
labelling=np.load('/scr/ilz3/myelinconnect/all_data_on_simple_surf/labels/%s_%s_highres2lowres_labels.npy'%(sub, hemi))
seeds=np.load('/scr/ilz3/myelinconnect/all_data_on_simple_surf/seeds/%s_%s_highres2lowres_seeds.npy'%(sub, hemi))
# make an edge label surface and save it
G = graph_from_mesh(complex_v, complex_f)
edge_labelling = np.zeros_like(labelling[:,1])
for v in range(edge_labelling.shape[0]):
if any(labelling[G.neighbors(v),1] != labelling[v,1]):
edge_labelling[v] = 1
write_vtk(edge_file%(sub, hemi), complex_v, complex_f,
data=edge_labelling[:,np.newaxis])
# make a no label surface and save it
if np.where(labelling[:,1]==-1)[0].shape[0] >0:
no_labelling = np.zeros_like(labelling[:,1])
for n in list(np.where(labelling[:,1]==-1)[0]):
no_labelling[n] = 1
write_vtk(no_file%(sub, hemi), complex_v, complex_f,
data=no_labelling[:,np.newaxis])
rv, rf = rsurf[0], rsurf[1]
cortex = np.concatenate((lidx, ridx+lv.shape[0]))
masked_myelin = np.load(myelin_file)
masked_embed = np.load(embed_file)
# unmask embedding and myelin
full_embed = np.zeros(((lv.shape[0]+rv.shape[0]),300))
full_embed[cortex] = masked_embed
full_myelin = np.zeros(lv.shape[0]+rv.shape[0],)
full_myelin[cortex] = masked_myelin
# estimate sigma2_res from noise in T1
Gl = graph_from_mesh(lv, lf)
Gr = graph_from_mesh(rv, rf)
left_medians= []
for li in range(lv.shape[0]):
neigh = Gl.neighbors(li)
neigh_myelin = [full_myelin[n] for n in neigh if not full_myelin[n]==0]
neigh_dist = [np.abs(full_myelin[li] - nt) for nt in neigh_myelin]
left_medians.append(np.median(neigh_dist))
right_medians = []
for ri in range(rv.shape[0]):
neigh = [x + lv.shape[0] for x in Gr.neighbors(ri)]
neigh_myelin = [full_myelin[n] for n in neigh if not full_myelin[n]==0]
neigh_dist = [np.abs(full_myelin[ri+lv.shape[0]] - nt) for nt in neigh_myelin]
right_medians.append(np.median(neigh_dist))