forked from qiwang321/MHN-parcellation
-
Notifications
You must be signed in to change notification settings - Fork 0
/
parcell_aal_spec_sa_patched.m
371 lines (320 loc) · 12.6 KB
/
parcell_aal_spec_sa_patched.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
function parcell_aal_spec_sa_patched(dirname, black_hole, dim_low, num_clusters, result_name) % the interface updated
%% spectral clustering with hopfield network post-processing
% function parcell_aal_spec_sa_patched(dirname, black_hole, dim_low, num_clusters, result_name)
%% load reference header structure
% nii_standard = load_untouch_nii('/usr/share/fsl/data/atlases/HarvardOxford/HarvardOxford-cort-maxprob-thr25-2mm.nii.gz');
nii_ref = load_untouch_nii([dirname, '/dti.bedpostX/nodif_brain.nii.gz']);
% nii_ref = load_untouch_nii([dirname, '/atlas_aal_90/aal_90_2mm_1.nii.gz']);
fprintf('processing %s...\n', dirname);
coord = load([dirname,'/track_aal_90_0/coords_for_fdt_matrix3']);
coord = single(coord(:,1:3));
% parameters
% num_clusters = 90;
num_voxels = size(coord,1);
% field_bound = [128,104,64];
field_bound = size(nii_ref.img);
%% neighbor_map
% construct the coordinate mapping first
fprintf('constructing coordinate mapping...\n');
tic;
neighbor_dist = 2; % distance of "neighbor nodes"
num_neighbors = 0;
for x = -neighbor_dist : neighbor_dist
for y = -floor(sqrt(neighbor_dist^2-x^2)) : floor(sqrt(neighbor_dist^2-x^2))
num_neighbors = num_neighbors + 2*floor(sqrt(neighbor_dist^2-x^2-y^2)) + 1;
end
end
num_neighbors = num_neighbors - 1;
grid_map_start = [min(coord(:,1)), min(coord(:,2)), min(coord(:,3))];
grid_map_end = [max(coord(:,1)), max(coord(:,2)), max(coord(:,3))];
grid_map = zeros(grid_map_end(1)-grid_map_start(1)+1+2*neighbor_dist, ...
grid_map_end(2)-grid_map_start(2)+1+2*neighbor_dist, ...
grid_map_end(3)-grid_map_start(3)+1+2*neighbor_dist);
coord_voxel_1d = zeros(num_voxels, 1);
for i = 1:num_voxels
coord_voxel = coord(i,1:3) - grid_map_start + 1 + neighbor_dist;
coord_voxel_1d(i) = 1 + coord(i,1) + field_bound(1) * (coord(i,2) + field_bound(2) * coord(i,3));
grid_map(coord_voxel(1), coord_voxel(2), coord_voxel(3)) = i;
end
dist = zeros(1, num_neighbors);
x_lim = neighbor_dist;
idn = 1;
for x = -x_lim:x_lim
y_lim = floor(sqrt(neighbor_dist^2-x^2));
for y = -y_lim:y_lim
z_lim = floor(sqrt(neighbor_dist^2-x^2-y^2));
for z = -z_lim:z_lim
if x==0 && y==0 && z==0, continue; end
dist(idn) = sqrt(x^2+y^2+z^2);
idn = idn+1;
end
end
end
neighbor_map = zeros(num_voxels, num_neighbors);
for i = 1:num_voxels
coord_voxel = coord(i,1:3) - grid_map_start + 1 + neighbor_dist;
idn = 1;
x_lim = neighbor_dist;
for x = -x_lim:x_lim
y_lim = floor(sqrt(neighbor_dist^2-x^2));
for y = -y_lim:y_lim
z_lim = floor(sqrt(neighbor_dist^2-x^2-y^2));
for z = -z_lim:z_lim
if x==0 && y==0 && z==0, continue; end
neighbor_map(i,idn) = grid_map(coord_voxel(1)+x, coord_voxel(2)+y, coord_voxel(3)+z);
idn = idn+1;
end
end
end
end
% add the virtual neighbors of the 0 node
neighbor_map = [zeros(1, num_neighbors); neighbor_map];
toc;
%% coarser map
fprintf('constructing the coarser map...\n');
% neighbor distance
neighbor_dist_low = 1; % distance of "neighbor nodes"
num_neighbors_low = 0;
for x = -neighbor_dist_low : neighbor_dist_low
for y = -floor(sqrt(neighbor_dist_low^2-x^2)) : floor(sqrt(neighbor_dist_low^2-x^2))
num_neighbors_low = num_neighbors_low + 2*floor(sqrt(neighbor_dist_low^2-x^2-y^2)) + 1;
end
end
num_neighbors_low = num_neighbors_low - 1;
% distance map
dist_low = zeros(1, num_neighbors_low);
x_lim = neighbor_dist_low;
idn = 1;
for x = -x_lim:x_lim
y_lim = floor(sqrt(neighbor_dist_low^2-x^2));
for y = -y_lim:y_lim
z_lim = floor(sqrt(neighbor_dist_low^2-x^2-y^2));
for z = -z_lim:z_lim
if x==0 && y==0 && z==0, continue; end
dist_low(idn) = sqrt(x^2+y^2+z^2);
idn = idn+1;
end
end
end
% high-low map: map the connectivity profile to low dimensional grid
% dim_low = 3;
grid_low = false( floor((grid_map_end(1)-grid_map_start(1))/dim_low ) + 1 + 2*neighbor_dist_low, ...
floor((grid_map_end(2)-grid_map_start(2))/dim_low ) + 1 + 2*neighbor_dist_low, ...
floor((grid_map_end(3)-grid_map_start(3))/dim_low ) + 1 + 2*neighbor_dist_low);
x = floor( (coord(:,1) - grid_map_start(1)) / dim_low ) + 1 + neighbor_dist_low;
y = floor( (coord(:,2) - grid_map_start(2)) / dim_low ) + 1 + neighbor_dist_low;
z = floor( (coord(:,3) - grid_map_start(3)) / dim_low ) + 1 + neighbor_dist_low;
for i = 1:num_voxels
grid_low(x(i),y(i),z(i)) = true;
end
ind = find(grid_low);
x_low = mod(ind-1, size(grid_low,1)) + 1;
y_low = mod((ind-x_low)/size(grid_low,1), size(grid_low,2)) + 1;
z_low = ((ind-x_low)/size(grid_low,1)-y_low+1)/size(grid_low,2) + 1;
num_voxels_low = length(x_low);
grid_map_low = zeros(size(grid_low));
grid_map_low(grid_low) = 1:num_voxels_low;
high_low_map = zeros(num_voxels,1);
for i = 1:num_voxels
high_low_map(i) = grid_map_low(x(i), y(i), z(i));
end
coord_map_low = [x_low, y_low, z_low];
neighbor_map_low = zeros(num_voxels_low, num_neighbors_low);
for i = 1:num_voxels_low
coord_voxel = coord_map_low(i,:);
idn = 1;
x_lim = neighbor_dist_low;
for x = -x_lim:x_lim
y_lim = floor(sqrt(neighbor_dist_low^2-x^2));
for y = -y_lim:y_lim
z_lim = floor(sqrt(neighbor_dist_low^2-x^2-y^2));
for z = -z_lim:z_lim
if x==0 && y==0 && z==0, continue; end
neighbor_map_low(i,idn) = grid_map_low(coord_voxel(1)+x, coord_voxel(2)+y, coord_voxel(3)+z);
idn = idn+1;
end
end
end
end
neighbor_map_low = [zeros(1, num_neighbors_low); neighbor_map_low];
%% prepare the connectivity profile
if ~exist([dirname, '/conn_profile_patched_dimlow',num2str(dim_low), '.mat'], 'file')
% load data
conn_profile = single(zeros(num_voxels, num_voxels_low));
for d = 0:0
fprintf('reading data %d...\n', d);
tic;
A = load([dirname,'/track_aal_90_',num2str(d),'/fdt_matrix3.dot']); A = single(A);
toc;
%% decide the boundary indices in the connectivity data
fprintf('updating the connectivity profiles...\n');
tic;
for i = 1:size(A,1)
term1 = high_low_map(A(i,1)); term2 = high_low_map(A(i,2));
conn_profile(A(i,1),term2) = conn_profile(A(i,1),term2) + A(i,3);
conn_profile(A(i,2),term1) = conn_profile(A(i,2),term1) + A(i,3);
end
toc;
clear A;
end
% add the virtual conn_profile of the 0 node
conn_profile = [zeros(1, num_voxels_low+1); zeros(num_voxels, 1), conn_profile];
save([dirname, '/conn_profile_raw_patched_dimlow', num2str(dim_low), '.mat'], 'conn_profile');
%% smoothing: smoothing need to use the new neighbor map which is of lower
% dimension
fprintf('gaussian smoothing...\n');
tic;
conn_profile_tmp = conn_profile;
for n = 1:num_neighbors_low
% append_conn = uint8( exp(-dist_low(n) ^2 / 4) * conn_profile(:,neighbor_map_low(:,n)+1) );
append_conn = single(exp(-dist_low(n) ^2)/(4/dim_low)^2) * conn_profile(:,neighbor_map_low(:,n)+1);
conn_profile_tmp = conn_profile_tmp + append_conn;
end
conn_profile = conn_profile_tmp;
clear conn_profile_tmp;
conn_norm = sqrt(sum(conn_profile.^2, 2)) + single(0.01);
toc;
save([dirname, '/conn_profile_patched_dimlow', num2str(dim_low), '.mat'], 'conn_profile', 'conn_norm');
else
load([dirname, '/conn_profile_patched_dimlow', num2str(dim_low), '.mat']);
end % endif exist conn_profile
thres_conn = 10;
invalid_idx = find(conn_norm(2:end) < thres_conn); % voxels with not too little total connectivity
valid_idx = find(conn_norm(2:end) >= thres_conn);
conn_profile(invalid_idx+1,:) = 0;
if ~exist([dirname, '/graph_patched_dimlow',num2str(dim_low), '.mat'], 'file')
%% create the similarity graph
fprintf('creating the similarity graph...\n');
tic;
G_sparse = zeros((num_voxels+1)*num_neighbors, 3);
G_sparse(:,1) = repmat( (1:num_voxels+1).', num_neighbors, 1 );
G_sparse(:,2) = reshape( neighbor_map+1, (num_voxels+1)*num_neighbors, 1 );
for n= 1:num_neighbors
% G_sparse( (n-1)*(num_voxels+1)+1 : n*(num_voxels+1), 3 ) = sum(conn_profile.*conn_profile(neighbor_map(:,n)+1, :), 2);
G_sparse( (n-1)*(num_voxels+1)+1 : n*(num_voxels+1), 3 ) = sum(conn_profile.*conn_profile(neighbor_map(:,n)+1, :), 2)...
./ conn_norm ./ conn_norm(neighbor_map(:,n)+1); % "cosine" distance
end
clear conn_profile;
balance = 0;
G_sparse(:,3) = G_sparse(:,3) - balance;
% form the sparse matrix; get rid of the virtual first row and first column
G = sparse(G_sparse(:,1), G_sparse(:,2), G_sparse(:,3), num_voxels+1, num_voxels+1);
G(1,:) = [];
G(:,1) = [];
% for i = 1:num_voxels
% if mod(i, 10000)==0, fprintf('%d\n', i); end
% j = nonzeros(neighbor_map(i,:));
% G(i,j) = sim_conn_one2m(conn_profile(i,:), conn_profile(j,:));
% G(j,i) = G(i,j);
% end
toc;
save([dirname, '/graph_patched_dimlow', num2str(dim_low), '.mat'], 'G');
else
load([dirname, '/graph_patched_dimlow', num2str(dim_low), '.mat']);
end % endif exist graph.mat
% G_support = G(G~=0);
% % G_support = G_support - mean(G_support);
% G_support = G_support - 0.5;
%
% G(G~=0) = G_support;
G = G(valid_idx, valid_idx);
%% grow initial parcellation with the aal atlas
fprintf('growing initial parcellation...\n');
tic;
thres = 0.2; % threshold for valid initial parcellation
mask = [];
for i = 1:num_clusters
nii_struct = load_untouch_nii([dirname, '/atlas_aal_90/aal_90_2mm_', num2str(i), '.nii.gz']);
nii_img = nii_struct.img(:);
mask = [mask, nii_img(coord_voxel_1d)];
end
labels = zeros(num_voxels,1);
for i = 1:num_voxels
[mv, mi] = max(mask(i,:), [], 2);
if mv > thres
labels(i) = mi;
end
end
labels = labels(valid_idx);
toc;
%% spectral clustering
fprintf('computing required matrices...\n');
G = (G+G')/2;
Dvec = sum(G,2);
D = diag(Dvec);
L = D - G;
fprintf('computing Dinv...\n');
Dinv12 = sqrt(diag(sparse(1./Dvec)));
fprintf('computing normalized laplacian...\n');
L = Dinv12 * L * Dinv12;
L = (L+L') /2;
fprintf('computing the eigen vectors...\n');
tic;
[U,S] = eigs(L, num_clusters, 'sa');
toc;
% initialize kmeans
fprintf('kmeans initializing...\n');
tic;
centroids = zeros(num_clusters, size(U,2)-1);
for k = 1:num_clusters
centroids(k,:) = mean(U((labels==k),2:end), 1);
end
toc;
fprintf('kmeans clustering...\n');
tic;
labels_new = kmeans(U(:,2:end), num_clusters, 'Distance', 'sqEuclidean', 'EmptyAction', 'singleton', 'Replicates', 1, 'start', centroids);
toc;
labels = zeros(num_voxels,1);
labels(valid_idx) = labels_new;
%% save the temporal results for spectral clustering
fprintf('generating nii file...\n');
view_field = zeros(field_bound(1), field_bound(2), field_bound(3));
for i = 1:num_voxels
view_field(coord(i,1)+1, coord(i,2)+1, coord(i,3)+1) = labels(i);
end
nii_struct = make_nii(view_field);
hdr = nii_ref.hdr;
hdr.dime.datatype = nii_struct.hdr.dime.datatype;
hdr.dime.bitpix = nii_struct.hdr.dime.bitpix;
hdr.dime.cal_max = nii_struct.hdr.dime.cal_max;
hdr.dime.cal_min = nii_struct.hdr.dime.cal_min;
hdr.dime.glmax = nii_struct.hdr.dime.glmax;
hdr.dime.glmin = nii_struct.hdr.dime.glmin;
nii_struct.hdr = hdr;
save_nii(nii_struct, [dirname,'/parcel_result_spec_sa_tmp.nii.gz']);
%% hopfield network dynamics
fprintf('doing hopfield network iterations: phase 1...\n');
labels_new = ind2vec(labels_new, num_clusters);
labels_new = labels_new*(1-black_hole) + black_hole;
% smoothing stage
fprintf('doing hopfield network iterations: phase 2...\n');
for iter = 1:100
fprintf('iter %d\n', iter);
labels_new = G * labels_new;
[mv, ind] = max(labels_new,[],2);
kept_ind = find(mv > 0);
ind = (ind(kept_ind)-1)*size(labels_new,1) + kept_ind;
labels_new = black_hole*ones(size(labels_new));
labels_new(ind) = 1; % constructed new labeling
end
% convert back to scalar label
labels_new = vec2ind(labels_new);
labels = zeros(num_voxels,1);
labels(valid_idx) = labels_new;
%% convert to index
% make nii file for view in fslview
fprintf('generating nii file...\n');
view_field = zeros(field_bound(1), field_bound(2), field_bound(3));
for i = 1:num_voxels
view_field(coord(i,1)+1, coord(i,2)+1, coord(i,3)+1) = labels(i);
end
nii_struct = make_nii(view_field);
hdr = nii_ref.hdr;
hdr.dime.datatype = nii_struct.hdr.dime.datatype;
hdr.dime.bitpix = nii_struct.hdr.dime.bitpix;
hdr.dime.cal_max = nii_struct.hdr.dime.cal_max;
hdr.dime.cal_min = nii_struct.hdr.dime.cal_min;
hdr.dime.glmax = nii_struct.hdr.dime.glmax;
hdr.dime.glmin = nii_struct.hdr.dime.glmin;
nii_struct.hdr = hdr;
save_nii(nii_struct, [dirname,'/',result_name]);