forked from FZJ-INM1-BDA/siibralab
-
Notifications
You must be signed in to change notification settings - Fork 0
/
walkthrough.m
352 lines (311 loc) · 16.1 KB
/
walkthrough.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
%% A quick walkthrough for siibralab - Matlab interface to the siibra toolsuite
% *siibra is a toolsuite for working with brain atlases that integrate parcellations
% and reference spaces at different spatial scales, covering multiple aspects
% of brain organization, and linking features from different data modalitites
% to brain structures. It aims to facilitate the programmatic and reproducible
% incorporation of brain region features from different sources into reproducible
% neuroscience workflows. siibraLab is a Matlab® toolbox for accessing functions
% provided in siibra.*
%
% _siibra_ provides structured acccess to parcellation schemes in different
% brain reference spaces, including volumetric reference templates at both macroscopic
% and microscopic resolutions as well as surface representations. It supports
% both discretely labelled and continuous (probabilistic) parcellation maps, which
% can be used to assign brain regions to spatial locations and image signals,
% to retrieve region-specific neuroscience datasets from multiple online repositories,
% and to sample information from high-resolution image data. Among the datasets
% anchored to brain regions are many different modalities from in-vivo and post
% mortem studies, including regional information about cell and transmitter receptor
% densties, structural and functional connectivity, gene expressions, and more.
%
% The main implementation of siibra is the Python client siibra-python, which
% as of now has a much more comprehensive set of features, document in the documentation
% at <https://siibra-python.readthedocs.io https://siibra-python.readthedocs.io>.
% There is also an interactive web application built around a 3D viewer - siibra-explorer
% -, which is hosted as part of the EBRAINS infrastructure at <https://atlases.ebrains.eu/viewer.
% https://atlases.ebrains.eu/viewer.>
%
% _siibralab_ is in very early development. Its API is not yet stable and the
% set of supported features is still rudimentary. We share releases of the toolbox
% on matlab file exchange (<https://www.mathworks.com/matlabcentral/fileexchange
% https://www.mathworks.com/matlabcentral/fileexchange>). If you are interested
% in the ongoing development and future releases, drop us a note to [email protected].
%
% This notebook provides a brief waltkhrough the already available functionality,
% focusing on access the Julich-Brain probabilistic cytoarchitectonic atlas.
%%
%
clear all; % uncomment rows to clear the workspace and close open figures
close all;
%% Accessing basic concepts: atlases, parcellations, and regions
% Predefined atlases
% siibralab provides access to atlases of different species. Note that on first
% call of a siibralab function, the package will retrieve preconfiguration data
% which might take a while. The data is then cached on your local disk for future
% calls.
siibra.atlasOverview()
%%
% As of now, siibra supports features of the human atlas, so let's access it.
atlas = siibra.getAtlas('human')
% Predefined parcellations
% The atlas groups a set of parcellations and reference spaces. We will work
% with the Julich-Brain cytoarchitectonic maps here, so we access the corresponding
% parcellation object. We can get a parcellation by providing a unique set of
% keywords.
[atlas.Parcellations.Name].'
julichbrain = atlas.getParcellation('Julich-Brain 2.9')
%%
% The parcellation is just a semantic object, defining a set of known regions
% that were delineated according to specific aspects of brain organization.
julichbrain.Description
% Brain regions from a parcellation
% We can search for regions, and also decode a partial name into a concrete
% region object which contains further information on the region.
regions = julichbrain.findRegion('V1')
v1l = julichbrain.decodeRegion('Area hoc1 V1 17 left')
%% Accessing the reference template for a given space
% Choosing a reference space
% The parcellation is just a semantic object. To obtain a parcellation map,
% a reference space needs to be specified. Just like a parcellation, the reference
% space can be accessed by keywords.
[julichbrain.Spaces.Name].'
mni_space = atlas.getSpace('MNI152 2009c nonl asym')
% Fetching the template as an image volume
% The space is a semantic object, while the template actually contains a reference
% image (vol_template). We fetch the template as a NIfTI object, which includes
% the actual image volume as a numeric array, as well as typical metadata such
% as an affine matrix relating the voxel space to the physical space, and various
% other metadata fields. Refer to the NIfTI documentation on details (https://nipy.org/nibabel/manual.html#manual).
%
% Here we inspect some metadata from its header.
mni_template_nii = mni_space.loadTemplate
mni_template_nii.Header.ImageSize
mni_template_nii.Header.Transform.T
mni_template_nii.Header.DisplayIntensityRange
%%
% The dimensions of the volume of the MNI152 template (mni_icbm152_nlin_asym_09c)
% are 193x229x193 px. The origin is located at x=-96, y=-132, z=-78]. Gray values
% range from 0 to 100 and represent the measured and callibrated T1 intensities
% of 152 individual subjects and scans.
%
% For the following, we explicitly extract the image volume from the NIfTI object:
mni_template_vol = mni_space.loadTemplate.loadData();
%% Fetching the probabilistic map of a given brain region in a given reference space
% Brain regions can be searched by name from the parcellation object. The parcellation's
% "findRegion" method would provide us a list of possible matches.
julichbrain.findRegion('Fp1')
%%
% Here we intend to specifiy a particular region directly, so we use "decodeRegion"
% which would raise an error if the specification is not unique. In the continuous
% maps of the Julich-Brain atlas, the intensities reflect the probabilty of being
% this specific area, so we denote them as probability maps (probmaps).
fp1l_region = julichbrain.decodeRegion('Fp1 left');
fp1l_probmap = fp1l_region.continuousMap(mni_space.Name);
%%
% We can now fetch the probability map as a NIfTI object, and retrieve the actual
% image volume, just as we did for the MNI template above.
fp1l_probmap_nii = fp1l_probmap.fetch();
fp1l_probmap_vol = fp1l_probmap_nii.loadData();
% Understanding the *probabilistic maps*
% The probabilistic maps have the same size and origin and can therefore interact
% with the template volume. The values of the volume of the probabilistic map
% range from 0 to 1 and reflect the probability that it is the specific area.
% 0 means that the specific area was found in 0 out of 10 brains at the specific
% position, 1 means that the area was found in 10 out of 10 brains at this specific
% position.
fprintf('max p value of area %s : %f ', fp1l_probmap.Name,max(fp1l_probmap_vol,[],'all'))
fprintf('min p value of area %s : %f ', fp1l_probmap.Name,min(fp1l_probmap_vol,[],'all'))
%% Visualizing parcellation maps
% Orthoslice view of the probability map superimposed with the template
z_plane = 90;
%%
% First we extract the 2D slice from the template volume.
hf = figure;
h1 = axes;
h_template_slice = slice(double(mni_template_vol),[],[],z_plane);
h_template_slice.EdgeColor='none';
colormap(h1,'gray');
%%
% Next we overlay it with a mask of the corresponding slice from the map.
h2 = axes;
h_map_slice = slice(fp1l_probmap_vol,[],[],90);
h_map_slice.EdgeColor = 'none';
mask = fp1l_probmap_vol(:,:,z_plane);
mask(mask>0) = 1;
h_map_slice.AlphaData = mask;
h_map_slice.FaceAlpha = 'flat';
set(h2,'color','none','visible','off')
colormap(h2,'parula');
h1.XColor='none'; h1.YColor='none'; h1.ZColor='none';
% Plot distribution of voxel wise probabilities in the continuous map
% There is always a larger periphery with lower probabilities because this is
% precisely the area due to the interindividual variability of brains. This is
% why the probability information is so valuable, because every brain is different
% in size and shape, which should be taken into account in neuroscientific analyses.
figure;
h_histo = histogram(fp1l_probmap_vol(fp1l_probmap_vol>0),100);
title('Probability distribution within the map of Fp1')
ylabel('number of voxel');
xlabel('probability of beeing specific area');
% Comparing probabilistic (continuous) map and maximum probabilistic (labelled) map
% The volume of the probability map is always larger than the volume of the
% maximum probability map. In the maximum probability map (MPM), also called "labeld
% map", each voxel is assigned exactly the one area label whose probability of
% being found at the specific voxel is the highest. It is therefore a "winner
% takes it all" model, which shrinks the volume compared to the probabilistic
% map, since some voxels within a probability map may be covered by neighboring
% areas with a higher probability.
%
% The MPM thus allows the representation of sharp boundaries, but it also does
% not represent the full degree of information of the probability map.
fprintf('The the size of the continous map (pmap) of Area Fp1 is %d voxel, or mm³, \nsince the the mni space has got a voxelsize of 1mm³.',size(find(fp1l_probmap_vol>0),1))
fp1l_mask = fp1l_region.getMask(mni_space.Name)
fp1l_mask_nii = fp1l_mask.fetch()
fp1l_mask_vol = fp1l_mask_nii.loadData;
fp1l_mask_vol_value = nnz(fp1l_mask_vol) % non zero elements in the mask
sum_pmap_voxel = 0;
for i=100:-1:1
sum_pmap_voxel = sum_pmap_voxel+h_histo.Values(i);
if sum_pmap_voxel > fp1l_mask_vol_value
disp(['The mpm representation of Area Fp1 with ' num2str(fp1l_mask_vol_value) ' voxel roughly corresponds in size with a ' num2str(i) '% thresholded ' newline 'version of the probabilistic map of area Fp1 with ' num2str(sum_pmap_voxel)]);
break;
end
end
% Plotting MPM (labelled map) representation of area Fp1 as isosurface with orthoslice
% We plot the same slice of the template as above, but with an isosurface representation
% of the mask.
close(gcf)
h_map_slice = slice(double(fp1l_probmap_vol),[],[],z_plane);
h_map_slice.EdgeAlpha = 0;
colormap(gca,"gray");
hold on
%%
% The isosurface is created from the mask by specifying a treshold value, here
% we choose 0.5 as the 50% probability.
s = isosurface(fp1l_mask_vol,0.5);
p = patch(s);
set(p,'FaceColor',[1 0 0]); %red color
set(p,'EdgeColor','none');
hold off
xlim(gca,[0 size(mni_template_vol,2)]);
ylim(gca,[0 size(mni_template_vol,1)]); %changed order!!
zlim(gca,[0 size(mni_template_vol,3)]);
view([28.6 25.7])
camlight;
lighting gouraud;
% Create a surface view of all regions
julichbrain_maxprobmap_nii = julichbrain.parcellationMap(mni_space.Name).fetch;
julichbrain_maxprobmap_right_vol = julichbrain_maxprobmap_nii(1).loadData;
mpm_idx = unique(julichbrain_maxprobmap_right_vol);
cmap = colormap('lines');
hold on
for i = 2:size(mpm_idx,1) % start with index 2 since 0 is background
single_mpm = julichbrain_maxprobmap_right_vol;
single_mpm(single_mpm ~= mpm_idx(i)) = 0;
s = isosurface(single_mpm,mpm_idx(i)-1);
p = patch(s);
set(p,'FaceColor',cmap(i,:));
set(p,'EdgeColor','none');
end
so = slice(double(mni_template_vol),[],[],z_plane);
so.EdgeAlpha = 0;
colormap(gca,"gray");
xlim(gca,[0 size(mni_template_vol,2)]);
ylim(gca,[0 size(mni_template_vol,1)]); %changed order!!
zlim(gca,[0 size(mni_template_vol,3)]);
hold off
view([51 18])
%% Assigning brain regions to volumes of interest given in a user-specified 3D image volume
% Let's assume you bring your own 3D volume depicting a region of interest,
% e.g. a functional activation, a segmented brain lesion, or 3D volumes of other
% origin. For the purpose of this example, we will here just use one of the probability
% maps (the one of Fp1 left) as a "fake activation" - just think of it as an fMRI
% activation. To load your own NIfTI file just use Matlab's <https://de.mathworks.com/help/images/ref/niftiread.html
% readnifti()> function (my_volume=|niftiread('brain.nii')|;). _Note that the
% volume of interest must be in the same reference space - here the MNI 152 space._
%
% *Question: Which map of the Jülich-Brain Atlas shows the closest spatial proximity
% to my volume?*
% define my volume
my_volume = fp1l_probmap_vol;
% loop over all Jülich-Brain areas included in the MPM (labeld_map)
mpm_idx = unique(julichbrain_maxprobmap_right_vol); % get all unique gray values of the Julich-Brain MPM
correlation_coeff = zeros(size(mpm_idx,1)-1,2); % predefine result variable
for i=2:size(mpm_idx,1) % start with index 2 since 0 is background
single_mpm = julichbrain_maxprobmap_right_vol;
single_mpm(single_mpm ~= mpm_idx(i)) = 0; % set all voxel which do not belong to the here adressed map to 0
r = pearson_3d_coefficient(uint16(my_volume),single_mpm); % function is located at the end of this script
correlation_coeff(i,:) = [mpm_idx(i) r]; % store the Julich-Brain gray value and the corresponding correlation coefficiant
end
[M,I] = max(correlation_coeff(:,2)); % find the area with the highest correlation
disp( ['The highest correlation coefficient is ' num2str(M) ', it was found with the area with gray value ' num2str(mpm_idx(I))])
%%
% From one of the first examples, we know that _gray level 212 in the Jülich
% Brain Atlas codes for area Fp1_.
%
% *Answer: Thus, it is obvious that the probability map of area Fp1 has the
% highest spatial correlation with the MPM representation of the Fp1 map.*
%% Extract regional properties from the probability map using the image_processing_toolbox
%if license('test', 'image_toolbox')
% regionprops3(fp1l_mask_vol,fp1l_probmap_vol,'MinIntensity','MeanIntensity',"MaxIntensity","WeightedCentroid")
% regionprops3(fp1l_mask_vol,'Volume','Centroid','SurfaceArea')
%end
%% Accessing multimodal data featrues
% Regional densities of neurotransmitter neuroreceptors
% Receptor densities, if available for a region, can then be retrieved using
% the getReceptorDensities() function. We choose region V1 here. Receptor densities
% are tabular data; they include a "fingerprint" of average densities measured
% in multiple tissue samples for differenty transmitter receptors.
v1l_region = julichbrain.decodeRegion("Area hOc1 (V1, 17, CalcS) left");
% This is currently not supported.
% To support fetching features, the matlab client needs to support the v3
% api first.
%receptorDensities = v1l_region.getReceptorDensities()
%%
% The fingerprints are typically visualized as a polarplot.
%fingerprint_table=receptorDensities.Fingerprint;
%fingerprint_table.plusSTD=receptorDensities.Fingerprint.("Mean")+receptorDensities.Fingerprint.("Std");
%fingerprint_table.minusSTD=receptorDensities.Fingerprint.("Mean")-receptorDensities.Fingerprint.("Std");
%
%figure;
%p=polarplot(fingerprint_table,["Mean" "plusSTD" "minusSTD"]);
%p(2).LineStyle='--';p(3).LineStyle='--';
%p(2).Color='c';p(3).Color='c';
%
%title(receptorDensities.Name)
%
%thetaticks(round(linspace(0,360,size(receptorDensities.Fingerprint,1))));
%thetaticklabels(receptorDensities.Fingerprint.Row);
% Structural connectivity grouped by parcellation
% As a feature of a parcellation, siibra provides access to streamline counts
% as a measure of structural connectivity. The connectivity has the form of a
% matrix, where columns and rows correspond to the regions defined in the parcellation.
% This is currently not supported.
% To support fetching features, the matlab client needs to support the v3
% api first.
%sc = julichbrain.getStreamlineCounts();
%%
% As with all datasets, descriptive meta information is part of the output of
% siibralab
%sc(1)
%%
% The actual streamline counts are available for 200 subjects from the HCP (range
% 1-200) and averaged from the 1000brains study (index 201). Here, for example,
% for subject number 1.
%head(sc(1).Matrix)
%%
% Plot the first 20 extracted streamline counts for "Area hOc1 (V1, 17, CalcS)
% right" as bar plot
%bar(sc(1).Matrix.("Area hOc1 (V1, 17, CalcS) right"))
%set(gca,'xticklabel',sc(1).Matrix.Properties.VariableNames)
%set(gca,'xtick',[1:size(sc(1).Matrix,1)])
%xlim([1 20])
%% Function definitions
function r = pearson_3d_coefficient(volume1,volume2)
A = volume1 - mean(volume1, 'all');
B = volume2 - mean(volume2, 'all');
num = sum(A .* B, 'all');
den1 = sum(A.^2, 'all');
den2 = sum(B.^2, 'all');
r = num ./ (den1 .* den2)^0.5;
end