-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathSample_cfeature_combination.m
430 lines (370 loc) · 27.6 KB
/
Sample_cfeature_combination.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
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
clc;
clear all;
% close all;
Dataset=3;
Subjects=[10];% instead of 1
bands=[1]; % 1=broad, 2=delta, 3=theta, 4=alpha, 5=beta, 6=gamma
Windows=[1:53];
Fs=1000;
%% These features provide 1 number for each trial and each channel
for band=bands % 1=broad, 2=delta, 3=theta, 4=alpha, 5=beta, 6=gamma
if band==2
lowband=0.5;
highband=4;
elseif band==3
lowband=4;
highband=8;
elseif band==4
lowband=8;
highband=12;
elseif band==5
lowband=12;
highband=16;
elseif band==6
lowband=25;
highband=200;
end
for Feat_selection=[1:9 12:19]
% for Feat_selection=[12:15]
for Subject=Subjects
signal = DatasetLoading(Dataset,Subject); % channel, time, cat, trial
Sel_feat=zeros(53,15,26);
accuracy = nan*ones(1,nchoosek(size(signal,3),2),length(Windows));
for windoww=Windows
windows=[1:1200];
step_size=20;
window_span=50;
wind=windows((windoww-1)*step_size+1:(windoww-1)*step_size+window_span);
features_chosen=nan*ones(size(signal,1),size(signal,3),size(signal,4),26);
feat=0;
for feature=[2:9 11:13 18:30 32 34] % features when evaluating time windows
feat=feat+1;
if feature==33
net= load ('imagenet-caffe-alex.mat');
end
for channel = 1:size(signal,1)
for category = 1:size(signal,3)
for trial = 1:size(signal,4)
if feature==1
trial_data=signal(channel,1:200,category,trial);
else
trial_data=signal(channel,wind,category,trial);
end
if band>1 && sum(isnan(trial_data))==0
trial_data=eegfilt(trial_data,Fs,lowband,highband,0,floor(length(trial_data)/3),0,'fir1');
end
if feature<28 && channel==1 && category==1 && trial==1 % single-valued
feature_chosen=nan.*ones(size(signal,1),size(signal,3),size(signal,4));
elseif feature==28 && channel==1 && category==1 && trial==1 % Wavelet
feature_chosen=nan.*ones(size(signal,1),size(signal,3),size(signal,4),length(wind)+20);
elseif (feature==29 || feature==30) && channel==1 && category==1 && trial==1 % Hilbert
feature_chosen=nan.*ones(size(signal,1),size(signal,3),size(signal,4),length(wind));
elseif feature==31 && channel==1 && category==1 && trial==1 % PAC
feature_chosen=nan.*ones(size(signal,1),size(signal,3),size(signal,4),size(signal,1)-1);
elseif feature==32 && channel==1 && category==1 && trial==1 % ICC
feature_chosen=nan.*ones(size(signal,1),size(signal,3),size(signal,4),size(signal,1));
elseif feature==33 && channel==1 && category==1 && trial==1 % CNN
feature_chosen=nan.*ones(size(signal,1),size(signal,3),size(signal,4),1000);
elseif feature==34 && channel==1 && category==1 && trial==1 % Samples
feature_chosen=nan.*ones(size(signal,1),size(signal,3),size(signal,4),length(wind));
elseif feature==35 && Dataset<3 && channel==1 && category==1 && trial==1 % Autocorr
feature_chosen=nan.*ones(size(signal,1),size(signal,3),size(signal,4),size(signal,1));
end
if sum(isnan(trial_data))>0
feature_chosen(channel,category,trial)=nan;
else
if feature==1
%% Baseline
feature_chosen(channel,category,trial) = mean(trial_data);
elseif feature==2
%% Time features: Sigal mean
feature_chosen(channel,category,trial) = mean(trial_data);
elseif feature==3
%% Signal median
feature_chosen(channel,category,trial) = median(trial_data);
elseif feature==4
%% Signal variance
feature_chosen(channel,category,trial) = var(trial_data);
elseif feature==5
%% Signal skewness
feature_chosen(channel,category,trial) = skewness(trial_data);
elseif feature==6
%% Signal Kurtosis
feature_chosen(channel,category,trial) = kurtosis(trial_data);
elseif feature==7
%% LZ complexity
threshold = median(trial_data);
% threshold = mean(data);
trial_data(trial_data>= threshold)=1;
trial_data(trial_data< threshold)=0;
[feature_chosen(channel,category,trial),~] = calc_lz_complexity(trial_data, 'exhaustive', 1);
elseif feature==8
%% Higuchi fractal dimension
maxtime = length(trial_data);
% Kmax = floor(maxtime./2);
Kmax = 10;
feature_chosen(channel,category,trial) = Higuchi_FD(trial_data,Kmax);
% % second implementation
% [HFD2(channel,category,trial),~,~,~] = HFD_LCALC(data);
% % thrid implementation
% HFD3(channel,category,trial) = hfd(data,Kmax);
elseif feature==9
%% Katz fractal dimensions
feature_chosen(channel,category,trial) = Katz_FD(trial_data);
elseif feature==10
%% Lyapunov exponent (largest LLE)
[feature_chosen(channel,category,trial),~] = lyaprosen(trial_data,0,0);
elseif feature==11
%% Hurst Exponent
feature_chosen(channel,category,trial) = estimate_hurst_exponent(trial_data);
% HE2(channel,category,trial) = genhurst(data);
% HE3(channel,category,trial) = hurstCC(data);
elseif feature==12
%% Sample entropy
feature_chosen(channel,category,trial) = entropy (trial_data);
% Ent2(channel,category,trial) = SampEn (2,0.2.* std(data),data,1);
elseif feature==13
%% Approximate Entropy
feature_chosen(channel,category,trial) = ApEn (2,0.2.* std(trial_data),trial_data,1);
% Ent4(channel,category,trial) = approx_entropy(2,0.2.* std(data),data);
elseif feature==14
%% ERP components P1/C1/P100, N1/N170, P200/P2a and P2b, chosen by visual inspection of PO3 channel, windows of components are 40, 60, 70 and 80 ms respectively.
% each trial can have 1 to 4 components
% P1
feature_chosen(channel,category,trial)=nanmean(trial_data([80:120]));
elseif feature==15
%% N1
feature_chosen(channel,category,trial)=nanmean(trial_data([120:200]));
elseif feature==16
%% P2a
feature_chosen(channel,category,trial)=nanmean(trial_data([150:220]));
elseif feature==17
%% P2b
feature_chosen(channel,category,trial)=nanmean(trial_data([200:275]));
elseif feature==18
%% Within-trial correlation
numLags=ceil(length(trial_data)./2);
[acf,lags,~] =autocorr(trial_data,numLags);
feature_chosen(channel,category,trial)= mean(acf(2:end));
elseif feature==19
%% Hjorth complexity
% this finds spread of the spectrum and represents the change in frequency
% Hcomplexity
step_size=1./Fs;
data_prime=(diff(trial_data)./step_size);
data_second=(diff(data_prime)./step_size);
feature_chosen(channel,category,trial)=(std(data_second).*std(trial_data))./(std(trial_data)).^2;
elseif feature==20
%% Hmobility
step_size=1./Fs;
data_prime=(diff(trial_data)./step_size);
data_second=(diff(data_prime)./step_size);
feature_chosen(channel,category,trial)=std(data_prime)./std(trial_data);
elseif feature==21
%% Mean Freq
feature_chosen(channel,category,trial) = meanfreq(trial_data,Fs);
elseif feature==22
%% Median Freq
feature_chosen(channel,category,trial) = medfreq(trial_data,Fs);
elseif feature==23
%% Average Freq
zeroscount=0;
for i=2:length(trial_data)
if (trial_data(i)>0 && trial_data(i-1)<0) || (trial_data(i)<0 && trial_data(i-1)>0)
zeroscount=zeroscount+1;
end
end
feature_chosen(channel,category,trial) =zeroscount.*(length(trial_data)./Fs);
elseif feature==24
%% Spectral edge frequency 95%
if var(trial_data)==0
feature_chosen(channel,category,trial) =0;
else
Fourier = fft(trial_data)/length(trial_data);
Fouriers = (abs(Fourier)); % Spectrum
Fv = linspace(0, 1, fix(length(trial_data)/2)+1)*Fs/2; % Frequency Vector
Iv = 1:length(Fv); % Index Vector
IntSpectrum = cumtrapz(Fv, Fouriers(Iv)); % Numeric Integration
feature_chosen(channel,category,trial) = interp1(IntSpectrum, Fv, 0.95*IntSpectrum(end), 'linear'); % Interploate To Find SEF
end
elseif feature==25
%% Power at Median Freq
amp = 2*abs(fft(trial_data))/length(trial_data);
phs = angle(fft(trial_data));
Fv = linspace(0, 1, fix(length(trial_data)/2)+1)*Fs/2; % Frequency Vector
Iv = 1:length(Fv); % Index Vector
fr_des = medfreq(trial_data,Fs); % Desired Frequency
ampv = amp(Iv); % Trim To Length Of Fv
phsv = phs(Iv); % Trim To Length Of Fv
ap = [ampv(:) phsv(:)]; % Amplitude & Phase Matrix
ap_des = interp1(Fv(:), ap, fr_des, 'linear');
feature_chosen(channel,category,trial) =ap_des(1);
elseif feature==26
%% Phase at Median Freq
amp = 2*abs(fft(trial_data))/length(trial_data);
phs = angle(fft(trial_data));
Fv = linspace(0, 1, fix(length(trial_data)/2)+1)*Fs/2; % Frequency Vector
Iv = 1:length(Fv); % Index Vector
fr_des = medfreq(trial_data,Fs); % Desired Frequency
ampv = amp(Iv); % Trim To Length Of Fv
phsv = phs(Iv); % Trim To Length Of Fv
ap = [ampv(:) phsv(:)]; % Amplitude & Phase Matrix
ap_des = interp1(Fv(:), ap, fr_des, 'linear');
feature_chosen(channel,category,trial) =ap_des(2);
elseif feature==27
%% Signal power
feature_chosen(channel,category,trial)=bandpower(trial_data);
elseif feature==28
%% Wavelet transform:
[c,l] = wavedec(trial_data,5,'sym2');
[ca5] = appcoef(c,l,'sym2',5);
[cd1,cd2,cd3,cd4,cd5] = detcoef(c,l,[1 2 3 4 5]);
feature_chosen(channel,category,trial,1:length([ca5,cd1,cd2,cd3,cd4,cd5]))=[ca5,cd1,cd2,cd3,cd4,cd5];
elseif feature==29
%% Hilbert transform amplitude
Hilb = hilbert(trial_data);
feature_chosen(channel,category,trial,:)=abs(Hilb);
elseif feature==30
%% Hilbert transform phase
Hilb = hilbert(trial_data);
feature_chosen(channel,category,trial,:)=angle(Hilb);
elseif feature==31
%% Phase-Amplitude Coupling
data.trial{1,1}= trial_data;
data.time{1,1}= [1:length(trial_data)]./Fs;
data.trialinfo=[100]';
data.label{1,1}='SampleData';
toi=[0.001 1.0]; % time of interest
phase=[0.5 12]; % phase(1):2.5:phase(2)
ampl=[24 120]; % amp(1):19:amp(2)
diag = 'no'; %'yes' or 'no' to turn on or off diagrams during computation
surrogates = 'no'; %'yes' or 'no' to turn on or off surrogates during computation
approach = 'tort';%,'ozkort','canolty','PLV';
[MI_matrix_raw,~] = calc_MI(data,toi,phase,ampl,diag,surrogates,approach);
feature_chosen(channel,category,trial,:)=reshape(MI_matrix_raw(1:6,1:5),[30 1]);
elseif feature==32
%% Inter-channel correlation 31 feature per trial
ICC=zeros(size(signal,1),1);
for ch2=1:size(signal,1)
trial_data2=signal(ch2,wind,category,trial);
if band>1
trial_data2=eegfilt(trial_data2,Fs,lowband,highband,0,floor(length(trial_data2)/3),0,'fir1');
end
ICC(ch2,1)=corr(trial_data',trial_data2');
end
feature_chosen(channel,category,trial,:)=ICC;
elseif feature==33
%% CNN
trial_data=(trial_data+abs(min(trial_data)));
trial_data=(trial_data).*(255./max(trial_data));
im_ = single(uint8(trial_data));
im_ = repmat(im_,[1 ceil(227*227./length(im_))]);
im_=im_(1,1:227*227);
im_=reshape(im_,[227 227]);
im_ = imresize(im_,net.meta.normalization.imageSize(1:2));
im_ = im_ - net.meta.normalization.averageImage;
res = vl_simplenn(net, im_);
feature_chosen(channel,category,trial,:)=squeeze(res.x);
elseif feature==34
%% Signal samples
feature_chosen(channel,category,trial,:)=trial_data;
elseif feature==35
%% Within-trial correlation
numLags=size(signal,1);
[acf,lags,~] =autocorr(trial_data,numLags);
feature_chosen(channel,category,trial,:)= acf(2:end);
end
end
end
end
end
if feature>27
features_chosen_tmp=[];
for Feat_sel=1:size(feature_chosen,4)
features_chosen_tmp=vertcat(features_chosen_tmp,reshape(feature_chosen(:,:,:,Feat_sel),[size(signal,1) size(signal,3)*size(signal,4)]));
end
nans=isnan(features_chosen_tmp(1,:));
label_tmp=repmat(1:size(signal,3),[1 size(signal,4)]);
label_tmp(nans)=[];
rmpath(genpath('/Hamid/Mojgan_analyses/Analyses/FSLib_v6.2.1_2018'))
features_reshaped_tmp=features_chosen_tmp(1:sum(~isnan(features_chosen_tmp(:,1))),logical(1-nans))';
[coeff,score,~,~,~,mu] =pca(features_reshaped_tmp);
features_reshaped_tmp = score*coeff(1:size(signal,1),:)' + repmat(mu(1:size(signal,1)),size(features_reshaped_tmp,1),1);
feature_chosen_tmp2=nan*ones(size(signal,3)*size(signal,4),size(signal,1));
g=0;
for i=1:size(signal,3)*size(signal,4)
if nans(i)==0
g=g+1;
feature_chosen_tmp2(i,:)=features_reshaped_tmp(g,:);
end
end
for chch=1:size(signal,1)
feature_chosenT(chch,:,:)=reshape(feature_chosen_tmp2(:,chch),[size(signal,3) size(signal,4)]);
end
feature_chosen=feature_chosenT;
addpath(genpath('/Hamid/Mojgan_analyses/Analyses/FSLib_v6.2.1_2018'))
end
features_chosen(:,:,:,feat)=zscore(feature_chosen);
% features_chosen(:,:,:,feat)=(feature_chosen);
end
%% Feature selection now all
features_chosen_tmp=[];
for Feat_sel=1:size(features_chosen,4)
features_chosen_tmp=vertcat(features_chosen_tmp,reshape(features_chosen(:,:,:,Feat_sel),[size(signal,1) size(signal,3)*size(signal,4)]));
end
nans=isnan(features_chosen_tmp(1,:));
features_chosen_tmp(:,nans)=[];
Y=[];
counts=[1:size(signal,3)];
counts=repmat(counts,[1 size(signal,4)]);
counts(nans)=[];
Y=counts;
X=features_chosen_tmp';
%% Classification
clearvars -except Feat_selection Sel_feat Windows Subjects bands lowband highband band windows windoww wind Dataset Fs True_Predicted_labels accuracy Subject signal feature X Y net
folds=10;
combinations=nchoosek(unique(Y),2);
for combination=1:size(combinations,1)
c=0;
for classes=combinations(combination,:)
for counter=1:length(Y)
if Y(counter)==classes
c=c+1;
YY(c)=Y(counter);
XX(c,:)=X(counter,:);
end
end
end
rmpath(genpath('/Hamid/Mojgan_analyses/Analyses/FSLib_v6.2.1_2018'))
t=zeros(size(XX,1),26);
for feats=1:26
[coeff,score,~,~,~,mu] =pca(XX(:,(feats-1)*size(signal,1)+1:feats*size(signal,1)));
t(:,feats) = score*coeff(1,:)' + repmat(mu(1),size(XX,1),1);
end
addpath(genpath('/Hamid/Mojgan_analyses/Analyses/FSLib_v6.2.1_2018'))
[fs,selection_method]=Feature_Selection_Ultimate(t,YY,Feat_selection,5);
rmpath(genpath('/Hamid/Mojgan_analyses/Analyses/FSLib_v6.2.1_2018'))
XXX=[];
for feats=fs
XXX =horzcat(XXX,XX(:,(feats-1)*size(signal,1)+1:feats*size(signal,1)));
end
YYY=YY;
[coeff,score,latent,tsquared,explained,mu] =pca(XXX);
XXX = score*coeff(1:size(signal,1),:)' + repmat(mu(1:size(signal,1)),size(XXX,1),1);
Xready=XXX;
Yready=YYY;
Sel_feat(windoww,combination,fs)=1;
clearvars XX YY YYY XXX
Classifier_Model = fitcdiscr(Xready,Yready,'DiscrimType','pseudoLinear');
cvmodel = crossval(Classifier_Model);
L = kfoldLoss(cvmodel);
accuracy(1,combination,windoww)=1-L;
[band Feat_selection Subject windoww combination]
end
Datasets={'Mine','Vhab','Stfd'};
Bands={'Broad','Delta','Theta','Alpha','Betta','Gamma'};
end
save(['Corrected_Dec_DS_',Datasets{Dataset},'_Band_',Bands{band},'_Wind_sliding_Subject_',num2str(Subject),'_CombFeat_',selection_method,'_PCA_cmplt.mat'],'accuracy','Sel_feat');
end
end
end