本程序意在解释这样几个问题：完整版代码在本文的最后。
1
2
3
4
5

1
.实验的设计如何转换成设计矩阵？
2
.设计矩阵的每列表示一个刺激条件，如何确定它们？
3
.如何根据设计矩阵和每个体素的信号求得该体素对刺激的敏感性？

程序详解：
1.构造hrf
1
2
3
4
5
6
7
8

hrf_small = [
0
4
2

1
0
];
figure(
1
);
clf;
plot(
0
:
4
,hrf_small,
'o'
);
grid on;
xlabel(
'Time (in units of TRs, 4s long each)'
);
ylabel(
'fMRI signal'
);
title(
'This is what an HRF would look like if you measure once every 4s'
)

2.构造刺激序列，并与hrf做卷积：
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

word_stim_time_series = [
0
1
0
0
0
0
];
object_stim_time_series= [
0
0
0
1
0
0
];
predicted_signal_that_word_would_evoke = conv(word_stim_time_series,hrf_small);
predicted_signal_that_object_would_evoke = conv(object_stim_time_series,hrf_small);
figure(
2
);
clf;
subplot(
3
,
1
,
1
);
hold on;
h1=stem(word_stim_time_series,
'b'
);
h2=stem(object_stim_time_series,
'r'
);
hold off;
grid on;
legend([h1(
1
) h2(
1
)],
'Word stim onset time'
,
'Object stim onset time'
);
axis([
1
9
0
1.2
]);
ylabel(
'Stimulus present / absent'
);
subplot(
3
,
1
,
2
);
plot(predicted_signal_that_word_would_evoke,
'b*'
);
grid on;
legend(
'Wordsensitive voxel would give this fMRI signal'
);
axis([
1
10

1.5
7
]);
ylabel(
'fMRI signal'
);
subplot(
3
,
1
,
3
);
plot(predicted_signal_that_object_would_evoke,
'r^'
);
grid on;
legend(
'Objectsensitive voxel would give this fMRI signal'
);
axis([
1
10

1.5
7
]);
xlabel(
'Time (measured in TRs, i.e. one timepoint every 4secs)'
);
ylabel(
'fMRI signal'
);

3.利用两个刺激构造设计矩阵，并绘图
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

predicted_word_response_column_vec = predicted_signal_that_word_would_evoke';
predicted_object_response_column_vec = predicted_signal_that_object_would_evoke';
%%% Now let's look at the actual vectors in the Matlab workspace window
predicted_word_response_column_vec % Because there is no semicolon after
this
,
% it will display in workspace window
predicted_object_response_column_vec
%%%%%% Now we can join these two column vectors together
%%%%%% to make the design matrix. We simply put the two columns sidebyside.
%%%%%% In Matlab, you make
new
matrices and vectors by
%%%%%% putting the contents inside [ square brackets ]
%%%%%% Note that to join them together in
this
way, they must be
%%%%%% the same length as each other.
%%%%%%
%%%%%% Because the names of my variables are so
long
and verbose,
%%%%%% the command below spills over onto two lines. In Matlab,
%%%%%% we can split a command over two lines by putting three dots ...
design_matrix = ... % The three dots here mean
"continued on the next line"
[ predicted_word_response_column_vec predicted_object_response_column_vec ];
design_matrix % No semicolon, so it displays in window
%%%%%% Translation guide:
%%%%%% In equations, the design matrix is almost always called X
%%%%%% Note that
this
is a capital
"X"
.
%%%%%%
%%%%%% X = design_matrix;
%%%%%%
%%%%%% Capitals are typically used
for
matrices, and small
case
is
%%%%%% used
for
vectors.
%%%%%% The only difference between a vector and a matrix is that
%%%%%% a vector is just a bunch of numbers in a row (a rowvector)
%%%%%% or a bunch of numbers in a column (a columnvector),
%%%%%% whereas a matrix is bunch of vectors stacked up next to each
%%%%%% other to make a rectangular grid, with rows *and* columns of numbers.
%%%%%% Now let's view a grayscale plot of the design matrix,
%%%%%% in the way that an fMRIanalysis
package
, such as SPM, would show it.
%%%%%% To
do
this
, we use the Matlab command
"imagesc"
.
%%%%%% This takes each number in the design matrix and represents
%%%%%% it as a colour, with the colour depending on how big the number is.
%%%%%% In
this
case
, we'll be using a gray colourscale, so low numbers
%%%%%% will be shown as darker grays, and high numbers are lighter grays.
%%%%%% The
"sc"
part at the end of
"imagesc"
stands
for
"scale"
, which
%%%%%% means that Matlab scales the mapping of numbers onto colours so
%%%%%% that the lowest number gets shown as black, and the highest as white.
%%%%%%
%%%%%% For examples of how to use the imagesc command to make
%%%%%% pictures of brainslices, see the companion program
%%%%%% showing_brain_images_tutorial.m
figure(
3
);
clf; % Clear the figure
imagesc(design_matrix); %
'imagesc'
maps the numbers to colors,
% normalising so that the max goes to white
% and the min goes to black
colormap gray; % Show everything in grayscale
colorbar; % Shows how the numbers lie on the colour scale
% Note that the highest number in the design matrix,
% which is
4
, is shown as white, and the lowest, 
1
,
% gets shown as black.
title(
'Grayscale view of design matrix'
);
xlabel(
'Each column represents one stimulus condition'
);
ylabel(
'Each row represents one point in time, one row per TR (every 4secs)'
);

4.设计矩阵和敏感度矩阵相乘，这里假设某一个体素仅仅对某一种刺激有反应，而对另外的刺激没有反应。
%% Now suppose we have a voxel which responds only to words, not to objects.
%% We can calculate how it would be predicted to respond
%% to our word+object display as follows:
%%
%% Predicted response from wordsensitive voxel =
%% 1 * Response which wordpresentation would evoke
%% + 0 * Response which objectpresentation would evoke
%%
%% Note that this is how the voxel would be predicted to respond
%% if there were no noise whatsoever in the system.
%% Clearly a real fMRI signal would never be this clean.
%%
%% Now, let's make a "sensitivity vector" for this voxel,
%% in which each entry will say how sensitive that voxel is to
%% the corresponding stimulus condition.
%%
%% This voxel is sensitive to words, which are our *first* stimulustype.
%% And we made the predicted word response into the first column of
%% the design matrix.
%% So, the sensitivity of this voxel to words will be the first element
%% in the sensitivityvector.
%%
%% Similarly, the sensitivity of this voxel to the second stimulustype,
%% which are objects, will be the second element in the sensitivity vector.
%%
%% So, the sensitivity vector for a voxel with
%% sensitivity = 1 to the first stimulustype, which are words
%% and sensitivity = 0 to the second stimulustype, which are objects
%%
%% will be [ 1 0 ]
%%
%% I know this seems trivial !!
%% Things will get more interesting in a minute...
sensitivity_vec = [ 1 0 ]'; % The dash makes this a column vector
sensitivity_vec % No semicolon, so it displays in window
%% Translation guide:
%% In equations, the numbers in the sensitivityvector are typically
%% called "betavalues", or sometimes "betacoefficients" or "betaweights".
%% The columns of the design matrix are called "regressors" and
%% the value that is assigned to each regressor is the betavalue.
%%
%% Note that in the example above, we are pretending that we already *know*
%% how sensitive our voxel is to the various stimuli, but in the real world
%% we don't know this. We're trying to figure out what stimuli our voxel
%% is sensitive to, using the fMRI data that we collect in the scanner.
%% This will be described more below.
%% In mathspeak, that means that we are trying to *estimate* the betas.
%% When people want to distinguish between the true betavalue
%% (which we don't know) and the estimated betavalue that we figure out
%% from our data, then they call the true one beta and
%% the estimated one "beta hat" (beta with a circumflex sign on top of it: ^
%% [ End of that part of the translation guide, back to the main theme... ]
%% So, we can now express our predicted voxel response in terms
%% of entries in the sensitivity vector multiplied by
%% columns in the design matrix:
%%
%% Predicted response from wordsensitive voxel =
%% 1 * Response which wordpresentation would evoke
%% + 0 * Response which objectpresentation would evoke
%%
%% And because of the way we made our sensitivity vector and design matrix,
%% this can be rewritten as:
%%
%% Predicted response from wordsensitive voxel =
%% (First element in sensitivity vector) * (First column in design matrix)
%% + (Second element in sensitivity vector) * (Second column in design matrix)
%%
%% Here's an important bit:
%% The process above, of going through the elements in a vector,
%% multiplying each element by the corresponding column in a matrix,
%% and then adding up the results of the multiplication,
%% is precisely what matrix multiplication does.
%%
%% In Matlab, everything is by default assumed to be a matrix,
%% (or a vector  you can think of a vector as simply a matrix that only
%% has one row or column in it), and every multiplication is
%% by default assumed to be a matrix multiplication.
%% So, to matrixmultiply our design matrix by our sensitivityvector,
%% we just use the standard "multiply by" sign, which is *
predicted_word_selective_voxel_response = design_matrix * sensitivity_vec;
predicted_word_selective_voxel_response
% Let's display this vector in the command window,
% by entering it without a semicolon after it.
%% When we multiply the design matrix by the sensitivity vector,
%% we make the ith row of the result by taking the ith row
%% of the matrix, rotating it 90 degrees, multiplying it elementbyelement
%% with the sensitivity vector, and then adding that all up.
%%
%% Since the sensitivity vector is in this case [ 1 0 ],
%% multiplying each matrix row by it elementbyelement means that
%% we end up getting 1* the first element in each row, and 0* the second
%% element in each row.
%%
%% So, by the time we have gone through all the rows, we have
%% 1* the first column of the design matrix, plus 0* the second column,
%% which is what we wanted.
%% Let's plot all this
figure(4);
clf; % Clear the figure
subplot(2,1,1); % This is just to make the plots line up prettily
hold on; % "Hold" is one way of putting more than one plot on a figure
h1=plot(predicted_word_response_column_vec,'b*');
h2=plot(predicted_object_response_column_vec,'r^');
hold off;
grid on;
legend([h1 h2],'Wordresponse column vector','Objectresponse column vector');
axis([1 10 1.5 7]);
xlabel('Time (measured in TRs, i.e. one timepoint every 4secs)');
ylabel('fMRI signal');
subplot(2,1,2);
plot(predicted_word_selective_voxel_response,'ms'); % Magenta squares
grid on;
legend('Wordselective voxelresponse: 1*wordresponse + 0*objectresponse');
axis([1 10 1.5 7]);
xlabel('Time (measured in TRs, i.e. one timepoint every 4secs)');
ylabel('fMRI signal');
5.假设某一个体素对两种刺激都会产生反应，则它的beta矩阵应当是[1,1]:
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

%%% Now let's
try
a voxel which responds equally to both words and objects
%%% So, it's sensitivity vector will be [
1
1
]
%%%
%%% This means that its response will be
%%%
1
* the first column of the design matrix, plus
1
* the second column
%%% i.e.
%%%
1
* the response which the word stimulus evokes +
%%%
1
* the response which the object stimulus evokes
sensitivity_vec = [
1
1
]'; % The dash makes
this
a column vector
predicted_unselective_voxel_response = design_matrix * sensitivity_vec;
predicted_unselective_voxel_response % Display in Matlab command window
%% Let's plot all
this
figure(
5
);
clf; % Clear the figure
subplot(
2
,
1
,
1
); % This is just to make the plots line up prettily
hold on; %
"Hold"
is one way of putting more than one plot on a figure
h1=plot(predicted_word_response_column_vec,
'b*'
);
h2=plot(predicted_object_response_column_vec,
'r^'
);
hold off;
grid on;
legend([h1 h2],
'Wordresponse column vector'
,
'Objectresponse column vector'
);
axis([
1
10

1.5
7
]);
xlabel(
'Time (measured in TRs, i.e. one timepoint every 4secs)'
);
ylabel(
'fMRI signal'
);
subplot(
2
,
1
,
2
);
plot(predicted_unselective_voxel_response,
'ms'
); % Magenta squares
grid on;
legend(
'Unselective voxelresponse: 1*wordresponse + 1*objectresponse'
);
axis([
1
10

1.5
7
]);
xlabel(
'Time (measured in TRs, i.e. one timepoint every 4secs)'
);
ylabel(
'fMRI signal'
);

6.假设某一个体素对两种刺激都会产生反应，但是它的beta矩阵是[1,2]，即对第二种刺激反应更强烈:
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

%%% Ok, I hope
this
isn
't overkill: let'
s
try
a voxel which gives a normal
%%% response to words, but which gives a response to objects which is
%%% *twice* as strong.
%%% So, it's sensitivity vector will be [
1
2
]
%%%
%%% This means that its response will be
%%%
1
* the first column of the design matrix, plus
2
* the second column
%%% i.e.
%%%
1
* the response which the word stimulus evokes +
%%%
2
* the response which the object stimulus evokes
sensitivity_vec = [
1
2
]'; % The dash makes
this
a column vector
predicted_object_preferring_voxel_response = design_matrix * sensitivity_vec;
predicted_object_preferring_voxel_response % Display in Matlab command window
%% Let's plot all
this
figure(
6
);
clf; % Clear the figure
subplot(
2
,
1
,
1
); % This is just to make the plots line up prettily
hold on;
h1=plot(predicted_word_response_column_vec,
'b*'
);
h2=plot(predicted_object_response_column_vec,
'r^'
);
hold off;
grid on;
legend([h1 h2],
'Wordresponse column vector'
,
'Objectresponse column vector'
);
axis([
1
10

2
10
]);
xlabel(
'Time (measured in TRs, i.e. one timepoint every 4secs)'
);
ylabel(
'fMRI signal'
);
subplot(
2
,
1
,
2
);
plot(predicted_object_preferring_voxel_response,
'ms'
); % Magenta squares
grid on;
legend(
'Objectpreferring voxelresponse: 1*wordresponse + 2*objectresponse'
);
axis([
1
10

2
10
]);
xlabel(
'Time (measured in TRs, i.e. one timepoint every 4secs)'
);
ylabel(
'fMRI signal'
);

7.我们现在模拟出一个真实测量得到的生理信号体素激活值：
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

measured_voxel_data = [
1

1
12
8

1
5

3
1

2

1
]';
% This is is what often gets called
"y"
.
% This measured signal is probably some kind of mixture of
% a response to the word stimulus and a response to the object stimulus,
% with random noise thrown on top.
% Let's plot it
figure(
7
);
clf; % Clear the figure
plot(measured_voxel_data,
'o'
);
% Plot HRF against time, with one timepoint every TR seconds.
% A line with circles on it
grid on;
xlabel(
'Time (in units of TRs, 4s long each)'
);
ylabel(
'fMRI signal'
);
title(
'Measured voxel data'
);

8.进行数据拟合，矩阵求逆，求伪逆，然后绘图plot，进行比对：
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

%%% What is the estimated sensitivity vector of
this
voxel ?
%
% Well, we make the pseudoinverse of the design matrix, and multiply
% it by the vector of measured voxel data:
estimated_voxel_sensitivity = pinv(design_matrix) * measured_voxel_data;
%%% This estimated_voxel_sensitivity is what gets called betahat in the math.
%%% Let's display
this
in the workspace, by typing it without a semicolon
estimated_voxel_sensitivity
%%% This makes the following show up in the Matlab command window:
%
% estimated_voxel_sensitivity =
%
%
3.2965
%
1.0565
%
%%% So, the estimate is that
this
voxel is around
3
times more sensitive to
%%% words than it is to objects
%%% Now, let's make a plot of what the predicted response would be of
%%% a voxel that has a sensitivity matrix which is *exactly* our estimate,
%%% and compare it to the voxel response which we measured.
%%% They won't be exactly the same, because of the noise in the signal.
predicted_voxel_output = design_matrix * estimated_voxel_sensitivity;
%%% This predicted overall voxel output is just the
%%% predicted response to the word, plus the predicted response to the object.
%%% As we saw in hrf_tutorial.m, the idea that we can calculate the overall
%%% response simply by adding up these two separate responses is what it
%%% means to say that we are assuming that the system is LINEAR.
%%%
%%% If we want to look at the predicted responses to the separate stimulus
%%% types, we can calculate them by separately multiplying the
%%% corresponding column of the design matrix by the corresponding element
%%% of the estimated sensitivity vector.
predicted_response_to_word = predicted_word_response_column_vec * ...
estimated_voxel_sensitivity(
1
);
predicted_response_to_object = predicted_object_response_column_vec * ...
estimated_voxel_sensitivity(
2
);
%%%%% Let's plot all
this
figure(
8
);
clf; % Clear the figure
subplot(
3
,
1
,
1
); % This is just to make the plots line up prettily
hold on;
h1=stem(word_stim_time_series,
'b'
);
h2=stem(object_stim_time_series,
'r'
); % Word onset in blue, object onset in red
hold off;
grid on;
legend([h1(
1
) h2(
1
)],
'Word stim onset time'
,
'Object stim onset time'
);
axis([
1
10
0
1.2
]); % This just sets the display graph axis size
ylabel(
'Stimulus present / absent'
);
subplot(
3
,
1
,
2
);
hold on;
h1=plot(predicted_response_to_word,
'b*'
);
h2=plot(predicted_response_to_object,
'ro'
);
h3=plot(predicted_voxel_output,
'ms:'
,
'linewidth'
,
2
);
%%%
'ms:'
means plot in the colour magenta (m),
%%% with squares as the markers (s), using a dotted line (:).
%%% Then we make the width of the line broader, linewidth=
2
,
%%% so that it shows up better.
%%% Note that the predicted_voxel_output is simply the sum of
%%% predicted_response_to_word and predicted_response_to_object
grid on;
legend([h1 h2 h3],
'Predicted response to word'
, ...
'Predicted response to object'
,
'Predicted total voxel response'
);
axis([
1
10

3
14
]);
ylabel(
'fMRI signal'
);
subplot(
3
,
1
,
3
);
hold on;
h1=plot(measured_voxel_data,
'g^'
);
h2=plot(predicted_voxel_output,
'ms:'
,
'linewidth'
,
2
);
hold off;
grid on;
legend([h1 h2],
'Measured voxel response'
,
'Predicted voxel response'
);
axis([
1
10

3
14
]);
xlabel(
'Time (measured in TRs, i.e. one timepoint every 4secs)'
);
ylabel(
'fMRI signal'
);

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
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816

%%%% Tutorial on the basic structure of an fMRI design matrix, using Matlab
%%%% Written by Rajeev Raizada, July
23
,
2002
.
%%%%
%%%% This file follows up on a preceding one: hrf_tutorial.m
%%%%
%%%% Neither file assumes any prior knowledge of linear algebra
%%%%
%%%% Please mail any comments or suggestions to: raizada at cornell dot edu
%%%%
%%%% Probably the best way to look at
this
program is to read through it
%%%% line by line, and paste each line into the Matlab command window
%%%% in turn. That way, you can see what effect each individual command has.
%%%%
%%%% Alternatively, you can run the program directly by typing
%%%%
%%%% design_matrix_tutorial
%%%%
%%%% into your Matlab command window.
%%%% Do not type
".m"
at the end
%%%% If you run the program all at once, all the Figure windows
%%%% will get made at once and will be sitting on top of each other.
%%%% You can move them around to see the ones that are hidden beneath.
%%%%
%%%% Note that
this
tutorial only shows the method where the
%%%% design matrix assumes a specific shape to the HRF.
%%%% It is also possible to estimate the HRF without making
%%%% any assumptions about its shape. This is called using the
%%%% Finite Impulse Response method, or FIR.
%%%% This involves using a slightly more complicated designmatrix
%%%% than the one we make below.
%%%%
%%%% First, let's make a pretend minihrf, just to show examples.
%%%% This is similar in shape to the HRFs that we looked at in
%%%% the program hrf_tutorial.m, but it doesn't have as many timepoints.
%%%% One reason to use a shortened HRF like
this
is just to save typing!
%%%% But in fact,
this
is approximately what a real HRF would look like
%%%%
if
you only measured from it once every four seconds.
%%%% In fMRI, the time it takes to make a wholebrain measurement is called
%%%% the TR (Time
for
Repetition, although people say
"Repetition Time"
).
%%%% So,
this
HRF is similar to what we'd measure
%%%%
if
our scanner had a TR of
4
seconds. These days, fast scanners
%%%% can usually manage to get a wholebrain full of data in only 2s.
hrf_small = [
0
4
2

1
0
];
%%%% Plot it
figure(
1
);
clf; % Clear the figure
plot(
0
:
4
,hrf_small,
'o'
); % Plot HRF against a timevector [
0
,
1
,
2
,
3
,
4
]
%
'o'
means
"use a line with circles on it"
% Type
"help plot"
in the Matlab command window
% to get a list of all the linestyles and markers
% that you can use. There are lots of them!
grid on; % Overlay a dottedline grid on top of the plot
xlabel(
'Time (in units of TRs, 4s long each)'
);
ylabel(
'fMRI signal'
);
title(
'This is what an HRF would look like if you measure once every 4s'
)
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% Just as we did in hrf_tutorial.m, now we're going to make
% a timeseries of
1
's and 0'
s representing the times when stimuli
% are shown. These timeseries will be convolved with the HRF,
% in order to see what kinds of fMRI signals would be evoked in voxels
% that respond to the stimuli. These predicted responses will form
% the columns of our design matrix, as is shown in more detail below.
%
% Just
for
purposes of illustration, we're going to imagine that
% one of our stimuli is flashing up a word on the screen, and that
% the other is flashing up a picture of an object.
%
% These stimulus onsets will probably produce more complex patterns
% of neural firing than the sudden flash of light that we talked about
% in HST_hrf_tutorial.m, but we're going to ignore that complication
%
for
now. We'll simply suppose that each stimulus instantly kicks off
% its own standardshaped HRF.
% This is what's typically done in eventrelated fMRI, and it turns
% out that it usually works pretty well.
%%%%%%%%%%%%%%% Now suppose we present a word at time t=
2
word_stim_time_series = [
0
1
0
0
0
0
];
%%%%%%%%%%%%%%% And let's present a picture of an object at time t=
4
object_stim_time_series= [
0
0
0
1
0
0
];
%%%% Let's convolve these with our miniHRF to see what kind of fMRI
%%%% signals they would evoke in voxels which respond to words or pictures
predicted_signal_that_word_would_evoke = conv(word_stim_time_series,hrf_small);
predicted_signal_that_object_would_evoke = conv(object_stim_time_series,hrf_small);
%%% Let's plot all
this
figure(
2
);
clf; % Clear the figure
subplot(
3
,
1
,
1
); % This is just to make the plots line up prettily.
% The first number is how many rows of subplots we have:
3
% The second number is how many columns:
1
% The third number is which subplot to draw in: the first one.
% So, we end up with three plots stacked on top of each other,
% and we draw in the first one (which is the uppermost subplot)
hold on; %
"Hold"
is one way of putting more than one plot on a figure
h1=stem(word_stim_time_series,
'b'
);
% Stem makes a nice looking plot with lines and circles
h2=stem(object_stim_time_series,
'r'
); % Word onset in blue, object onset in red
% The
"h1="
and
"h2="
bits are called
"handles"
.
% They are pointers to the plots that we are making,
% which are the stem plots in
this
case
.
% Making handles like
this
is useful
for
manipulating
% pretty much any aspect of the plot afterwards.
% In
this
instance, we use them to put a legend on the plot.
% That's done by the
"legend"
command, a couple of lines below.
% There's no need to worry about these handles at
this
stage,
% I just wanted to explain what those mysteriouslooking h's
% were doing there. Usually you can make a nicelooking legend
% without worrying about handles, but it turns out that
%
for
stem plots they help to make the legend look better.
hold off;
grid on;
legend([h1(
1
) h2(
1
)],
'Word stim onset time'
,
'Object stim onset time'
);
% We use the h1 and h2 handles here.
% This helps us to get the right symbols displayed in the legend,
% in
this
case
, blue and red circles.
axis([
1
9
0
1.2
]); % This just sets the display graph axis size
% The first two numbers are the xaxis range:
1
to
9
% The last two numbers are the yaxis range:
0
to
1.2
ylabel(
'Stimulus present / absent'
);
subplot(
3
,
1
,
2
);
plot(predicted_signal_that_word_would_evoke,
'b*'
);
%
'b*'
means blue stars on a solid line
grid on;
legend(
'Wordsensitive voxel would give this fMRI signal'
);
axis([
1
10

1.5
7
]);
ylabel(
'fMRI signal'
);
subplot(
3
,
1
,
3
);
plot(predicted_signal_that_object_would_evoke,
'r^'
);
%
'r^'
means red triangles
% pointing up, lying on a solid line.
grid on;
legend(
'Objectsensitive voxel would give this fMRI signal'
);
axis([
1
10

1.5
7
]);
xlabel(
'Time (measured in TRs, i.e. one timepoint every 4secs)'
);
ylabel(
'fMRI signal'
);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%
%%%%%% What the design matrix has in it
%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%
%%%%%% Here's the key part.
%%%%%% The design matrix is built up out of these predicted responses.
%%%%%%
%%%%%% Each column of the design matrix is the predicted fMRI signal
%%%%%% that a voxel would give,
if
it were perfectly and exclusively
%%%%%% sensitive to a particular stimuluscondition.
%%%%%%
%%%%%% In our
case
, the first column of the design matrix
%%%%%% would be the vector
"predicted_signal_that_word_would_evoke"
%%%%%% that we made just above, and the second column would be
%%%%%% the vector
"predicted_signal_that_object_would_evoke"
%%%%%%
%%%%%% So, the most important part of the design matrix
%%%%%% is simply these two vectors sidebyside.
%%%%%%
%%%%%% A real design matrix would have some other columns in it too,
%%%%%% which have other types of predicted fMRI signals in them,
%%%%%% e.g. what the signal would look like
if
the scanner's output
%%%%%% were slowly drifting in time.
%%%%%% But those other columns don't deal with the signal that the stimuli
%%%%%% would be predicted to evoke in the brain, and so we can ignore them
%%%%%%
for
now. (Later in the HST583 course, Doug will talk more about
%%%%%% how you might model slowscanner drift etc.)
%%%%%%
%%%%%% It's the *columns* of the design matrix that get built up out of
%%%%%% these predicted responses to the different stimulus types,
%%%%%% but the actual vectors that we made above are row vectors,
%%%%%% i.e. just a bunch of numbers in a row.
%%%%%% So, to match the format of the design matrix, we need to turn
%%%%%% these into column vectors, by transposing them (i.e. flipping them).
%%%%%% We
do
this
by putting a dash/apostrophe at the end of the vector
predicted_word_response_column_vec = predicted_signal_that_word_would_evoke';
predicted_object_response_column_vec = predicted_signal_that_object_would_evoke';
%%% Now let's look at the actual vectors in the Matlab workspace window
predicted_word_response_column_vec % Because there is no semicolon after
this
,
% it will display in workspace window
predicted_object_response_column_vec
%%%%%% Now we can join these two column vectors together
%%%%%% to make the design matrix. We simply put the two columns sidebyside.
%%%%%% In Matlab, you make
new
matrices and vectors by
%%%%%% putting the contents inside [ square brackets ]
%%%%%% Note that to join them together in
this
way, they must be
%%%%%% the same length as each other.
%%%%%%
%%%%%% Because the names of my variables are so
long
and verbose,
%%%%%% the command below spills over onto two lines. In Matlab,
%%%%%% we can split a command over two lines by putting three dots ...
design_matrix = ... % The three dots here mean
"continued on the next line"
[ predicted_word_response_column_vec predicted_object_response_column_vec ];
design_matrix % No semicolon, so it displays in window
%%%%%% Translation guide:
%%%%%% In equations, the design matrix is almost always called X
%%%%%% Note that
this
is a capital
"X"
.
%%%%%%
%%%%%% X = design_matrix;
%%%%%%
%%%%%% Capitals are typically used
for
matrices, and small
case
is
%%%%%% used
for
vectors.
%%%%%% The only difference between a vector and a matrix is that
%%%%%% a vector is just a bunch of numbers in a row (a rowvector)
%%%%%% or a bunch of numbers in a column (a columnvector),
%%%%%% whereas a matrix is bunch of vectors stacked up next to each
%%%%%% other to make a rectangular grid, with rows *and* columns of numbers.
%%%%%% Now let's view a grayscale plot of the design matrix,
%%%%%% in the way that an fMRIanalysis
package
, such as SPM, would show it.
%%%%%% To
do
this
, we use the Matlab command
"imagesc"
.
%%%%%% This takes each number in the design matrix and represents
%%%%%% it as a colour, with the colour depending on how big the number is.
%%%%%% In
this
case
, we'll be using a gray colourscale, so low numbers
%%%%%% will be shown as darker grays, and high numbers are lighter grays.
%%%%%% The
"sc"
part at the end of
"imagesc"
stands
for
"scale"
, which
%%%%%% means that Matlab scales the mapping of numbers onto colours so
%%%%%% that the lowest number gets shown as black, and the highest as white.
%%%%%%
%%%%%% For examples of how to use the imagesc command to make
%%%%%% pictures of brainslices, see the companion program
%%%%%% showing_brain_images_tutorial.m
figure(
3
);
clf; % Clear the figure
imagesc(design_matrix); %
'imagesc'
maps the numbers to colors,
% normalising so that the max goes to white
% and the min goes to black
colormap gray; % Show everything in grayscale
colorbar; % Shows how the numbers lie on the colour scale
% Note that the highest number in the design matrix,
% which is
4
, is shown as white, and the lowest, 
1
,
% gets shown as black.
title(
'Grayscale view of design matrix'
);
xlabel(
'Each column represents one stimulus condition'
);
ylabel(
'Each row represents one point in time, one row per TR (every 4secs)'
);
%% Now suppose we have a voxel which responds only to words, not to objects.
%% We can calculate how it would be predicted to respond
%% to our word+object display as follows:
%%
%% Predicted response from wordsensitive voxel =
%%
1
* Response which wordpresentation would evoke
%% +
0
* Response which objectpresentation would evoke
%%
%% Note that
this
is how the voxel would be predicted to respond
%%
if
there were no noise whatsoever in the system.
%% Clearly a real fMRI signal would never be
this
clean.
%%
%% Now, let's make a
"sensitivity vector"
for
this
voxel,
%% in which each entry will say how sensitive that voxel is to
%% the corresponding stimulus condition.
%%
%% This voxel is sensitive to words, which are our *first* stimulustype.
%% And we made the predicted word response into the first column of
%% the design matrix.
%% So, the sensitivity of
this
voxel to words will be the first element
%% in the sensitivityvector.
%%
%% Similarly, the sensitivity of
this
voxel to the second stimulustype,
%% which are objects, will be the second element in the sensitivity vector.
%%
%% So, the sensitivity vector
for
a voxel with
%% sensitivity =
1
to the first stimulustype, which are words
%% and sensitivity =
0
to the second stimulustype, which are objects
%%
%% will be [
1
0
]
%%
%% I know
this
seems trivial !!
%% Things will get more interesting in a minute...
sensitivity_vec = [
1
0
]'; % The dash makes
this
a column vector
sensitivity_vec % No semicolon, so it displays in window
%% Translation guide:
%% In equations, the numbers in the sensitivityvector are typically
%% called
"betavalues"
, or sometimes
"betacoefficients"
or
"betaweights"
.
%% The columns of the design matrix are called
"regressors"
and
%% the value that is assigned to each regressor is the betavalue.
%%
%% Note that in the example above, we are pretending that we already *know*
%% how sensitive our voxel is to the various stimuli, but in the real world
%% we don
't know this. We'
re trying to figure out what stimuli our voxel
%% is sensitive to, using the fMRI data that we collect in the scanner.
%% This will be described more below.
%% In mathspeak, that means that we are trying to *estimate* the betas.
%% When people want to distinguish between the
true
betavalue
%% (which we don't know) and the estimated betavalue that we figure out
%% from our data, then they call the
true
one beta and
%% the estimated one
"beta hat"
(beta with a circumflex sign on top of it: ^
%% [ End of that part of the translation guide, back to the main theme... ]
%% So, we can now express our predicted voxel response in terms
%% of entries in the sensitivity vector multiplied by
%% columns in the design matrix:
%%
%% Predicted response from wordsensitive voxel =
%%
