-
Notifications
You must be signed in to change notification settings - Fork 6
/
evaluateMotion.m
66 lines (60 loc) · 2.75 KB
/
evaluateMotion.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
function [outMotionData, stimulusCorrelatedMotion] = evaluateMotion(motionFile, vectorFile, tr)
% FORMAT [outMotionData, stimulusCorrelatedMotion] = evaluateMotion(motionFile, vectorFile, tr)
% Calculates framewise displacement and stimulus-correlated motion based on
% condition onsets/durations, motion information provided by SPM8's
% realignment tool, and TR.
% Author: Taylor Salo 141002-141209
%
%
% Inputs:
% motionFile: String pointing to rp*.txt file outputted by
% SPM's realignment tool. Contains nVolumes X 6
% array.
% vectorFile: String pointing to mat file where single
% block's vector information is stored.
% tr: Time to repetition, in seconds. Double.
%
% Outputs:
% outMotionData: Motion parameter array from motionFile with
% framewise displacement appended as 7th column.
% stimulusCorrelatedMotion: Cell array of structures. One cell for each
% condition. Structure contains name field with
% condition name and corr field with correlation
% between expected BOLD response and motion.
motionData = importdata(motionFile);
% In Power et al. 2012, "rotational displacements were converted from
% degrees to millimeters by calculating displacement on the surface of a
% sphere of radius 50 mm, which is approximately the mean distance from the
% cerebral cortex to the center of the head." Rotation is given in radians
% by SPM8, and the formula for length of an arc from radians and radius is
% length = radians * radius.
motionDataDisp = motionData;
motionDataDisp(:, 4:6) = motionData(:, 4:6) .* 50;
motionDataDeriv = [0 0 0 0 0 0; diff(motionDataDisp)];
framewiseDisplacement = sum(abs(motionDataDeriv), 2);
motionDataDerivAndFd = [motionDataDeriv framewiseDisplacement];
[nScans, ~] = size(framewiseDisplacement);
outMotionData = [motionData framewiseDisplacement];
vectors = load(vectorFile);
% Quantify Stimulus-Correlated Motion.
% Convolution method from spm.martinpyka.de/?p=41
stimulusCorrelatedMotion = cell(size(vectors.onsets));
for iCond = 1:length(vectors.onsets)
vec = zeros(nScans, 1);
singleCond = (int16(vectors.onsets{iCond}) / tr) + 1;
vec(singleCond) = 1;
for jDur = 1:int16(vectors.durations{iCond} / tr)
vec(singleCond + jDur) = 1;
end
xBF.dt = tr;
xBF.name = 'hrf';
bf = spm_get_bf(xBF);
U.u = vec;
U.name = {'reg'};
convreg = spm_Volterra(U, bf.bf);
stimulusCorrelatedMotion{iCond}.name = vectors.names{iCond};
for jMot = 1:7
stimulusCorrelatedMotion{iCond}.corr(jMot) = abs(corr(convreg, motionDataDerivAndFd(:, jMot)));
end
end
end