-
Notifications
You must be signed in to change notification settings - Fork 54
/
spm_P_FDR.m
153 lines (138 loc) · 5.58 KB
/
spm_P_FDR.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
function [P] = spm_P_FDR(Z,df,STAT,n,Ps)
% Returns the corrected FDR P value
%
% FORMAT [P] = spm_P_FDR(Z,df,STAT,n,Ps)
%
% Z - height (minimum of n statistics)
% df - [df{interest} df{error}]
% STAT - Statistical field
% 'Z' - Gaussian field
% 'T' - T - field
% 'X' - Chi squared field
% 'F' - F - field
% 'P' - P - value
% n - Conjunction number
% Ps - Vector of sorted (ascending) p-values in search volume
%
% P - corrected FDR P value
%
%
% FORMAT [P] = spm_P_FDR(p)
%
% p - vector or array of all uncorrected P values, from which
% non-finite values will be excluded (but zeros and ones are kept)
%
% P - corrected FDR P values (a vector or array of the same shape as p)
%
%__________________________________________________________________________
%
% The Benjamini & Hochberch (1995) False Discovery Rate (FDR) procedure
% finds a threshold u such that the expected FDR is at most q. spm_P_FDR
% returns the smallest q such that Z>u.
%
% Background
%
% For a given threshold on a statistic image, the False Discovery Rate
% is the proportion of suprathreshold voxels which are false positives.
% Recall that the thresholding of each voxel consists of a hypothesis
% test, where the null hypothesis is rejected if the statistic is larger
% than threshold. In this terminology, the FDR is the proportion of
% rejected tests where the null hypothesis is actually true.
%
% A FDR proceedure produces a threshold that controls the expected FDR
% at or below q. The FDR adjusted p-value for a voxel is the smallest q
% such that the voxel would be suprathreshold.
%
% In comparison, a traditional multiple comparisons proceedure
% (e.g. Bonferroni or random field methods) controls Familywise Error
% rate (FWER) at or below alpha. FWER is the *chance* of one or more
% false positives anywhere (whereas FDR is a *proportion* of false
% positives). A FWER adjusted p-value for a voxel is the smallest alpha
% such that the voxel would be suprathreshold.
%
% If there is truely no signal in the image anywhere, then a FDR
% proceedure controls FWER, just as Bonferroni and random field methods
% do. (Precisely, controlling E(FDR) yields weak control of FWE). If
% there is some signal in the image, a FDR method should be more powerful
% than a traditional method.
%
% For careful definition of FDR-adjusted p-values (and distinction between
% corrected and adjusted p-values) see Yekutieli & Benjamini (1999).
%
%
% References
%
% Benjamini & Hochberg (1995), "Controlling the False Discovery Rate: A
% Practical and Powerful Approach to Multiple Testing". J Royal Stat Soc,
% Ser B. 57:289-300.
%
% Benjamini & Yekutieli (2001), "The Control of the false discovery rate
% in multiple testing under dependency". Annals of Statistics,
% 29(4):1165-1188.
%
% Yekutieli & Benjamini (1999). "Resampling-based false discovery rate
% controlling multiple test procedures for correlated test
% statistics". J of Statistical Planning and Inference, 82:171-196.
%__________________________________________________________________________
% Copyright (C) 2008-2014 Wellcome Trust Centre for Neuroimaging
% Thomas Nichols and Ged Ridgway
% $Id: spm_P_FDR.m 5824 2014-01-02 14:50:13Z guillaume $
%-Set Benjamini & Yeuketeli cV for independence/PosRegDep case
%--------------------------------------------------------------------------
cV = 1;
if nargin == 1
%-FORMAT [P] = spm_P_FDR(p), with p -> Z, (all uncorrected p-values)
%======================================================================
%-Exclude non-finite (allowing use on a masked image) and sort
%----------------------------------------------------------------------
mask = isfinite(Z);
Z = Z(mask);
S = numel(Z);
[Z, si] = sort(Z(:));
[i, ui] = sort(si); % "unsorting indices"
%-"Corrected" p-values
%----------------------------------------------------------------------
Z = Z*S./(1:S)'*cV;
%-"Adjusted" p-values
%----------------------------------------------------------------------
Z(end + 1) = 1; % (sentinel 1 at end, dropped in ui indexing below)
for i = S:-1:1
Z(i) = min(Z([i i+1]));
end
%-Unsort, unmask and return
%----------------------------------------------------------------------
Z = Z(ui);
P = nan(size(mask));
P(mask) = Z;
return
end
%-FORMAT [P] = spm_P_FDR(Z,df,STAT,n,Ps)
%==========================================================================
%-Calculate p-value of Z
%--------------------------------------------------------------------------
PZ = spm_z2p(Z,df,STAT,n);
%-Calculate FDR p-values
%--------------------------------------------------------------------------
% If Z is a value in the statistic image, then the adjusted p-value
% defined in Yekutieli & Benjamini (1999) (eqn 3) is obtained. If Z
% isn't a value in the image, then the adjusted p-value for the next
% smallest statistic value (next largest uncorrected p) is returned.
%-"Corrected" p-values
%--------------------------------------------------------------------------
S = length(Ps);
Qs = Ps*S./(1:S)'*cV;
%-"Adjusted" p-values
%--------------------------------------------------------------------------
P = zeros(size(PZ));
for i = 1:numel(PZ)
%-Find PZ(i) in Ps, or smallest Ps(j) s.t. Ps(j) >= PZ(i)
%----------------------------------------------------------------------
I = find(Ps>=PZ(i), 1 );
%-"Adjusted" p-values
%----------------------------------------------------------------------
if isempty(I)
P(i) = 1;
else
P(i) = min(Qs(I:S));
end
end