ReHo: quickly get the ranks for matrix

Hi, Xiao-Wei

I just write these here and suggest you use methods below to get the ranks more quickly:
(here I suppose symbol V means a 4D fMRI matrix)
%-----------------------------------------------
[Y,I]=sort(V,4);
[X,R]=sort(I,4);
%-----------------------------------------------
R should be the ranks 4D matrix, which can be used for ReHo computation. These two lines code make computation faster in MATLAB (around 10s for one 4D fMRI data).

Say hi to your guys
XINIAN

Dr. Zuo:

Thanks very much. One sort is enough. Dr. Wang Liang ever told me about this. And we also checked it about two years ago.
Example from Matlab is:

A = [ 3 7 5

      0 4 2 ];

>>> [B,IX] = sort(A,2)
B =
     3     5     7
     0     2     4
 
IX =
     1     3     2

     1     3     2 

IX is the rank.

You may check Chaogan's code, because the sort generate two big 4D matrix, so he divides the 4D matrix into many pieces first, and then use command sort to get the rank before rebuilding the full 4D rank.

Xiaowei

Hi, all! Dr. Zuo is right.
>> a=[2 7 5 3 5];
>> [B IX]=sort(a);
IX = 1 4 3 5 2
IX is not the correct rank!
>> [B IXIX]=sort(IX);
IXIX = 1 5 3 2 4
Then IXIX is the correct rank.

If you do not want to perform the “sort” calculation two times, you can use the following codes:
[B IX]=sort(a);
Ranks=1:length(a);
Ranks(IX)=Ranks;

But this is not the biggest challenge in rewriting the reho algorithm, the biggest one is how to solve the problem of “ties” (several time points which have the same value results in the same rank).
For example: A=[10 11 11 13 14 15 15 15 16];
Then the rank is not 1 2 3 4 5 6 7 8 9.
The correct rank which used in KCC computation should be: 1 2.5 2.5 4 5 7 7 7 9

I have solved this problem in the REST V1.3 Release 090422.
Maybe Dr. Zuo can provide some further revising advices.


function [] = reho(ADataDir, NVoxel, AMaskFilename, AResultFilename)
% Calculate regional homogeneity (i.e. ReHo) from the 3D EPI images.
% FORMAT function [] = (ADataDir, NVoxel, AMaskFilename, AResultFilename)
% Input:
% ADataDir Where the 3d+time dataset stay, and there should be 3d EPI functional image files. It must not contain / or \ at the end.
% NVoxel The number of the voxel for a given cluster during calculating the KCC (e.g. 27, 19, or 7); Recommand: NVoxel=27;
% AMaskFilename the mask file name, I only compute the point within the mask
% AResultFilename the output filename
% Output:
% AResultFilename the filename of ReHo result

% Written by Yong He, April,2004
% Medical Imaging and Computing Group (MIC), National Laboratory of Pattern Recognition (NLPR),
% Chinese Academy of Sciences (CAS), China.
% E-mail: yhe@nlpr.ia.ac.cn
% Copywrite (c) 2004,

% Info on the approach based on the reho:
% Zang YF, Jiang TZ, Lu YL, He Y and Tian LX, Regional Homogeneity Approach
% to fMRI Data Analysis. NeuroImage, Vol.22, No.1, 2004, 394-400.

% Info on the interesting and potential applications about the reho:
% He Y, Zang YF, Jiang TZ, Lu YL and Weng XC, Detection of Functional Networks
% in the Resting Brain. 2nd IEEE International Symposium on Biomedical Imaging:
% From Nano to Macro (ISBI'04), April 15-18, 2004, Arlington, USA.

% Info on the above can be found in:
% http://www.nlpr.ia.ac.cn/english/mic/YongHe
% ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
%Revised by Xiaowei Song, 20070421
%1.Add another parameter to allow user defined mask, so change some in code of selecting mask
% user defined mask has priority
% And delete the old parameter 'bMask', if the AMaskFilename is null(bMask=0) or 'Default'(bMask=1), then I would use 'ones mask' or 'default mask'
%2.Add waitbar for gui waiting and progress showing
%Revised by Xiaowei Song 20070903
% ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
% Revised by YAN Chao-Gan 080610: NIFTI compatible
% Revised by YAN Chao-Gan 080808: also support NIFTI images.
% Revised by YAN Chao-Gan, 090321. Result data will be saved in the format 'single'.
% Revised by YAN Chao-Gan, 090419. 1. Add Parameters: ADataDir and
% AResultFilename; Remove the old parameter 'NVolume'. 2. Fixed the bug
% of computing ReHo with 7 voxels or 19 voxels in a cluster.
% Algorithm re-written by YAN Chao-Gan, 090422. The algorithm has been re-written to speed up the calculation of ReHo.

if nargin~=4
error(' Error using ==> reho. 4 arguments wanted.');
end
theElapsedTime =cputime;

% Examine the Nvoxel
% --------------------------------------------------------------------------
if NVoxel ~= 27 & NVoxel ~= 19 & NVoxel ~= 7
error('The second parameter should be 7, 19 or 27. Please re-exmamin it.');
end

%read the normalized functional images
% -------------------------------------------------------------------------
fprintf('\n\t Read these 3D EPI functional images.\twait...');
[AllVolume,vsize,theImgFileList, Header] =rest_to4d(ADataDir);
[nDim1 nDim2 nDim3 nDim4]=size(AllVolume);
isize = [nDim1 nDim2 nDim3];
mask=rest_loadmask(nDim1, nDim2, nDim3, AMaskFilename);

%Algorithm re-written by YAN Chao-Gan, 090422. Speed up the calculation of ReHo.
%Saving a big 3D+time Dataset to small pieces by its first dimension to make this process run at least
% put pieces of 4D dataset to the temp dir determined by the current time
theTempDatasetDirName =sprintf('ReHo_%d', fix((1e4) *rem(now, 1) ));
theTempDatasetDir =[tempdir theTempDatasetDirName] ;
ans=rmdir(theTempDatasetDir, 's');%suppress the error msg
mkdir(tempdir, theTempDatasetDirName); %Matlab 6.5 compatible

Save1stDimPieces(theTempDatasetDir, AllVolume, 'dim1_');
clear AllVolume;%Free large memory

%rank the 3d+time functional images voxel by voxel
% -------------------------------------------------------------------------
fprintf('\n\t Calculate the rank of time series on voxel by voxel');

NumPieces_Dim1=10; %Constant number to divide the first dimension to "NumPieces_Dim1" pieces
NumComputingCount =ceil(nDim1/NumPieces_Dim1);
for x=1:(NumComputingCount),
rest_waitbar(x/NumComputingCount, ...
'Calculate the rank of time series on voxel by voxel. Please wait...', ...
'REST working','Child','NeedCancelBtn');

theFilename =fullfile(theTempDatasetDir, sprintf('dim1_%.8d', x));
theDim1Volume4D =Load1stDimVolume(theFilename);
theDim1Volume4D =double(theDim1Volume4D);

% Algorithm re-written by YAN Chao-Gan, 090422. Speed up the calculation of ReHo.
theDim1Volume4D=permute(theDim1Volume4D,[4,1,2,3]); % Change the Time Course to the first dimention
[TimePoints,M,N,O]=size(theDim1Volume4D);
[theDim1Volume4D,SortIndex] = sort(theDim1Volume4D,1);
db=diff(theDim1Volume4D,1,1);
clear theDim1Volume4D;
db = db == 0;
if x~=NumComputingCount
mask_x = mask(((x-1)*NumPieces_Dim1+1):(x*NumPieces_Dim1), :,:);
sumdb=squeeze(sum(db,1)).*mask_x;
else
mask_x = mask(((x-1)*NumPieces_Dim1+1):end, :,:);
sumdb=sum(db,1);
sumdb=reshape(sumdb,size(mask_x)).*mask_x;
end
SortedRanks=reshape(repmat([1:TimePoints]',M*N*O,1),[TimePoints,M,N,O]);
if any(sumdb(:))
TieAdjustIndex=find(sumdb);
for i=1:length(TieAdjustIndex)
[iM iN iO]=ind2sub([M N O],TieAdjustIndex(i));
ranks=SortedRanks(:,iM,iN,iO);
ties=db(:,iM,iN,iO);
tieloc = [find(ties); TimePoints+2];
maxTies = numel(tieloc);
tiecount = 1;
while (tiecount < maxTies)
tiestart = tieloc(tiecount);
ntied = 2;
while(tieloc(tiecount+1) == tieloc(tiecount)+1)
tiecount = tiecount+1;
ntied = ntied+1;
end
% Compute mean of tied ranks
ranks(tiestart:tiestart+ntied-1) = ...
sum(ranks(tiestart:tiestart+ntied-1)) / ntied;
tiecount = tiecount + 1;
end
SortedRanks(:,iM,iN,iO)=ranks;
end
end
clear db sumdb;
SortIndexBase=reshape(([1:M*N*O]'-1).*TimePoints,[M,N,O]);
SortIndexBase=repmat(SortIndexBase,[1,1,1,TimePoints]);
SortIndexBase=permute(SortIndexBase,[4,1,2,3]);
SortIndex=SortIndex+SortIndexBase;
clear SortIndexBase;
theDim1Volume4D(SortIndex)=SortedRanks;
theDim1Volume4D=reshape(theDim1Volume4D,size(SortedRanks));
clear SortIndex SortedRanks;
theDim1Volume4D=uint16(theDim1Volume4D); % Change to uint16 to get the same results of previous version.
%Save to file
theFilename =fullfile(theTempDatasetDir, sprintf('result_%.8d', x));
save(theFilename, 'theDim1Volume4D');
fprintf('.');

end

for x=1:(NumComputingCount)
rest_waitbar(x/(floor(nDim1/NumPieces_Dim1)+1), ...
'Rank 3D Brain reconstructing. Please wait...', ...
'REST working','Child','NeedCancelBtn');
theFilename =fullfile(theTempDatasetDir, sprintf('result_%.8d', x));
if x==1
I=Load1stDimVolume(theFilename);
else
I=cat(2,I,Load1stDimVolume(theFilename));
end
fprintf('.');
end
fprintf('\n');
[TimePoints,M,N,O]=size(I);

ans=rmdir(theTempDatasetDir, 's');%suppress the error msg

% calulate the kcc for the data set
% ------------------------------------------------------------------------
fprintf('\t Calculate the kcc on voxel by voxel for the data set.\n');
K = zeros(M,N,O);
switch NVoxel
case 27
for i = 2:M-1
for j = 2:N-1
for k = 2:O-1
block = I(:,i-1:i+1,j-1:j+1,k-1:k+1);
mask_block = mask(i-1:i+1,j-1:j+1,k-1:k+1);
if all(mask_block(:))
R_block=reshape(block,[],27); %Revised by YAN Chao-Gan, 090420. Speed up the calculation.
mask_R_block = R_block(:,reshape(mask_block,1,27) > 0);
K(i,j,k) = f_kendall(mask_R_block);
end %end if
end%end k
end% end j
rest_waitbar(i/M, ...
sprintf('Calculate the kcc\nwait...'), ...
'ReHo Computing','Child','NeedCancelBtn');
end%end i
fprintf('\t The reho of the data set was finished.\n');
rest_writefile(single(K),AResultFilename,isize,vsize,Header, 'single'); %Revised by YAN Chao-Gan, 090321. Result data will be stored in 'single' format. %'double');
case 19
mask_cluster_19=ones(3,3,3);
mask_cluster_19(1,1,1) = 0; mask_cluster_19(1,3,1) = 0; mask_cluster_19(3,1,1) = 0; mask_cluster_19(3,3,1) = 0;
mask_cluster_19(1,1,3) = 0; mask_cluster_19(1,3,3) = 0; mask_cluster_19(3,1,3) = 0; mask_cluster_19(3,3,3) = 0;
%Revised by YAN Chao-Gan, 090420. The element in the mask could be 1 other than 127. Fixed the bug of computing ReHo with 7 voxels or 19 voxels.
for i = 2:M-1
for j = 2:N-1
for k = 2:O-1
block = I(:,i-1:i+1,j-1:j+1,k-1:k+1);
mask_block = mask(i-1:i+1,j-1:j+1,k-1:k+1);
if all(mask_block(:))
mask_block=mask_block.*mask_cluster_19;
%Revised by YAN Chao-Gan, 090419. The element in the mask could be 1 other than 127. Fixed the bug of computing ReHo with 7 voxels or 19 voxels.
R_block=reshape(block,[],27); %Revised by YAN Chao-Gan, 090420. Speed up the calculation.
mask_R_block = R_block(:,reshape(mask_block,1,27) > 0); %Revised by YAN Chao-Gan, 090419. The element in the mask could be 1 other than 127. Fixed the bug of computing ReHo with 7 voxels or 19 voxels. %> 2);
K(i,j,k) = f_kendall(mask_R_block);
end%end if
end%end k
end%end j
rest_waitbar(i/M, ...
sprintf('Calculate the kcc\nwait...'), ...
'ReHo Computing','Child','NeedCancelBtn');
end%end i
fprintf('\t The reho of the data set was finished.\n');
rest_writefile(single(K),AResultFilename,isize,vsize,Header, 'single'); %Revised by YAN Chao-Gan, 090321. Result data will be stored in 'single' format. %'double');
case 7
mask_cluster_7=ones(3,3,3);
mask_cluster_7(1,1,1) = 0; mask_cluster_7(1,2,1) = 0; mask_cluster_7(1,3,1) = 0; mask_cluster_7(1,1,2) = 0;
mask_cluster_7(1,3,2) = 0; mask_cluster_7(1,1,3) = 0; mask_cluster_7(1,2,3) = 0; mask_cluster_7(1,3,3) = 0;
mask_cluster_7(2,1,1) = 0; mask_cluster_7(2,3,1) = 0; mask_cluster_7(2,1,3) = 0; mask_cluster_7(2,3,3) = 0;
mask_cluster_7(3,1,1) = 0; mask_cluster_7(3,2,1) = 0; mask_cluster_7(3,3,1) = 0; mask_cluster_7(3,1,2) = 0;
mask_cluster_7(3,3,2) = 0; mask_cluster_7(3,1,3) = 0; mask_cluster_7(3,2,3) = 0; mask_cluster_7(3,3,3) = 0;
%Revised by YAN Chao-Gan, 090420. The element in the mask could be 1 other than 127. Fixed the bug of computing ReHo with 7 voxels or 19 voxels.
for i = 2:M-1
for j = 2:N-1
for k = 2:O-1
block = I(:,i-1:i+1,j-1:j+1,k-1:k+1);
mask_block = mask(i-1:i+1,j-1:j+1,k-1:k+1);
if all(mask_block(:))
mask_block=mask_block.*mask_cluster_7;
%Revised by YAN Chao-Gan, 090419. The element in the mask could be 1 other than 127. Fixed the bug of computing ReHo with 7 voxels or 19 voxels.
R_block=reshape(block,[],27); %Revised by YAN Chao-Gan, 090420. Speed up the calculation.
mask_R_block = R_block(:,reshape(mask_block,1,27) > 0); %Revised by YAN Chao-Gan, 090419. The element in the mask could be 1 other than 127. Fixed the bug of computing ReHo with 7 voxels or 19 voxels. %> 2);
K(i,j,k) = f_kendall(mask_R_block);
end%end if
end%end k
end%end j
rest_waitbar(i/M, ...
sprintf('Calculate the kcc\nwait...'), ...
'ReHo Computing','Child','NeedCancelBtn');
end%end i
fprintf('\t The reho of the data set was finished.\n');
rest_writefile(single(K),AResultFilename,isize,vsize,Header,'single'); %Revised by YAN Chao-Gan, 090321. Result data will be stored in 'single' format. %'double');
otherwise
error('The second parameter should be 7, 19 or 27. Please re-exmamin it.');
end %end switch
Ken = K;
theElapsedTime =cputime - theElapsedTime;
fprintf('\n\tRegional Homogeneity computation over, elapsed time: %g seconds\n', theElapsedTime);

% calculate kcc for a time series
%---------------------------------------------------------------------------
function B = f_kendall(A)
nk = size(A); n = nk(1); k = nk(2);
SR = sum(A,2); SRBAR = mean(SR);
S = sum(SR.^2) - n*SRBAR^2;
B = 12*S/k^2/(n^3-n);

function Save1stDimPieces(ATempDir, A4DVolume, AFilenamePrefix)
%Save the 1st dimension of the 4D dataset to files
NumPieces_Dim1=10; %Constant number to divide the first dimension to "NumPieces_Dim1" pieces
NumComputingCount =ceil(size(A4DVolume,1)/NumPieces_Dim1);
for x = 1:(NumComputingCount),
%for x = 1:(floor(size(A4DVolume,1)/NumPieces_Dim1)+1)
rest_waitbar((x/NumComputingCount), ...
'Cut one Big 3D+time Dataset into pieces of 3D+time Dataset Before ReHo. Please wait...', ...
'REST working','Child','NeedCancelBtn');

theFilename =fullfile(ATempDir, sprintf('%s%.8d',AFilenamePrefix, x));
if x~=NumComputingCount
the1stDim = A4DVolume(((x-1)*NumPieces_Dim1+1):(x*NumPieces_Dim1), :,:,:);
else
the1stDim = A4DVolume(((x-1)*NumPieces_Dim1+1):end, :,:,:);
end
save(theFilename, 'the1stDim');
end

function Result=Load1stDimVolume(AFilename)
%Load the 1st dimension of the 4D dataset from files, return a Matrix not a struct
Result =load(AFilename);
theFieldnames=fieldnames(Result);
% Result =eval(sprintf('Result.%s',the1stField));%remove the struct variable to any named variable with a matrix
Result = Result.(theFieldnames{1});

Hi Chao-Gan,

Please refer to this matlab function: tiedrank.

hope this helps!

XINIAN

     Thanks a lot!
     Tiedrank only can calculate with vector but not matrix.
     Actually, the ReHo algorithm re-written by me was referenced from tiderank.m

Error | Forum of resting-state fMRI

Error

The website encountered an unexpected error. Please try again later.