function summary=stat_summary(input_file, fwhm, df, mask_file, mask_thresh, ...
   input_thresh, flip, k);

%STAT_SUMMARY
%
% SUMMARY = STAT_SUMMARY( INPUT_FILE , [FWHM , [DF , [MASK_FILE, MASK_THRESH ...
%   [, INPUT_THRESH [, FLIP [, K ]]]]]] )
%
% Produces an SPM-style glass brain and summary analysis of a T or F statistic
% image. P-values for local maxima and cluster sizes are based on non-isotropic
% random field theory if an FWHM image is provided (or a Bonferroni correction, 
% if smaller). The random field theory is based on the assumption that
% the search region is a sphere (in isotropic space), which is a very tight 
% lower bound for any non-spherical region.
%
% Finds local maxima and clusters of INPUT_FILE above INPUT_THRESH
% and MASK_FILE above MASK_THRESH (ignored if empty; empty by default).
% If INPUT_THRESH <= 0.1 then it is taken as a probability and threshold is
% chosen so that the uncorrected P-value is INPUT_THRESH (0.001 by default),
% If MASK_THRESH is a vector [a b], a<=b, then mask is a<MASK_FILE<=b. 
% Clusters are voxels connected in any of the 6 diresctions. A local maximum 
% is a voxel which is greater than or equal to all its 6 neighbours,
% and strictly greater than at least one of them. 
%
% FWHM is the fwhm of a smoothing kernel applied to the data, either as a
% fwhm file from fmrilm or multistat, or as a scalar. Default is
% 0.0, i.e. no smoothing, which is roughly the case for raw fMRI data.
% For motion corrected fMRI data, use at least 6mm;
% for PET data, this would be the scanner fwhm of about 6mm.
%
% DF=[DF1 DF2; DFW1 DFW2] is a 2 x 2 matrix of degrees of freedom.
% If DF2 is 0, then DF1 is the df of the T statistic image.
% If DF1=Inf then it calculates thresholds for the Gaussian image. 
% If DF2>0 then DF1 and DF2 are the df's of the F statistic image.
%
% DFW1 and DFW2 are the numerator and denominator df of the FWHM image. 
% If DF=[DF1 DF2] (row) then DFW1=DFW2=Inf, i.e. FWHM is fixed.
% If DF=[DF1; DFW1] (column) then DF2=0 and DFW2=DFW1, i.e. t statistic. 
% If DF=DF1 (scalar) then DF2=0 and DFW1=DFW2=Inf, i.e. t stat, fixed FWHM.
% If any component of DF >= 1000 then it is set to Inf. Default is Inf. 
%
% If FWHM is estimated by FMRILM, set DFW1=DFW2=DF outputted by FMRILM;
% if FWHM is estimated by MULTISAT, set DFW1=DF_RESID, DFW2=DF outputted 
% by MULTISTAT.
%
% FLIP: INPUT_FILE is multiplied by FLIP before processing. For T statistic
% images, FLIP = -1 will look for negative peaks and clusters. Default is 1.
%
% K is the number of conjunctions. If K > 1, calculates P-values for peaks 
% (not clusters) of the minimum of K independent t or F maps. Default is 1.
%
% SUMMARY is a matrix with 12 columns. 
% Col 1: index of cluster, in descending order of cluster volume. 
% Col 2: volume of cluster in mm^3.
% Col 3: resels of cluster.
% Col 4: P-value of cluster extent.
% Col 5: P-value if the cluster was chosen in advance, e.g. nearest to an ROI.
% Col 6: values of local maxima, sorted in descending order.
% Col 7: P-value of local maxima.
% Col 8: P-value if the peak was chosen in advance, e.g. nearest to an ROI.
% Col 9: Q-value or false discovery rate ~ probability that voxel is not signal.
% Cols 10-12: x,y,z coords of local maxima in C notation i.e. starting at 0
%        so that they agree with 'register'.

%############################################################################
% COPYRIGHT:   Copyright 2002 K.J. Worsley, 
%              Department of Mathematics and Statistics,
%              McConnell Brain Imaging Center, 
%              Montreal Neurological Institute,
%              McGill University, Montreal, Quebec, Canada. 
%              worsley@math.mcgill.ca
%
%              Permission to use, copy, modify, and distribute this
%              software and its documentation for any purpose and without
%              fee is hereby granted, provided that this copyright
%              notice appears in all copies. The author and McGill University
%              make no representations about the suitability of this
%              software for any purpose.  It is provided "as is" without
%              express or implied warranty.
%############################################################################

% Defaults:

if nargin < 2
   fwhm=0
end
if nargin < 3
   df=Inf
end
if nargin < 4
   mask_file=[];
end
if nargin < 6
   input_thresh=0.001
end
if nargin < 7
   flip=1
end
if nargin < 8
   k = 1
end

if input_thresh <= 0.1
   input_thresh=stat_threshold(1,1,0,df,input_thresh,0.001,0.05,k)
end

[search_volume, num_voxels]= ...
   glass_brain(input_file,input_thresh,mask_file,mask_thresh,flip);

lm=locmax(input_file,input_thresh,mask_file,mask_thresh,fwhm,flip);

if isstr(fwhm)
    d=fmris_read_image(fwhm,0,0);
    numslices=d.dim(3);
    if ~isempty(mask_file)
        mask_thresh1=mask_thresh(1);
        if length(mask_thresh)>=2
            mask_thresh2=mask_thresh(2);
        else
            mask_thresh2=Inf;
        end
        d=fmris_read_image(mask_file,1:numslices,1);
        mask=d.data;
        mask= (mask>mask_thresh1 & mask<=mask_thresh2);
        d=fmris_read_image(fwhm,1:numslices,2);
        search_resels=sum(sum(sum(mask.*d.data)))
        num_voxels=sum(sum(sum(mask)))
    else
        d=fmris_read_image(fwhm,1:numslices,2);
        search_resels=sum(sum(sum(d.data)))
        num_voxels=d.dim(1)*d.dim(2)*d.dim(3)
    end
    average_fwhm=(search_volume/search_resels)^(1/3)
    [p_peak, p_cluster, p_peak1, p_cluster1]= ...
        stat_threshold(search_resels, num_voxels, 1, ... 
        df, lm(:,1), input_thresh, lm(:,7), k);
else
   [p_peak, p_cluster, p_peak1, p_cluster1]= ...
      stat_threshold(search_volume, num_voxels, fwhm, ... 
      df, lm(:,1), input_thresh, lm(:,6), k);
end

q_value = fdr_threshold( input_file, input_thresh, ... 
   mask_file, mask_thresh, df, lm(:,1), flip, k);

n=size(lm,1);
['clus    vol  resel  p-val  (one)   peak   p-val  (one)   q-val   x   y   z']
[repmat('  ',n,1) num2str(round(lm(:,5))) repmat('  ',n,1) ...
   num2str(round(lm(:,6))) repmat('  ',n,1) ...
   num2str(round(lm(:,7)*100)/100) repmat('  ',n,1) ...
   num2str(round(p_cluster*1000)/1000) repmat(' (',n,1)  ... 
   num2str(round(p_cluster1*1000)/1000) repmat(')  ',n,1)  ... 
   num2str(round(lm(:,1)*100)/100) repmat('  ',n,1)  ...  
   num2str(round(p_peak*1000)/1000) repmat(' (',n,1)  ...  
   num2str(round(p_peak1*1000)/1000) repmat(')  ',n,1)  ... 
   num2str(round(q_value'*1000)/1000) repmat('  ',n,1)  ... 
   num2str(round(lm(:,2:4)))]
summary=[lm(:,5:7) p_cluster p_cluster1 lm(:,1) p_peak p_peak1 q_value' lm(:,2:4)];
return