Generate figures illustrating the DNB results

Contents

Download the DNB results (if necessary)

% download results
DNBdownloadresults;

% construct path to the DNB root directory
dnbdir = absolutepath(strrep(which('DNBrun'),'DNBrun.m',''));

% create directory for figures
mkdirquiet(fullfile(dnbdir,'figures'));
mkdirquiet(fullfile(dnbdir,'figures','scatter'));

Downloading results.mat (please be patient).
Downloading is done!
Downloading figures.tar (please be patient).
Downloading is done!

Load in a .mat file that contains the DNB results

load(fullfile(dnbdir,'DNBresults','results.mat'));

Check the workspace

whos

  Name                  Size               Bytes  Class    Attributes

  allR2                21x11            92086208  cell               
  brainmasks            1x21             2331648  cell               
  denoisemethods        1x11                1522  cell               
  denoisenames          1x11                1550  cell               
  dnbdir                1x14                  28  char               
  meanvols              1x21             9319536  cell               
  voxelselections       1x21             2331648  cell               

Write out spatial maps of cross-validated R^2 values

% for each dataset
for p=1:size(allR2,1)
  fprintf('writing cross-validated R^2 maps for dataset %d...',p);

  % for each denoising method
  for q=1:size(allR2,2)

    % if we have results available
    if ~isempty(allR2{p,q})

      % construct the filename
      file0 = sprintf(fullfile(dnbdir,'figures','dataset%02d_%s.png'),p,denoisemethods{q});

      % write out the spatial map with a hot colormap that ranges from 0% to 100%.
      % the R^2 values are square-rooted before conversion to colors such that
      % the dynamic range of the color map is concentrated at the low end.
      imwrite(uint8(255*makeimagestack(signedarraypower(allR2{p,q}/100,0.5),[0 1])),hot(256),file0);

    end

  end

  fprintf('done.\n');
end

writing cross-validated R^2 maps for dataset 1...done.
writing cross-validated R^2 maps for dataset 2...done.
writing cross-validated R^2 maps for dataset 3...done.
writing cross-validated R^2 maps for dataset 4...done.
writing cross-validated R^2 maps for dataset 5...done.
writing cross-validated R^2 maps for dataset 6...done.
writing cross-validated R^2 maps for dataset 7...done.
writing cross-validated R^2 maps for dataset 8...done.
writing cross-validated R^2 maps for dataset 9...done.
writing cross-validated R^2 maps for dataset 10...done.
writing cross-validated R^2 maps for dataset 11...done.
writing cross-validated R^2 maps for dataset 12...done.
writing cross-validated R^2 maps for dataset 13...done.
writing cross-validated R^2 maps for dataset 14...done.
writing cross-validated R^2 maps for dataset 15...done.
writing cross-validated R^2 maps for dataset 16...done.
writing cross-validated R^2 maps for dataset 17...done.
writing cross-validated R^2 maps for dataset 18...done.
writing cross-validated R^2 maps for dataset 19...done.
writing cross-validated R^2 maps for dataset 20...done.
writing cross-validated R^2 maps for dataset 21...done.

Inspect a cross-validated R^2 map

figure;
imageactual(fullfile(dnbdir,'figures','dataset01_GLMdenoise.png'));
% Cross-validated R^2 for GLMdenoise on Dataset 1

Write out spatial maps that provide additional information

% for each dataset
for p=1:length(meanvols)
  fprintf('writing additional maps for dataset %d...',p);

  % write out the mean volume
  file0 = sprintf(fullfile(dnbdir,'figures','dataset%02d_meanvol.png'),p);
  imwrite(uint8(255*makeimagestack(meanvols{p},1)),gray(256),file0);

  % write out a binary mask indicating which voxels were selected to compare denoising methods
  file0 = sprintf(fullfile(dnbdir,'figures','dataset%02d_voxelselection.png'),p);
  imwrite(uint8(255*makeimagestack(voxelselections{p},[0 1])),gray(256),file0);

  % write out the brain mask which helped determine the voxel selection
  file0 = sprintf(fullfile(dnbdir,'figures','dataset%02d_brainmask.png'),p);
  imwrite(uint8(255*makeimagestack(brainmasks{p},[0 1])),gray(256),file0);

  fprintf('done.\n');
end

writing additional maps for dataset 1...done.
writing additional maps for dataset 2...done.
writing additional maps for dataset 3...done.
writing additional maps for dataset 4...done.
writing additional maps for dataset 5...done.
writing additional maps for dataset 6...done.
writing additional maps for dataset 7...done.
writing additional maps for dataset 8...done.
writing additional maps for dataset 9...done.
writing additional maps for dataset 10...done.
writing additional maps for dataset 11...done.
writing additional maps for dataset 12...done.
writing additional maps for dataset 13...done.
writing additional maps for dataset 14...done.
writing additional maps for dataset 15...done.
writing additional maps for dataset 16...done.
writing additional maps for dataset 17...done.
writing additional maps for dataset 18...done.
writing additional maps for dataset 19...done.
writing additional maps for dataset 20...done.
writing additional maps for dataset 21...done.

Inspect these additional maps

figure;
imageactual(fullfile(dnbdir,'figures','dataset01_meanvol.png'));
% Mean volume for Dataset 1

figure;
imageactual(fullfile(dnbdir,'figures','dataset01_voxelselection.png'));
% Voxel selection for Dataset 1

figure;
imageactual(fullfile(dnbdir,'figures','dataset01_brainmask.png'));
% Brain mask for Dataset 1

For each denoising method, calculate the median cross-validated R^2 value achieved on each dataset

% initialize
summary = zeros([size(allR2) 3]);  % this will hold the median and the 16th and 84th percentiles of the bootstrapped medians

% for each dataset
for p=1:size(allR2,1)
  fprintf('calculating median R^2 values for dataset %d...',p);

  % for each denoising method
  for q=1:size(allR2,2)

    % if we don't have results, just insert NaNs
    if isempty(allR2{p,q})
      summary(p,q,1:3) = NaN;

    % otherwise, calculate the median and then a confidence interval via bootstrapping.
    % note that we consider only those voxels in the binary mask voxelselections{p}.
    else
      [d,d,summary(p,q,1:3)] = calcmdsepct(allR2{p,q}(voxelselections{p}));
    end

  end

  fprintf('done.\n');
end

calculating median R^2 values for dataset 1...done.
calculating median R^2 values for dataset 2...done.
calculating median R^2 values for dataset 3...done.
calculating median R^2 values for dataset 4...done.
calculating median R^2 values for dataset 5...done.
calculating median R^2 values for dataset 6...done.
calculating median R^2 values for dataset 7...done.
calculating median R^2 values for dataset 8...done.
calculating median R^2 values for dataset 9...done.
calculating median R^2 values for dataset 10...done.
calculating median R^2 values for dataset 11...done.
calculating median R^2 values for dataset 12...done.
calculating median R^2 values for dataset 13...done.
calculating median R^2 values for dataset 14...done.
calculating median R^2 values for dataset 15...done.
calculating median R^2 values for dataset 16...done.
calculating median R^2 values for dataset 17...done.
calculating median R^2 values for dataset 18...done.
calculating median R^2 values for dataset 19...done.
calculating median R^2 values for dataset 20...done.
calculating median R^2 values for dataset 21...done.

Visualize a summary of the cross-validation results on a bar chart

% for visibility, we will divide up the figure into several plots.
% here we define which groups of datasets to show on a single plot.
ixs = {1:7 8:13 14:21};

% loop over groups of datasets
for p=1:length(ixs)

  % first, write out the figure to a .png file
  figure; set(gcf,'Visible','off','PaperPositionMode','auto');
  setfigurepos([100 100 900 275]); hold on;
  h = bar(summary(ixs{p},:,1),1);  % plot the medians
  colormap(jet);
  for mm=1:length(h)
    tempxd = get(get(h(mm),'Children'),'XData');
    errorbar2(mean(tempxd([1 3],:),1),summary(ixs{p},mm,1)', ...
              squish(summary(ixs{p},mm,2:3),2)','v','r-','LineWidth',1);
  end
  set(gca,'XTick',1:length(ixs{p}),'XTickLabel',ixs{p});
  xlabel('Dataset');
  ylabel('Cross-validated R^2');
  legend(h,denoisenames,'Location','EastOutside');
  filename = sprintf(fullfile(dnbdir,'figures','summary%d.png'),p);
  print('-dpng','-r72',filename);
  close;

  % then, show this .png file in a figure window
  figure;
  imageactual(filename);

end

Write out scatter plots that show a detailed comparison of each pair of denoising methods on each dataset

% for each dataset
for p=1:size(allR2,1)
  fprintf('writing scatter plots for dataset %d...',p);

  % for each denoising method
  for d1=1:size(allR2,2)

    % for each denoising method
    for d2=1:size(allR2,2)

      % skip if results are unavailable or if we are comparing a method to itself
      if isempty(allR2{p,d1}) || isempty(allR2{p,d2}) || d1==d2
        continue;
      end

      % write out the figure to a .png file
      figure; set(gcf,'Visible','off','PaperPositionMode','auto');
      setfigurepos([100 100 500 500]); hold on;
      scatter(allR2{p,d1}(voxelselections{p}), ...
              allR2{p,d2}(voxelselections{p}),25,'r.');
      ax = axis;
      mn = min(ax([1 3]));
      mx = max(ax([2 4]));
      axis([mn mx mn mx]);
      axissquarify;
      axis([mn mx mn mx]);
      straightline(0,'h','y-');
      straightline(0,'v','y-');
      xlabel(sprintf('%s (cross-validated R^2)',denoisenames{d1}));
      ylabel(sprintf('%s (cross-validated R^2)',denoisenames{d2}));
      title(sprintf('Dataset %d',p));
      filename = sprintf(fullfile(dnbdir,'figures','scatter','scatter_%s_vs_%s_dataset%02d.png'), ...
                                  denoisemethods{d1},denoisemethods{d2},p);
      print('-dpng','-r72',filename);
      close;

    end

  end

  fprintf('done.\n');
end

writing scatter plots for dataset 1...done.
writing scatter plots for dataset 2...done.
writing scatter plots for dataset 3...done.
writing scatter plots for dataset 4...done.
writing scatter plots for dataset 5...done.
writing scatter plots for dataset 6...done.
writing scatter plots for dataset 7...done.
writing scatter plots for dataset 8...done.
writing scatter plots for dataset 9...done.
writing scatter plots for dataset 10...done.
writing scatter plots for dataset 11...done.
writing scatter plots for dataset 12...done.
writing scatter plots for dataset 13...done.
writing scatter plots for dataset 14...done.
writing scatter plots for dataset 15...done.
writing scatter plots for dataset 16...done.
writing scatter plots for dataset 17...done.
writing scatter plots for dataset 18...done.
writing scatter plots for dataset 19...done.
writing scatter plots for dataset 20...done.
writing scatter plots for dataset 21...done.

Inspect one of these scatter plots

figure;
imageactual(fullfile(dnbdir,'figures','scatter','scatter_GLMstandard_vs_GLMdenoise_dataset01.png'));

Published with MATLAB® R2012b