Linking it all together

Back to Table of Contents

The individual processing steps that we saw in Basic Workflow can be linked together into a core pipeline for along tract processing.

trkPath         = fullfile(exDir, 'cst.trk');
volPath         = fullfile(exDir, 'dti_fa.nii.gz');
[header tracks] = trk_read(trkPath);
volume          = read_avw(volPath);

tracks_interp           = trk_interp(tracks, 50);
tracks_interp           = trk_flip(header, tracks_interp, [97 110 4]);
tracks_interp_str       = trk_restruc(tracks_interp);
[header_sc tracks_sc]   = trk_add_sc(header, tracks_interp_str, volume, 'FA');
[scalar_mean scalar_sd] = trk_mean_sc(header_sc, tracks_sc);

Serial processing

We can time these steps to get a feel for how long it might take if we were to scale up the process over many tracts and subjects.

profile on
tic

trkPath         = fullfile(exDir, 'cst.trk');
volPath         = fullfile(exDir, 'dti_fa.nii.gz');
[header tracks] = trk_read(trkPath);
volume          = read_avw(volPath);

tracks_interp           = trk_interp(tracks, 50);
tracks_interp           = trk_flip(header, tracks_interp, [97 110 4]);
tracks_interp_str       = trk_restruc(tracks_interp);
[header_sc tracks_sc]   = trk_add_sc(header, tracks_interp_str, volume, 'FA');
[scalar_mean scalar_sd] = trk_mean_sc(header_sc, tracks_sc);

toc
profile viewer

Output:

Elapsed time is 3.764054 seconds.

Parallel processing

Since most of the time is being spent fitting the splines and resampling the streamlines, we can get a nice boost in speed if we parallelize this step.

matlabpool 7
profile on
tic

trkPath         = fullfile(exDir, 'cst.trk');
volPath         = fullfile(exDir, 'dti_fa.nii.gz');
[header tracks] = trk_read(trkPath);
volume          = read_avw(volPath);

tracks_interp           = trk_interp(tracks, 50);
tracks_interp           = trk_flip(header, tracks_interp, [97 110 4]);
tracks_interp_str       = trk_restruc(tracks_interp);
[header_sc tracks_sc]   = trk_add_sc(header, tracks_interp_str, volume, 'FA');
[scalar_mean scalar_sd] = trk_mean_sc(header_sc, tracks_sc);

toc
profile viewer
matlabpool close

Output:

Elapsed time is 1.238922 seconds.

Scaling it up

The processing modules that we’ve seen so far can form the core of a loop as we process many subjects and tracts. A bit of pseudo-code looks like this

loop over subjects
    loop over tracts
        loop over hemispheres
            do along-tract processing steps
        end
    end
end

Back to Table of Contents