Lyman

--------------------------

This is a bugfix release that anyone using v0.0.5 should upgrade to.

Preprocessing workflow
~~~~~~~~~~~~~~~~~~~~~~

- Fix a bug that was introduced in v0.0.5 where the preprocessed timeseries
was not being written out by the DataSink.

-------------------------

Preprocessing workflow
~~~~~~~~~~~~~~~~~~~~~~

- Added a workaround some changes in later versions of FSL
that now return a de-meaned timeseries from the highpass filter.
In FEAT, the mean is replaced, and the rest of the processing carries
on as usual. Because I don't want to break compatability with older
versions of FSL, this adds back in the mean but only if it looks
like the filtered timeseries has been de-meaned. **Note**: This uses
a simple heuristic, which may not be robust in all cases, so it is
important to check that the signal-to-noise maps make sense if you are
doing something that expects a nonzero timeseries mean.

-------------------------

Infrastructure
~~~~~~~~~~~~~~

- Added continuous integration with TravisCI.

Mixed effects workflow
~~~~~~~~~~~~~~~~~~~~~~

- Fixed a bug where the analysis mask was getting smoothed on the surface.

FNIRT-based normalization workflow
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- Fixed a bug where the outputs of FNIRT were not properly renamed and thus
were not correctly picked up by the registration stage of the functional
pipeline.

Surface snapshots script
~~~~~~~~~~~~~~~~~~~~~~~~

- Fixed a bug where surface visualization would crash when the analysis mask
includes all vertices.

- Added a brief pause between updating the view and saving a snapshot to allow
redrawing to finish.

---------------------------

Preprocessing workflow
~~~~~~~~~~~~~~~~~~~~~~

- Added the ``coreg_init`` field to the experiment file. This is ``"fsl"`` by
default, which uses ``FLIRT`` to get a rough coregistration before using the
boundary-based algorithm (this was the old behavior). It can also be set to
``"header"``, which assumes that the functional and anatomy are roughly in
register in real space and that a mapping can be found with the header
geometry.

Model workflow
~~~~~~~~~~~~~~

- Added the ``memory_request`` field to the experiment file so that you can
request more memory on memory-intensive nodes (those involving model
estimation) when using a managed distribution engine (such as SGE). This can
be helpful for whole-brain high-resolution studies.

Anatomical normalization
~~~~~~~~~~~~~~~~~~~~~~~~

- Fixed a bug in ANTS-based anatomical normalization that affected non-OSX
systems. This bug caused a workflow crash, so if you haven't seen it, don't
worry about it.

----------------------

Anatomical normalization
~~~~~~~~~~~~~~~~~~~~~~~~

- Added ANTS-based volume normalization. This provides substantial improvements
over the FSL-based normalization that was previously used. However, ANTS can
be difficult to install, so this is optional and off by default. It controled
through a variable in the ``project.py`` file, ``ants_normalization``, which
should be either ``True`` or ``False``. After enabling it, you can use the
command-line tools as before, and ANTS will be used in ``run_warp.py`` and
``run_fmri.py -workflow reg``.

Preprocessing workflow
~~~~~~~~~~~~~~~~~~~~~~

- The artifact detection code now uses robust metrics (median and median
absolute deviation). Previously, it used mean and standard deviation.
**Importantly**, this means that the your intensity threshold should be
adjusted by a scaling factor to provide a similarly stringent threshold.
As a general rule of thumb, 1 SD is about 1.48 MADs.

- Added white noise spike detection. This is controlled through the
``spike_threshod`` in the experiment file. It is also in units of median
absolute deviation. It is ``None`` by default, indicating that no volumes
will be excluded for white noise spikes. Additionally, a plot that can be
used to diagnose spikes has been added to the artifact detection report.

- Changed the derivation of the brain mask. Previously, this mask was
intensity based (although the intensity threshold was determined within a
mask output by BET). Now, the Freesurfer segmentation is used to define
an anatomical brain mask, which is then transformed into native run
space. This should avoid losing voxels in magnetic susceptibility areas
like ventral temporal cortex.

- Otherwise updated the preproc report with better summary figures.

Subject-level modelling
~~~~~~~~~~~~~~~~~~~~~~~

- It should now be possible to run the model workflow on task-free data
(i.e. for functional connectivity analysis) by setting "``design_name``"
to ``None`` in the experiment file.

- Added computation and reporting of residual tSNR.

- Improved the colormaps used for reporting summary statistics about the
mode (residual variance, R squared, etc.)

- Improved the plot showing correlations between confound and task
variables

- Otherwise improved the logic and testing of the model workflow.

- Added to and improved the model report at the fixed effects stage.

Mixed effects workflow
~~~~~~~~~~~~~~~~~~~~~~

- Updated the mixed effects model reporting and simplified the workflow graph.

- The boxplot of COPE effect sizes in the mixed effects report is now taken
from a sphere (with the same size as in the activation peak image)
centered at each peak voxel rather than just from the single voxel
itself.