oleksiiparf / plsnpairs Goto Github PK

Test the results displayer with PET files and check for possible errors.

Plot only the points included in input mask.  Eventually want to go the
other way, i.e., select area of scatterplot and produce mask (probably
containing multiple disjoing clusters).

We have a ton of warnings that appear in Eclipse, which makes it hard to notice 
when we add more 
warnings.  Let's get rid of all the ones we can.

Provide an option for the user to either manually resize the left panel 
horizontally or provide an 
option to hide it.

For each brain image, we could probably just draw the crosshair and lag/slice 
labels onto the 
graphics of it on each call to the repaint() method rather than modify the 
image directly.  This would 
cause repaint to be slightly slower but would eliminate delays when we modify 
things such as the 
transparency and colour of the crosshair and lag/slice labels. 

Note: Labels have already been changed.

The histogram lacks certain features which could be useful, such as extra 
numbers and tick marks, 
as well as the 95% percentile marks or other percentile marks.

Create a plot manager class that performs checks on each results file as it is 
loaded by the results 
displayer and determines which can be allowed for which plot, and also 
displays/removes certain 
plots depending on what data is provided by the results files.

By doing this, we can potentially have both PLS and NPAIRS results files loaded 
in the results 
displayer together, with both their plots available.

Flip/mirror brain slices in the results displayer by either LVs/CVs or the 
entire results file. Also, 
consider how this would be done depending the brain view (axial, sagittal, 
coronal).

Want to create r vs p curves for a range of NPAIRS results.  Note that 
this is a meta-analysis plot.  

Currently need to click 'PLOT' before View updates; want it to happen
automatically when new View is chosen

Make the matrix library used by PLS consistent with what NPAIRS is using, and 
make sure the most 
efficient one is being used. Profile the speed, if possible.

Go through the JMatio code and try to optimize it as much as possible, or 
consider a new package 
for reading .mat files if possible

Currently, user must choose 'Deselect' in menu bar and select conditions to
include in analysis before saving Analysis setup params file. If user does
not do this, condition selection info is not initialized so analysis
crashes when user tries to run it.  

Solve this by initializing condition selection info to include all
conditions by default.

* Note also that deselect conditions option could be moved into main
Analysis window to make it more prominent; otherwise user can forget to
specify this info.  It was originally included in menu bar instead in order
to be as close as possible to PLS Analysis window.

see summary

In the progress dialog, there is a "Stop" button provided, which so far
just disposes of the dialog itself and its parent frame. Depending on what
the progress dialog is being used for, disposing the parent frame may not
be enough to end the currently-running process. This is the case for
running a PLS or NPAIRS analysis. This so far only works for loading the
results displayer since it completely relies on the parent frame, since
that is the window containing all the plots.

Edit option lets user change data path of input data volumes.  In PLS, 
data for each run is required to live in its own directory, and all run 
directories must be located in a single parent directory.  The Edit option 
lets the user change only the parent directory for all runs 
simultaneously.  

For NPAIRS, no such requirement exists for the data.  The Edit option for 
NPAIRS should let the user change the data path of each run's data 
separately.  The path change should be applied directly to the data files, 
not to the directory containing the data files as in PLS. 

Determine a way of displaying error messages when the application is being run 
from a jar file or 
from PLS.jnlp and not from Eclipse.

Review the PLS guide to determine which plots are not yet implemented in PLS 
and implement them.

We can currently save each individual plot due to JFreeChart, but we currently 
lack the ability to save 
anything in the main brain viewer.

TBD - must speak with Stephen to determine details of implementation

Include all variables required for PLS analysis in an Npairs session file;
fill with dummy or default values as appropriate (e.g. datamat prefix can
be resultsFilePrefix).  Then PLS analysis can load Npairs session files. 
Note that PLS users will still have to create datamat files before running
PLS analysis, since datamat files aren't generated for Npairs.

The ImageEventHandler class should be removed since it is unnecessary and 
redundant.

What steps will reproduce the problem?
1. Run and save Npairs analysis
2. Try to view results
3. Images can't be read in

Problem: CV eigenimages (full-data,avg and zs-avg) are transposed when
saved in results .mat file

Solution: don't transpose before saving in ResultSaver.java

Results are saved using whatever full-path prefix is supplied in Results
text field; if user wants all results saved in a single directory, the
directory must be explicitly created when using the file browser to select
the Results file name/prefix.  Should improve this so directory is
automatically created and files with same prefix as directory name are
saved in the directory. 

In the BrainData results display:

Since we are able to control the transparency for both the cross-hair and 
the slice/lag numbers, we can set both transparency values to 100% to make 
them disappear.  This makes the checkboxes to toggle cross-hair, and slice/
lag numbers redundant and add needless options for the user.  Can we 
remove the checkboxes?

Also I like that there is a button that brings up dialogues for these 
options now reducing clutter, but perhaps it would be *even better* to 
remove the buttons on the left hand side of the result panel and put the 
transparency sliders on the right hand side above the brain slices 
instead.  To me that's where they belong since those controls all 
manipulate the contents of the slice images (ex. magnification, 
orientation) whereas the ones on the left hand side seem more to 
manipulate what slices are displayed (ex. which slices/lags, which files).

If splits .vols file is uploaded, no check is done to see whether list of
indices in file 
(a) are within range
(b) include each input vol in at least one sample.

If some vols are not included in any samples, then the average
training/test CV Scores for those vols will be Nan.  

Note that we might eventually want the option to analyze only a subset of
the data, so it's not necessarily the case that we always require every vol
to be included somewhere.  

Want to be able to input analyze and nifti volumes as overlay (foreground)
images.  (Currently only able to load them as underlay (background) images.)

We do not want the user to be able to select coordinates and/or lags that are 
not currently visible. 
Do not give them the ability to select non-visible lags by filtering the drop 
down combo box.

Users currently have to re-type values in input fields whenever they run
the app. Provide a load/save option for them to restore/store the values
from these input fields so they do not have to keep manually typing them back.

NPAIRS currently runs using parameters found in an NPAIRSAnalysisSetup 
(npairs setup params) .mat file.  If the user fills in Analysis window 
fields and hits 'Run' without saving, then an error window pops up telling 
the user to save first.  But if the user loads a .mat file into the 
Analysis window and then changes some fields without resaving before 
pressing 'Run', NPAIRS actually runs using the parameters saved in the 
original .mat file instead of reading values directly from the GUI 
fields.  The user is not warned that this is happening.

SOLUTION: A message should pop up telling the user to save the new field 
info first.  

Note:
Could also refactor code to read Analysis field info directly into an 
NpairsSetupParams object instead of reading the info in via the .mat file, 
but in that case we should still warn the user that field info has changed 
and either ask if the new info should be saved or else save it 
automatically.  But the first Solution is the easiest to implement.

Viewer should be generalized so that a different background images can be
uploaded for each results file.  Currently, only a single background images
can be loaded, and it is applied to all results files.  

Note that the dimensions of each of the background images are still
expected to be the same.

Can currently set up and run only one NPAIRS analysis at a time from GUI.  
Each NPAIRS analysis requires user to fill in which PC dimensions to enter 
into CVA after PCA (PC dims given for full and split data as reg exp, 
e.g. '2, 4, 9-22').  Want to be able to give a range of number of PCs to 
keep, including incrementation step size.  

e.g.: PC range = '2-100', Step size = '2' 

--> do 50 npairs analyses: 
Analysis 1: use PC dims 1-2
Analysis 2: use PC dims 1-4
Analysis 3: use PC dims 1-6...
Analysis 50: use PC dims 1-100

Current Attach/Detach button location varies depending on the tab.  Can we 
make it consistent across all tabs?  Top-left as in the first tab seems 
fine.  Or even a small icon on the tab itself would suffice.

Ensure that each specific plot maintains the same range (and domain) even after 
switching between 
LVs/CVs or result files. (ie. the same max and min range values are always used)

This makes it easier for the user to compare data of different result files for 
the same plot.

Currently, all input scans are expected to have an assigned condition
(except for any 'scans skipped' at the beginning of a run.  If number of
input scans is more than number of scans assigned a condition, then we get
the following error:

Stack Trace:
java.lang.ArrayIndexOutOfBoundsException: -1
    at npairs.NpairsSetupSessionInfo.getSesStack Trace:
java.lang.ArrayIndexOutOfBoundsException: -1
    at
npairs.NpairsSetupSessionInfo.getSessionFileInfo(NpairsjSetupParams.java:1019)
    at npairs.NpairsSetupSessionInfo.<init>(NpairsjSetupParams.java:867)
    at npairs.NpairsjSetupParams.<init>(NpairsjSetupParams.java:348)
    at npairs.Npairsj.<init>(Npairsj.java:143)
    at pls.analysis.NpairsjAnalysis.doTask(NpairsjAnalysis.java:19)
    at
pls.chrome.shared.ProgressDialogWatcher.run(ProgressDialogWatcher.java:13)sionFi
leInfo(NpairsjSetupParams.java:1019)
    at npairs.NpairsSetupSessionInfo.<init>(NpairsjSetupParams.java:867)
    at npairs.NpairsjSetupParams.<init>(NpairsjSetupParams.java:348)
    at npairs.Npairsj.<init>(Npairsj.java:143)
    at pls.analysis.NpairsjAnalysis.doTask(NpairsjAnalysis.java:19)
    at pls.chrome.shared.ProgressDialogWatcher.run(ProgressDialogWatcher.java:13)

--------------------------------------------
Solution: apply default condition value to scans not included in onset info
and then 'exclude' this condition.

Provide a stacked bar chart option where applicable for plots when more than 
one result file is 
loaded (ie. an "all" option in the dropdown menu when switching between result 
files).

It would be nice if we had a script that could create signed versions of 
PLS.jar or PLS-experimental.jar and check it in under tags.  Not even 
sure if this is doable, I'll look into if Fatjar is able to sign jars 
which would be great because then this would probably be super easy.

When domain and range max and min are very small, they are not displayed
correctly.  Presumably the same would be true for very large values, since
it seems that the problem arises when they're small enough to be written in
exponential notation (e.g. 3.14159E-8); then they are rounded without
exponent info (e.g. just 3.1416).  

When images are rotated (90, 180, 270 degrees) then crosshairs do not
coincide with wherever the mouse is pointing.  If the images are rotated to
original orientation again, then crosshairs align properly.

Currently, user must save Analysis setup params info in .mat file before
running analysis, since Analysis info is read in directly from .mat file. 
If user does not save before running, then either:
(1) Error pops up telling user to save before running, if file wasn't loaded.
(2) If file was loaded and changes were made to some Analysis fields, then
analysis will run, but using orig field info saved in file loaded.

Also, when 'Save Npairs Setup Parameters' button is pressed, Save As window
always pops up with no default filename info displayed, hence user always
has to navigate path and enter filename from scratch - a big pain. 

TODO:
(1) Populate Save As window with loaded filename as default in the case
that file has been uploaded instead of making user nagivate from scratch.
(2) Consider what to do when fields have not been saved in file: should
user be forced to save?  

Want to be able to run 'agnostic CVA', i.e. CVA with every input timepoint
(scan) assigned to its own class.  The user currently has to input class
labels manually (but see Issue 22, adding options for textfile input). 
Want to add 'aCVA' checkbox or button as option on either sessionfile or
analysis screen.

Conditions/classes are currently laid out in a single row across Deselect 
frame; this gets awkward when there are many conds/classes.  Laying them 
out vertically without changing dims of frame is awkward as well, since 
the deselect frame seems to be fixed.  Try laying them out vertically but 
displaying more of them at once. 

missing approx. P-value in display - check where it's displayed in PLS
matlab gui and add it to Java

In the Matlab version of PLS, it is possible to look at an overview of the
Brain vs Behav scores, as well as an overview of the correlations.  It
basically displays all the groups and conditions / lvs at once.

This is useful because in the Matlab version of PLS, the user can print
that overview.  So this task is twofold:

1 - allow the user to look at an overview
2 - allow the user to print that overview as a single image, otherwise it
is mostly useless.

Currently a user has to mouse-over an item in a plot to see its value.  Perhaps 
the numbers 
themselves could also be displayed on the graphs. For example, display a value 
above each bar in a 
bar graph.

Many of the widgets are hidden when there is only one lag (i.e. when you view a 
block fMRI).  If you 
use the add/remove files dialog to remove the block file and add an event 
related file, the widgets 
do not update and you are stuck looking at lag 1 only.

Zooming in on domain/range just zooms towards middle of original
domain/range.  This means the plot itself can be zoomed right out of view.
 The zoom should centre on the middle of the plot instead.

Currently, class labels for CVA are assumed to be condition labels entered 
into session files.  Want to be able to run CVA using any user-specified 
class labels.  
* See also Issue

In Windows, the progress dialog is modal but the error dialogs that are 
generated if for example a datamat for a group is not found are also modal 
but show up underneath the progress dialog so it's very hard to figure out 
what's going on until you start getting bored and start switching windows 
and realise that there are errors hidden underneath this dialog.  I would 
much prefer these errors be displayed in the progress text area since 
there is no point in bringing up an error in which the user has to press 
okay to continue.  Just list all the errors unless the user can physically 
do something to avoid the error in which case a dialog is necessary.

Must update user guide whenever changes are made, e.g.:

- PLS analysis window info can now be saved
- NPAIRS analysis window file now saved as _NPAIRSAnalysisSetup.mat 
- Onsets can now be loaded from textfile
- other changes to NPAIRS analysis window: deselect,save,run options