5.3. Log of progress 2018

  • Author: Felipe Orihuela-Espina
  • Created: December 17, 2018
  • Revised: January 21, 2018
  • Copyright (c) 2018 INAOE

Please note that advances indicated at a particular date, may actually refer to advances in the previous days/weeks.

5.3.1. Advances 17-Dec-2018

  • Version: v0.2
  • Responsible: FOE

Summary of changes:

  • Incorporation of the measuring of thickness to the GUI finished.
  • Depuration of the technical documentation with Sphinx. This affected documentation in most classes. As far as I can tell, it is now up-to-date, and ready for v0.2 delivery.
  • Integration of the informal reports to the Sphinx documentation done.
  • Started working on the translation of matlab’s code for manual edition of segmentation borders. This is yet unfinished.

5.3.2. Advances 13-Dec-2018

  • Version: v0.2
  • Responsible: FOE

Summary of changes:

  • Incorporation of the measuring of thickness to the GUI started but unfinished. This has involved among other things;
    • Definition of two new classes; src.IOT_GUI_MeasureThicknessParameterSettings for the controls and src.IOT_GUI_UtilitiesDock for the panel dock of the main window.
    • Adjustments of several methods spread throughout several classes.

5.3.3. Advances 3-Dec-2018

  • Version: v0.2
  • Responsible: FOE

Summary of changes:

  • Intensive testing: +20 bugs captured including syntax and logic. All sorted except for one of the logical ones. Initialization of a class attribute depends on the value of another attribute. Even if the later is declared in advance, it is non-existent until the object is created and thus, trying to check its value in the property methods yields an error.
  • Substantial changes to class src.IOT_Document, including encapsulation of attributes, and links with GUI, deprecation of all pairs get/set, the study is now an src.IOT_OCTscan (it cannot further be an np-array) affording greater consistency, and rebranding of attributes e.g. scanSegmented to segmentation (even though this might sound trivial, but it helps to avoid conceptual “link” to scans instead of volumes).
    • Marked task for v0.3: Upgrade to src.IOT_OCTvolume
  • New attribute .shape for src.IOT_OCTscanSegmentation for further internal consistency checks
  • The layer thickness measurement has been even further improved with just a trick. This algorithm although ready, is not yet available through the GUI.
  • Redefinition of the signature of the abstract method execute() in class src.IOT_Operation to permit the pass of parameters. Also, the return value changes from None to the result of the operation (this nevertheless remains to be stored in the attribute .result, but capturing it on the fly improves efficiency and code readability).

5.3.4. Advances 16-Nov-2018

  • Version: v0.2
  • Responsible: FOE

NOTE: Rodrigo has reported today that all basic functions in v0.1 are working correctly.

Summary of changes:

  • Attendance to urgent demands from Rodrigo:
    • Longitudinal reflectivity profiles (a.k.a. perfilometer): Ready and activated in the GUI. Either pixel or window can be chosen.
    • Segmentation lines modification: The spline base edition remains ready in matlab from ACA but pending translation to python and adaptation to the new data model by FOE. This is of outmost priority
    • Zoom: Ready and active. This function is automatically provided by matplotlib but we have it hidden for testing purposes. Acknowledging Rodrigo’s request, this has now been liberated.
      • This also include the option to save the images to .png, .ps/eps, .pdf and .svd. Since these are plain images and not our file format the saved image loses the document information. In reopening the file, this will therefore, be a plain picture, not an OCT segmentation.
      • Both the main OCT scan canvas and the perfilometer respond to the Zoom but they do it separatedly. It may be convenient to have these to respond in synchrony.
      • This also includes panning.
  • Improved measuring of layer thicknesses. If the pixel size is in the Amira file, then the pixel width is height from there, otherwise a default value is assumed.
  • Dummy segmentation has been largely improved. Now it covers the whole segmentation -which is what one should expect in segmentation- and it has a better handling of the BACKGROUND label.
  • Improved documentation of classes’ logs.

5.3.5. Advances 14-Nov-2018

  • Version: v0.2 beta
  • Responsible: FOE

Summary of changes:

  • New tabbed settings panel in the tools window. This will hold other parameter settings.
  • GUI controls for perfilometer settings have been added.
  • New src.IOT_GUI_PerfilometerParameterSettings for separated managing of GUI controls for the different operations. In the next weeks the panels for other operations will be added.
  • Communication between the main window and the tools window is now bidirectional.
  • Further testing on the improved stitching operation.
  • Annotated a task to “concentrate” the mail reports formally in the documentation for easier historical documentation.

Bug found:

  • Stitching works for the first execution (first 2 images), but crashes if a second execution is attempted (third image is stitched).

In addition to the above:

  • FOE has consumed the available quota of Git LFS. This will be a problem when the time comes to upload the new installers to GitHub. No clear solution is now available.

5.3.6. Advances 7-Nov-2018

  • Version: v0.2 beta
  • Responsible: FOE

Summary of changes:

  • Data model for v0.2 is now finished.
  • Improved stitching; no black stripes. Further, the resulting image now adapts its output size to the image size. Initial testing looks good, but more testing is needed.
  • Debugging of all classes inheriting from IOT_Operation
  • Loading of images now works from the GUI with the new data model. Now, a IOT_OCTscan is loaded instead of an nd.array.
  • Perfilometer is also responding now to the new data model. In the GUI, only the global behaviour is currently available. FOE will implement access to Perfilometer settings in the GUI asap.
  • Bug fixing:
    • Found and circumvented a bug in the panorama external library. This library does not work well with grayscale images, so I force an artificial conversion to RGB prior to calling panorama functions.

5.3.7. Advances 23-Oct-2018

  • Version: v0.2 beta
  • Responsible: FOE

Summary of changes:

  • Some tasks have been reorganized to give priority to things that are more pressing for Rodrigo (e.g. those less urgent delayed to v0.3, and those more urgent bring forward to v0.2)
  • Parameters of the perfilometer operation have been encapsulated so that they can be made accesible through the GUI. GUI access has yet to be implemented though.
  • Improvements to the data model; new class src.IOT_OCTscanSegmentation to support segmentation maps.
    • As a corollary, we have to revisit the rendering of the segmentation.
  • Migration of all operation to used the abstract method execute() has now been completed. This provides a uniform call for operations. Further, this has been made already considering classes src.IOT_OCTscan and src.IOT_OCTscanSegmentation. Initial testing is showing excellent performance.
    • All original operation specific methods have been declared deprecated.
  • String representation (method __str__) for some further classes have been created. Some others still remains. Those which inherit from src.IOT_Operation are presenting a bug due to some recursive calling.
  • Code cleaning. Some unnecessary import have been removed.
  • Bug fixing:
    • “leakage” of the algorithm for measuring the layers thickness has now been attended. Some further testing is needed.

5.3.8. Advances 1-Oct-2018

  • Version: v0.2 beta
  • Responsible: FOE

Summary of changes:

  • New classes src.IOT_OCTscan e src.IOT_OCTvolume for a more homogeneuous and extensible data model. Before this, we were working directly over the separated images which resulted in a code poor in efficiency and difficult to maintain and extend. These two classes are now ready but testing is needed.
  • Yet another large cleaning of code. Although without inmeadiate effects this is expected to be critical for forthcoming changes and implementation of new features.
  • Substantial improvement of the technical and user documentation. This is now available through readthedocs.
  • Improve error control with the warnings library.
  • Incorporation of the string representation (implementaiton of method __str__) for some of the classes. Others will follow.
  • Improved encapsulation through the use of decorator @property in several classes. Others will follow.
  • Attention to one of the pendings from v0.1; Class src.IOT_operation is now abstract. The abstract method execute() must be implemeted by child classes. Operands has been moved to superclass src.IOT_operation and arity is now calculated on the fly. A few operations have already been updated to deal with this new method, and the rest will follow soon.
  • Code for the src.AmiraReader has been liberated to return all scans again. Although this was available at earlier versions, but it was disabled while we develop v0.1 for the sake of sanity. Support for dealing with different scans has yet to be added to the GUI.

5.3.9. Advances 13-Sep-2018

  • Version: v0.1
  • Responsible: FOE

Summary of changes:

5.3.10. Advances 6-Sep-2018

  • Version: v0.1
  • Responsible: FOE

Summary of changes:

  • The installer appears to be working, but the installed .pyw (python’s equivalent to .exe) is not. The problem seems to be in the “linking” with pytonw.exe (python’s equivalent to command.com in Windows). JHV and FOW are now looking at this.

5.3.11. Advances 4-Sep-2018

  • Version: v0.1
  • Responsible: FOE

Summary of changes:

  • Version compiler and installer working. FOE opted for packing python on the installation to minimize risks of the application not working at Rodrigo’s machine. The price to pay is a very large installer (almost 600Mb -84Mb zipped-). Overhead is brutal! Over 450Mb! …but hopefully worth it.
  • JHV and SMH are now testing.

5.3.12. Advances 2-Sep-2018

  • Version: v0.1 alpha
  • Responsible: FOE

Summary of changes:

  • Improved separation of model (src.IOT_Document), view (src.IOT_GUI_* classes) and controllers (src.IOT_Operation and subclasses)
  • Polished GUI does no longer depend on QTDesigner
  • Mouse control is now working
  • All src.IOT_operations are now correctly connected to Document through the GUI
  • New class src.IOT_RetinalLayers for easier control of retinal layer informations
  • Connected GUI with basic delect and changeLabel EditSegmentation operations for ROI and COI.

Known issues:

  • The stitching algorithm still leaves the “black” regions
  • src.IOT_Document only follows one scan at a time. Liberate all scans in the Amira reader
  • Transformation from screen pixels to image pixels is missing. Algorithms for which the input depend on the mouse work as long as the document window is not resized. Upon resizing, there is risk of “index out of bounds”.
  • Dummy segmentation “only” paints default edges instead of a full image. This will make the src.IOT_OperationMeasureLayerThickness class to measure incorrectly. It is necessary to separate the segmentation map itself (all pixels in layer painted) from its representation (only top edge painted)
  • Document saving not released. Pictures can be saved by print screen only at this moment.
  • Advanced segmentation editing tools e.g. cubic splines line modification not yet incorporated.
  • Color of layers fixed. We need to provide a tool for selecting color layers
  • There is a need for an src.IOT_Settings class to store settings, both application-wide and study-specific. A simple map will do the job. We need one instance of this src.IOT_Settings for application settings and the another for the study. JSON can be used to save these to a .txt file if we do not want to get a full XML parser.

5.3.13. Advances 28-Aug-2018

  • Version: v0.1 beta
  • Responsible: FOE

Summary of changes:

  • The suboperations for edition of segmentation; remove and change label have been added to the GUI. This has been made both for COI (class of interest -global changes-) and ROI (region of interest -local changes-) based operations. Some testing is needed.
  • To avoid a third window with the operations settings/options the GUI has been modified.
  • Dependence on Qt’s Designer and on .ui files have been eliminated.
  • Class src.IOT_RetinalLayers has been created. This provides a better manipulation of tissue layers.

5.3.14. Advances 24-Aug-2018

  • Version: v0.1 beta
  • Responsible: FOE

Summary of changes:

  • Finally cracked on the mouse listening problem! A dummy matplotlib embedded in Qt window example has been prepared. The solution did not came from using QMouseEvent -this listens to events within the window, but NOT within the matplotlib canvas axes-. The solution required bypassing the matplotlib own events ( https://matplotlib.org/users/event_handling.html ) so that they can be listen by the container window. Now that the solution has been found, this should be incorportated to the application in the next few days.

GitHub commit/pull/push should be made as soon as this is attended.

5.3.15. Advances 17-Aug-2018

  • Version: v0.1 beta
  • Responsible: FOE

Summary of changes:

  • Not good news. The problem with the listening to mouse events freezing the application after just a few clicks remains open. Yesterday, JHV and FOE work on this for a while without spotting anything obvious. In the next few days we will be trying a plan B using Qt class QMouseEvent.

5.3.16. Advances 13-Aug-2018

  • Version: v0.1 beta
  • Responsible: FOE

Summary of changes:

  • Work on the issue of the listening to mouse events leaving the application frozen. This is a well known issue of pynput library for Windows* as reported in:

    https://pynput.readthedocs.io/en/latest/mouse.html#monitoring-the-mouse

    “The listener callbacks are invoked directly from an operating thread on some platforms, notably Windows. This means that long running procedures and blocking operations should not be invoked from the callback, as this risks freezing input for all processes. A possible workaround is to just dispatch incoming messages to a queue, and let a separate thread handle them.”

    …ergo, FOE has started to work on isolating the thread (done!) and queue messages (working on it). Hopefully it willbe sort out soon.

5.3.17. Advances 9-Aug-2018

  • Version: v0.1 beta
  • Responsible: FOE

Summary of changes:

  • The problem with the compilation remains open. Error on modules have been addressed and fixed. It seems to be working on console mode (it prints the message “OCT-Tools Initiating…” but there seems to be a problem with importing Qt.

5.3.18. Advances 6-Aug-2018

  • Version: v0.1 beta
  • Responsible: FOE

Summary of changes:

  • All operations have been now encapsulated and are working from the GUI.
  • There is no longer need to operate the steps in sequence (except of course opening the image for obvious reasons). Once the image is open, the work flow steps can be done in any order that fits the clinician. This sorts out the issue that operations have to be carried out step-by-step.
  • If an editSegmentation operation is attempted when there is still no “automatic” operation, an automatic default dummy segmentation is generated on the fly. This is important for Rodrigo who needs NOT to depend on an automatic segmentation.
  • Mouse monitoring for the manipulation of segmentation has noe been included over ACA functions. Nevertheless, this is currently disabled as it seems that listening to mouse events freezes the application.
    • We are currently using pynput but we should not discard alternative solutions.
  • A full set of labelled images from Duke university has been downloaded. It may be convenient to test the segmentation algorithm.
  • FOE has attempted a first full compilation with pyinstaller.
  • Bug fixing:
    • Alteration of the colour pallete by the perfilometer operation has been fixed.

The performance of the segmentation operation is pauper! Although, not a bug in the sense that it works, but this is not acceptable.

Pending for v1.0:

  • Manual manipulation is ready from ACA functions. From code, things can be manipulated, but without access to these from a GUI and with adequate mouse support, this is still insufficient. For practical matters, still useless.
  • Compilation; The distribution folder is created and the .exe is generated (sized >200Mb), but errors are reported during the generation of the .exe.
    • An alternative is to pack miniconda on the distribution and prepare a batch file that calls the python interpreter and executes run.py.
  • The executable of the advanced segmentation algorithm prepared by ACA in Matlab requires Matlab Runtime environment to be executed. Obviously, we CANNOT force the user to buy Matlab, and hence we must find a alternative; whether compile so that it can be run without Matlab Runtime environment (not sure if this is possible), or translate it to python, but this may not be trivial.

5.3.19. Advances 5-Aug-2018

  • Version: v0.1 beta
  • Responsible: FOE

Summary of changes:

  • All hardcoded paths have now been cleaned. As far as I can tell there is none remaining.
  • I have generated a minimal version where I have eliminated much of the code with a lot of internal tests that we have.
  • The processing functions (flattening, stitching and perfilometer) have been encapsulated. Still pending are segmentation and editSegmentation.
  • Bugs fixing:
    • File opening
    • Exiting from the menu option
  • Added button “0” to open the initial imagen
  • Revised and updated the AmiraReader which has now been encapsulated in a class.
  • Improved and enlarged code comments.
  • I have force the stitching operation to work on only 2 images at a time. One may still join as many as desired, but it will have to be done in pairs. For instance, if 3 images have to be stitched; you will have to make first 2, and then to the result add the 3rd. Although this works now, but it is not a desirable situation.

The above changes have improve this version a lot, although still some work is pending on the segmentation and editSegmentation operations. No commit to GitHub should be done until the version is fully functional.

  • Bugs found:
    • The perfilometer function does not get the image size correctly. The problem appears to be that Python’s skimage stores the images “linearized” (as ACA has previously warned FOE!).
    • I have forced the Amira reader to return only the first scan. RGB images for skimage are <width, height, filter(x3)> and typed uint8, whereas Amira images are grayscaled scans sized <width, height, scan(xn)> and typed float. Casting is needed here. Ideally, we should store in; <width, height, filter(x3), scan(xn)>
    • Flattening distorts the colour palette. I think I have provoked this during code cleaning.

5.3.20. Advances 3-Aug-2018

  • Version: v0.1 beta
  • Responsible: FOE

Summary of changes:

  • After some adjustements, the program now runs in my machine (under Windows) and still using the interpreter. For compilation, a few other issues have to be attended.

    • NOTE: FOE is using Pyzo with Miniconda, whereas ACA uses PyCharm which gives some problems because uses some non-standard libraries.

  • We have dependencies on Qt5 (for the interface), but also with SciPy (this is not too serious as it is a standard library of Python, yet it must be installed in the interpreter).

  • We have dependencies on OpenCV for the stitching operation which currently relies on external algorithm Panorama. While this is not ideal, but it works by now.

  • Hardcoded path in the perfilometer function has been removed.

  • Hardcoded paths in the mosaic function have been removed.

  • Bugs found

    ** Upon attempting to open a new image (menu File->New) but the open file operation is cancelled, you get an “out of index” error. This is easy to removed, it only requires a parameter checking (If ... is None) but I did not have the time to finish it today. ** The exit option on the File menu is not working. To exit the application, right now it is only possible using the “x” button on the window.

Right now, the operations flow works but separatedly; each operation on its in own. It would be convenient to modify the function emergentes so that it stores the working image, so that this is passed down from one step to the next. Although, this is not critical from the point of view of the algorithms actually working, but it is very inconvenient for the user. We CANNOT force the clinician to manually call every operation separatedly.

5.3.21. Advances 31-Jul-2018

  • Version: v0.1 beta
  • Responsible: ACA/FOE

Summary of changes:

  • Still unsolved the issue with hard paths.

5.3.22. Advances 17-Jul-2018

  • Version: v0.1 beta
  • Responsible: ACA/PHW

Summary of changes:

  • Reported by Rodrigo that the program does not starts up. This was found to be due to some remaining hard “paths”.

5.3.23. Advances 11-Jul-2018

  • Version: v0.1 beta
  • Responsible: ACA

Summary of changes:

  • Uploaded first version of the program and report to OSF. This version still has severe integration issues.