serhii.net

DAMN.

Figures – Quarto:

:::: {layout="1,1,1],[1],[1,1,1" layout-valign="center"}

if it’s a layout I do get a valign thing. I just have to use layouts, not columns. Damn.

This doesn’t work on SVGs for me:

![](images/02/svgs/02_seqsiii.svg){.absolute top=0 left=0}

This does:

::: {.absolute top=0 left=0}
![](images/02/svgs/02_seqsiii.svg)
:::

SO.

One can literally generate a plot w/ graphviz, export as svg, open and edit in inkscape, save as optimized svg, paste into quarto, and manually add the correct class="fragment" data-fragment-index=4 bits to the relevant groups.

Insert with the usual {=html} thing.

Damn.

LeaderLine

almost works with quarto, except that line.position needs to be called on slide change — ergo likely won’t work in pdf export mode.

maurosilber/pyfragments: Animated figures for Quarto installs as-is

IT WORKS FOR PDF MODE TOO! And uses fragments

# | fig-width: 15
# | fig-height: 10
# | output: asis
import matplotlib.pyplot as plt
from pyfragments import AnimatedFigure

with AnimatedFigure() as ani:
# Data for plotting
  train_loss = [1.0, 0.6, 0.3, 0.2, 0.1]
  val_loss = [1.2, 0.8, 0.5, 0.4, 0.6]
  iterations = [1, 2, 3, 4, 5]

  plt.xlim(0, 6)
  plt.ylim(0, 1.5)

# Labels and legend
  plt.xlabel('Trainingsiterationen', fontsize=14)
  plt.ylabel('Loss', fontsize=14)
  plt.legend(loc='upper right', fontsize=12)

# Plot
  for x in range(len(train_loss)+1):
    with ani.fragment():
      plt.plot(iterations[:x], train_loss[:x], 'bo-', label='Training', linewidth=2)
      plt.plot(iterations[:x], val_loss[:x], 'go-', label='Validation', linewidth=2)

# Add a dashed vertical line in the middle
  mid_iteration = (iterations[0] + iterations[-1]) / 2

  with ani.fragment():
    plt.axvline(x=mid_iteration, color='r', linestyle='--', linewidth=1.5)

I couldn’t make a graph large enough to fill the screen, increasing fig-width didn’t help.

Solution: increase both fig-width and fig-height!

{dot}
//| fig-width: 12
//| fig-height: 9
digraph G {
    rankdir=LR;
	..

Execution Options – Quarto has figure options that lists the default sizes of figures based on output formats. Couldn’t find them because was looking in figures/graphviz etc. pages, not execution. …

For reveal slides, it’s 9 x 5.

Apparently it didn’t want to increase the 5 till I explicitly told it to, then width started increasing as well.

Both superscripts and subscripts

… are hard and you have to use a table.

digraph Neural_Network {
    rankdir=LR;
    ranksep=1.3;
    
    node [shape=circle, style=filled, fontcolor=white, fontsize=25, fillcolor="blue", color="black"];

    subgraph cluster_0 {
        node [fillcolor="#2c3e50", style="filled"];
        x1 [label=<
        <TABLE border="0" cellborder="0" cellspacing="0">
        <TR><TD rowspan="2" style="...">X</TD><TD style="...">1</TD></TR>
        <TR>                                  <TD style="...">1</TD></TR>
        </TABLE>>];
        x2 [label=<
        <TABLE border="0" cellborder="0" cellspacing="0">
        <TR><TD rowspan="2" style="...">X</TD><TD style="...">1</TD></TR>
        <TR>                                  <TD style="...">2</TD></TR>
        </TABLE>>];
    }

    sum [label=<∑<FONT color="yellow" point-size="10">(⎰)</FONT>>,  fillcolor="#27ae60", width=0.8, height=0.8, fixedsize=true];
    y [label=<y<sup>1</sup>>];
    
    edge [style=solid, color="#2c3e50"];
    x2 -> sum;
    x1 -> sum;

    edge [style=solid, color="#27ae60"];
    sum -> y;

    {rank=same; x1; x2;}
}

Misc

• Papers
  • Comparison of Three 3D Segmentation Software Tools for Hip Surgical Planning - PMC: compares 3d slicer, Syngo, and Materialize¹
    • 
  • Comprehensive Review of 3D Segmentation Software Tools for MRI Usable for Pelvic Surgery Planning - PMC²: 12 software tools
    • really nice criteria, incl. extensibility etc.
    • 
    • 
    • TL;DR: ITK-SNAP most usable, Myrian Studio second but windows-only, 3d Slicer third but neat because easily extendable. Seg3d as also mentioned as aving very good usability
• Misc
  • https://github.com/neurolabusc/niivue-images sample nii images for evaluating nii viewers/editors

+ 3d Slicer

• Main file: 240715-1653 Notes on 3d-slicer, 240718-1530 First python 3d slicer interactions
• License:
  • About 3D Slicer — 3D Slicer documentation
  • custom license: BSD-style, broadly open source
    • no GPL/copyleft code allowed because it’d make it harder to mix slicer w/ private intellectual property
    • commercial use OK
    • 3d slicer based products — only some are listed, because one can use it and not disclose it and it’s fine
• Language: python
• extensible in python as well
  • has many cool extensions: @KitwareMedical/slicer-extensions-webapp
  • includes MONAILabel active learning extension plugin
  • a whole category for developer tools! @KitwareMedical/slicer-extensions-webapp
  • Extension Wizard — 3D Slicer documentation
    • https://extensions.slicer.org/catalog/Shape Analysis/32952/linux has extensions about stats
  • ai-assisted-annotation-client/slicer-plugin/README.md at master · NVIDIA/ai-assisted-annotation-client
• Very active community and frequent updates
• Downsides
  • J:
    • too specific for medical images
    • presets hard to customize (I agree!)
    • bad documentation (I think fixed now)
  • but: “Best of the free ones”
• Ha: https://github.com/NVIDIA/ai-assisted-annotation-client/blob/master/slicer-plugin/README.md

+ Syngo

• syngo.via OpenApps – Siemens Healthineers Deutschland
• Paid.
• Best according to ¹
• No easy way to test.

MeVisLab

• https://www.mevislab.de/
• Commercial and non-commercial version available
• Documentation:
  • Getting Started
• SDK for scripting exists
  • python QT bindings available
  • and generally feels expandable w/ python, shows python errors in the main screen for example, which is a good sign
• Official video tutorials exist, incl. for application development: MeVisLab - YouTube
• Download: a single .bin file that has to be run
• I don’t really like the website but the program itself feels nice!
• Nice pipeline visualization: and adding blocks — much less chaotic than 3d slicer (at first glance)
• overall good impression
• python scripting has a nice reference etc.

ITK-SNAP

• TODO
• Winner of ² due to its usability, does segmentation only but does it well
• available on AUR
• tutorial: TutorialSectionVolumes
• I really like it as first impression
• Has a distributed segmentation service already implemented!
  • ITK-SNAP Distributed Segmentation Services
  • 
  • ASHS Documentation - Overview sample of such a service
  • (can’t test it because pyqt6 present but not found)
• has volumes and statistics, maybe expandable
• BUT can’t find easy ways to add plugins

+ 3D Seg

• Open Source, segmentation-first
• for linux has to be compiled from source
• last meaningful commits from ~3 years ago: Commits · SCIInstitute/Seg3D, last release 2021 as well
• tried to build it, as ./build, failed because can’t find svn for
• Has detailed-ish developer documentation as videos! https://www.sci.utah.edu/cibc-software/seg3d.html
• Couldn’t build locally sadly, error in the middle of it
• Giving up

Myrian

• commercial, but non-commercial exists, and can be extended with plugins

NeuroMorph

• NeuroMorph ‐ EPFL
  • NeuroMorph | BIII
• Blender-based!

• NeuroMorph is a set of tools designed to import, analyze, and visualize mesh models in Blender. It has been developed specifically for the morphological analysis of 3D objects derived from serial electron microscopy images of brain tissue, but much of its functionality can be applied to any 3D mesh.

ilastic

• ilastik - ilastik
• integrat-able with fiji: ilastik/ilastik4ij: ImageJ plugins to run ilastik workflows and with everything else
• BioImage.IO integration as well, btw another model zoo

Fiji

• Fiji: ImageJ, with “Batteries Included”
• Legendary
• infinite plugins, very well known, downloadable as executable
• not too easy to use in my subjective opinion, menus inside menus, but…

Kitware glance

https://kitware.github.io/glance/doc/index.html

• web-based, extendable
• Kitware/glance: Kitware Glance is an open-source web application developed at Kitware for visualizing volumetric images, molecular structures, geometric objects, and point clouds. It is part of the ParaView platform and can serve as a foundation for building custom web-based visualization applications involving ITK.js and VTK.js.

Biaflows

• https://biaflows-sandbox.neubias.org/#/
• fully-online and can run models
• I don’t like the 3d viewer and I think it’s segmentation only, no easy way to generate/show stats

Commit messages starting with #14 whatever are awkward as # is the default comment in git rebase and friends.

git config core.commentchar ";"

fixes that for me.

For a one-time thing this works as well:

git -c core.commentChar="|" commit --amend

(escaping - Escape comment character (#) in git commit message - Stack Overflow)

Why is the number of voxels different from the number of matrices? - Support - 3D Slicer Community

>>> raw = getNode('probe1_0000.nii.gz')
>>> slicer.util.arrayFromVolume(raw).shape
(180, 180, 500)
>>> 

doesn’t work for segmentations because not a volume.

Image dimensions can be seen from volume information And are identical in both segmentations and original

I found smoothing in view controllerS!

In Slice controllers there’s this arrow button that allows to set basically layers AND SMOOTHING per slice!

So everything is as expected wrt smoothing — there was none.

Zooming in slices w/ CTRL!

// ref: 240701-1752 NII MRI annotation tools

• 3D Slicer image computing platform | 3D Slicer
• DICOM seems primary format¹ but nifti is supported
• Installing from AUR

• Importing 2x .nii as directory worked as-is

• Volumes can do settings per file, incl.:

  • annotations threshold starts at 1

• Volume renderings gives 3d views?

• Annotations can be imported as annotations/segmentations through add data! Then they are parsed semantically

• One can segment N slices and it magically creates a 3d shape out of it!

• Exporting annos as nifti possible through “export” in Segmentation module ²(not Segment Editor!)

• https://www.youtube.com/watch?v=CxwxYtizdhE resampling volumes