LabelGeometryImageFilter not using physical space

dzenanz · January 30, 2019, 1:51pm

Looking at the code of LabelGeometryImageFilter, it seems to be doing all its calculation in index space, including volume and centroid. To add insult to injury, centroid is representated using PointType which is usually used for physical points.

Should this class be fixed to use ContinuousIndex instead of PointType to represent and return these parameters? Or should it start using physical space, perhaps with a backwards compatibility switch?

matt.mccormick · January 30, 2019, 2:20pm

Good catch @dzenanz.

Before the ITK 5 release is a good time to make the change to physical space with a backwards compatibility switch.

blowekamp · January 30, 2019, 2:54pm

It has been a few years but when I look into this filter there were quite a number of issues including several places where there were issues with physical spacing, orientation, and efficiency in the computation. As I recall the way this filter computes some geometry is to convert all pixels of a label into continuous index space, then do matrix computation on the full list. This was very memory inefficient for me causing up to 16X memory usage compared to the number of pixels.

I have spent effort adding all the features and computations that the LabelGeometryImageFilter into the LabelMap statistics filters. This includes quite comprehensive testing of many complex and corner cases. This OBB are computed very efficiently from just the started and ending pixels in the line.

IMHO, this filter is redundant and should be removed to a remote module. If there is something missing from the LabelMaps statistics filter we can add that there.

EDIT:
The other option is to make the LabelGeometry filter an adapter class to the calculations done in the LabelShapeStatistiscsImageFilter…

dzenanz · January 31, 2019, 4:49pm

I changed my code to use LabelImageToShapeLabelMapFilter, thanks for pointing it out. Part of the reason I asked about it is because I remember @blowekamp adding tests and fixing something about a shape filter.

Or it could be moved into ITKDeprecated module. Refactoring LabelGeometryImageFilter will take time, and it is indeed redundant. What are other opinions about this?

blowekamp · January 31, 2019, 5:08pm

I am in favor of moving into the deprecated module.

I was just looking at interfaces, it looks like the LabelGeometryFilter also can optionally produce images for each label. In my only work on this topic, I created a LabelObject which can store the image, and separate filters to computer either axis aligned or oriented images for each label. Recently I have just manually resampled for label object/image out side of a filter like what was done here:

github.com

InsightSoftwareConsortium/SimpleITK-Notebooks/blob/master/Python/35_Segmentation_Shape_Analysis.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h1 align=\"center\">Focused Ion Beam Scanning Electron Microscopy Image Segmentation</h1>\n",
    "\n",
    "\n",
    "**Summary:**\n",
    "1. SimpleITK supports a large number of filters that facilitate classical segmentation algorithms (variety of thresholding algorithms, watersheds...).\n",
    "2. Once your data is segmented SimpleITK enables you to efficiently post process the segmentation (e.g. label distinct objects, analyze object shapes).\n",
    "\n",
    "This notebook will illustrate the use of SimpleITK for segmentation of bacteria from a 3D Focused Ion Beam Scanning Electron Microscopy (FIB-SEM) image. The specific bacterium is <a href=\"https://en.wikipedia.org/wiki/Bacillus_subtilis\">bacillus subtilis</a>, a rod shaped organism naturally found in soil and plants. The bacteria have been subjected to stress to initiate the process of forming an endospore. These endospores can be seen as a generally dark ellipsoid inside the individual bacterium. \n",
    "\n",
    "Data provided courtesy of the [Laboratory of Cell Biology](https://electron.nci.nih.gov/), Center for Cancer Research, National Cancer Institute, National Institutes of Health."
   ]
  },
  {
   "cell_type": "code",

This file has been truncated. show original