Aivia Software

(Pre-version 10.5.1) 3D EM Object Analysis

Owned by Hoyin Lai
Last updated: Jan 10, 2023 by Aiden Maloney-Bertelli (Unlicensed)
4 min read

The 3D EM Object Analysis recipe in Aivia is designed to segment neuron structures in 3D electron microscopy images and create surface reconstructions of the detected structures. It is strongly recommended that you specify a deep learning model for EM segmentation as a preprocessing step for this recipe or otherwise feed a confidence map for neuron structures into the recipe. There are EM segmentation models offered in our model library.

The 3D EM Object Analysis recipe generates standard mesh surfaces. To create dendrite objects, you can use the 3D Neuron Analysis - EM recipe.

Parameters and Presets

Parameters

Recipe parameters for 3D EM Object Analysis and their descriptions are summarized in the table below.

Preset Group

Parameter Name

Min Value

Max Value

Description

Preset Group

Parameter Name

Min Value

Max Value

Description

Detection



Intensity Threshold

0

255 (8-bit)

65,535 (16-bit)

Specifies the minimum intensity for segmentation on the input image or the output of the preprocessing deep learning model applied to the input image, if a model is given; a lower value leads to detection of larger objects as well as regions of higher uncertainty

Reconstruction Type

Not applicable

Allows the user to specify the 3D reconstruction algorithm to use for segmentation; there are two (2) options:

  • Use Robust Reconstruction uses the robust reconstruction algorithm for segmenting 3D objects based on connectivity between adjacent slices of detection; additionally, the algorithm applies slight smoothing and fill holes to enhance the object morphology; use this option for segmenting objects in a very dense environment.

  • Use 3D Watershed uses the watershed algorithm for segmenting 3D objects based on the intensity gradient of the probability map; use this option for segmenting sparse objects.

Subset Filtering



Min Object Size

0

1 x 1012 (px3 or µm3)

Specifies the range of objects to be included in the analysis results based on the volumes of the detected objects

Smoothing Factor

0

10

Adjusts the amount of smoothing applied to the surface reconstructions of the detected objects; a lower value leads to the generation of surfaces with greater similarity to the input image

 

Presets

There are two (2) preset groups in the recipe: Detection and Subset Filtering. Each group has three (3) pre-configured parameter groupings to help you get started on the analysis. The default preset values are given in the sections to follow.

Detection

Parameter Name

Low

Medium

High

Parameter Name

Low

Medium

High

Intensity Threshold

51 (8-bit)

13,107 (16-bit)

128 (8-bit)

32,768 (16-bit)

242 (8-bit)

62,258 (16-bit)

 

Subset Filtering

Parameter Name

Small

Medium

Large

Parameter Name

Small

Medium

Large

Min Object Size

5 (px3 or µm3)

20 (px3 or µm3)

10,000 (px3 or µm3)

Smoothing Factor

1

2

5

 

Measurements

The 3D EM Object Analysis recipe generates morphological measurements for each detected 3D object. You can add additional measurements to the analysis results using the Measurement Tool in Aivia. The measurements generated by the recipe are as follows:

  • Surface Area

  • Volume

 

Tutorial

Before beginning the tutorial, please download the 3D EM Analysis Demo image as well as the this deep learning model based on a 3D U-Net architecture trained on the ISBI 2013 dataset from Kasthuri et al. (If you have run the models installer for Aivia, the previously linked deep learning model is located by default at "C:\Program Files\SVision LLC\Aivia {version number}\deeplearning\models\Segmentation_3D_EM_ISBI_2013.acmodel.") For information on how to select presets or modify parameter values, please refer to the tutorial on how to use the Recipe Console.



  1. Unzip the demo file and load the demo image, "3DEMAnalysisDemo.tif," into Aivia.

  2. In the Recipe Console, click on the Recipe selection dropdown menu and select the 3D EM Object Analysis recipe.

  3. In the Input and Output section of the Recipe Console, click on the Browse icon for the Deep Learning Model field, navigate to and select the "Segmentation_3D_EM_ISBI_2013.acmodel" model in the dialog, and then click Open.

  4. Select the Low or Medium preset for the Detection preset group.

  5. Click on the caret to the left of the Subset Filtering preset group to show the preset parameters.

  6. Change the Subset Filtering preset group parameter values to the following:

    • Min Object Size: 100,000

    • Smoothing Factor: 2

  7. Click the Start button or press the F4 key on your keyboard to begin applying the recipe to the image.

The 3D EM Object Analysis recipe generates two (2) output object sets: Cross Sections and Surfaces. The Cross Sections output allows you to view the cross-section outlines of the detected objects in Main View (2D); the Surfaces output allows you to view the surface reconstructions of the detected objects in 3D View.

Browse for a deep learning model to use for preprocessing

 

 

Results

Output cross sections (at z = 5)

 

 

Image credits

Narayanan "Bobby" Kasthuri and Jeff Lichtman, Harvard University, http://docs.neurodata.io/kasthuri2015/

Related articles

Â