Skip to content

2. Cell counts and ganglia analysis

Jump to bottom
Pradeep Rajasekhar edited this page Mar 21, 2022 · 2 revisions

Analysing number of neurons

Neurons can be detected in a field of view as long as the cells are stained with Hu, a pan neuronal marker that labels the cell soma. The workflow below will show you how to analyze the number of neurons.

Sample image used for example below: GFAP/Hu

  • Select GAT -> Analyze Neurons (Hu only)

gat_neuron_tut_1

  • For this image,

    • Channel 1: GFAP
    • Channel 3: Hu

  • As we are only interested in the number of neurons, we enter the channel number for Hu. If you do not know, enter NA into the field. You will have the option for entering it earlier, however, it is easier to prefill information as much as possible.

gat_neuron_tut_1

  • We will use the default pixel size, probability and overlap threshold values. Click OK when done.
  • GAT will open the image and run the segmentation after which you will get the prompt below.

gat_neuron_tut_1

  • You can use this step to add or delete ROIs from the ROI Manager as you please. Once you are done, click OK.
  • Once the analysis is complete, you will have a folder called “Analysis” within which there will be a subfolder with the name of the image you analysed. This will contain:
  • Table with no. of neurons
  • ROI Manager with neuron outlines
  • Label image where each neuron is an individual colour.

gat_neuron_tut_1

Optimising cell detection

On occasion, you may run the analysis and realize that the software doesn't detect all cells.

low detection

Within GAT, you can optimise the probability of detection during analysis:

Test segmentation Analyze Neuron (Tick Finetune detection)
test analyse

Increasing probability will detect brighter objects and reduce background detection, whereas decreasing probability will detect dim cells, but may pick up background. Refer to section 1 on using Test segmentation for choosing the right parameters.

In the example above, decreasing the probability threshold gave better accuracy:

Good detection GAT Stardist

Segmenting ganglia

Two channels

In the above image, we had GFAP in channel 1 which can be used to detect the ganglia. This can be used to determine the no of ganglia and cells per ganglia.

  • To do this, select Analyse Neurons (Hu only) and make sure you tick Cell_counts_per_ganglia. Choose DeepImageJ and enter the channel number for GFAP in the box.

Tip: It doesn’t have to be GFAP, it could be any channel that labels the neuron or glial fibres and/or soma. As long as you have a combination of Hu and a “ganglia” marker, you should be able to segment the ganglia.

ganglia_seg

  • Click OK and follow the same procedure as neuronal segmentation.
  • As we selected to count cells per ganglia, deepImageJ will be run to segment the ganglia based on colocalization of Hu and GFAP.

ganglia_seg

DeepImageJ segmentation Overlay of segmentation
  • You can now correct the ganglia outline using the brush tool. Click on the image below to watch the instruction video.
Correct ganglia outline

Single channel

In the latest release, Hu staining can be used to define the ganglia. This requires the neuron ROIs. Essentially, the cells are expanded by a user-specified radius or distance. If they touch each other, they are part of the same ganglia. You can test this out using the Test_ganglia_segmentation_Hu_only script.

Original image Segmented neurons Cells dilated by 12 pixels Touching cells form individual ganglia
original segmented dilation ganglia

To perform the analysis, you select the Define ganglia using Hu option: ganglia seg using Hu

A good starting value is around 12 microns. It's still experimental, so please use it with caution and make sure you run the tests using Test_ganglia_segmentation_Hu_only before using it for your analysis.

You can checkout the tutorials on Youtube here.

Detecting multiple markers

Please note that GAT expects multichannel images when studying different markers. Within FIJI, this is called a composite image.

Analysing a single neuronal subtype

A common workflow in ENS research is to quantify the number of neurons, detect neuronal subtypes based on neurochemical coding and quantify the proportion of cells. We will use the sample image of the myenteric plexus of an adult mouse colon labelled with DAPI, nNOS (neuron subtype) and Hu (neuron). This image has:

  • Select Analyse Neuron (Multi-channel)

multiple_markers_1

  • Enter the channel number for Hu.
  • Tick Enter_channel_details_now so you can enter the name (nNOS) and corresponding channel no. (2).
  • Tick Calculate_Neuron_Subtype as we are going to analyse the no. of nNOS neurons and the proportion of Hu +ve neurons that are nNOS.
  • Tick Cell_counts_per_ganglia to calculate cell no. per ganglia. Click OK when done.

1. Segmentation of neurons and verification.

  • It is extremely important to verify the neuron segmentation and prediction. This will be used as a basis for predicting nNOS cells. Click OK when you’ve verified neuron ROIs.

2. Ganglia segmentation and verification.

The ganglia below has been corrected. Check this video for more information on how to correct the ganglia outline.

3. nNOS segmentation and verification

In this step, the neuron subtype model is used to predict nNOS positive cells. This model “over predicts” the no. of cells and suggests more nNOS candidates than the image has. GAT then uses the Hu prediction from above and verifies if the nNOS candidate is a neuron. If not, the nNOS cell is deleted.

  • Click OK when you’re done and the following table will be displayed:

multiple_markers_table

GAT will calculate:

  • No. of neuron
  • No. of ganglia
  • Neuron per ganglia
  • No. of nNOS neurons
  • No. of nNOS cells per ganglia

These results will be saved in the “Analysis” folder.

Analysing multiple neuronal subtypes

In many occasions, there is more than one neuronal subtype that needs to be analysed. Technically, there is no limit on the number of markers that can be analysed in GAT. For this workflow, we will use a dataset that GAT hasn't been trained on. We will use a sample dataset from the SPARC portal, which is also in the sample images folder. This image has the following channels:

  • Channel 1: ChAT (subtype)
  • Channel 2: nNOS (subtype)
  • Channel 3: VIP (subtype)
  • Channel 4: Hu (neuron)

For this workflow, we will look at the proportion of ChAT and nNOS neurons.

The order of the Channel names and the channel numbers must be the same. It can be entered like:

multiple_markers_8

OR

multiple_markers_9

The order should be the same in both rows. The channel for segmenting ganglia can be either nNOS or ChAT.

Video tutorial

As the analysis involves correcting ganglia outline and neuron numbers, it has been recorded as a video and published here (no audio):

Correct ganglia outline

The resulting table from the analysis will have:

  • No. of neuron
  • No. of ganglia
  • Neuron per ganglia
  • Marker combinations
    • No. of nNOS neurons
    • No. of ChAT neurons
    • No. of ChAT+nNOS neurons
  • No. of nNOS cells per ganglia
  • No. of ChAT cells per ganglia
  • No. of ChAT+nNOS cells per ganglia

It is possible that the model files may not work well for your set of images. If you find yourself making corrections to the cell detection and ganglia segmentation, you can use the corrections to train a model that works on your data. Please contact me if you’d like to generate models for your images. It is recommended you have approx. 8 annotated or corrected images for this. The StarDist2D notebooks from ZeroCostDL4Mic were used to generate the models.

Analysing multiple neuronal subtypes (without Hu) - experimental

In the latest release, we have added an option to analyse multi-channel images without Hu. It is still experimental, but it allows flexibility to analyse neuronal subtypes without the need to label all neurons.

The caveat is that the ganglia segmentation will not be accurate and may need significant manual adjustments. Alternatively, it may save time to manually define it from the start.

Dataset reference:

Clone this wiki locally