# System Requirements

GAT has been tested on Windows (10, 11) and on Mac OS.

Large images may lead to out-of-memory errors during analysis. The size of your image (pixels) and the pixel size (microns/pixel) affect the resizing required for accurate neuronal segmentation predictions. The necessary hardware specifications depend on the size of your image datasets.

`The images are resized to a pixel size of 0.568 µm/px , so if your images have pixel sizes lower than this, images will be resized to a larger resolution -> making it bigger.`

## Hardware Requirements:

* Any recent Intel/AMD CPU should work
* RAM:
  *  Minimum of 8GB
  * Recommended: 32 GB or more.
* GPU: Not necessary, but it will make segmentation faster.[ Requires extra configuration](https://imagej.net/develop/tensorflow).

## Tests for neuronal segmentation

Summary of tests on sample images of different sizes for neuronal segmentation. Some of the example images are available [here](https://github.com/pr4deepr/GutAnalysisToolbox/tree/main/Sample%20Images). 

This was tested on a Windows 11 PC with 32 GB RAM without using a GPU. 

Benchmarking was done using scripts from **GAT -> Tools -> Test Neuron** **Probability.**

{% tabs %}
{% tab title="Small image" %}
Resolution: 1024 x 1024 pixels

Magnification: 40X

Pixel Size: 0.378 µm/px 

Rescaled resolution used segmentation: 683 x 683 pixels

No. of tiles: 1

RAM used (MB): 200 - 300 MB RAM

Example dataset: [ms\_distal\_colon\_Hu\_40X\_1.tif](https://github.com/pr4deepr/GutAnalysisToolbox/blob/main/Sample%20Images/ms_distal_colon_Hu_40X_1.tif)
{% endtab %}

{% tab title="Medium image" %}
Resolution: 2048x 2048 pixels

Magnification: 20X

Pixel Size: 0.283 µm/px 

Rescaled resolution used segmentation: 1024 x 1024 pixels

No. of tiles: 1

RAM used (MB): \~700 MB

Example dataset: [ms\_distal\_colon\_Hu\_20X.tif](https://github.com/pr4deepr/GutAnalysisToolbox/blob/main/Sample%20Images/ms_distal_colon_Hu_20X.tif)
{% endtab %}

{% tab title="Large image" %}
Resolution: 4856x 3672 pixels

Magnification: 20X

Pixel Size: 0.9098 µm/px 

Rescaled resolution used segmentation: 7780 x 5583 pixels

No. of tiles: 1 

RAM used (MB): Crashed; >32 GB

No. of tiles: 4

RAM used (MB): Maxed out at 32 GB -> successful segmentation

Example dataset: [Tilescan\_GAT\_ms\_distal\_colon\_MP\_hu.tif](https://github.com/pr4deepr/GutAnalysisToolbox/blob/main/Sample%20Images/Tilescan_GAT_ms_distal_colon_MP_hu.tif)
{% endtab %}
{% endtabs %}

Within GAT, the no. of tiles is dynamically adjusted based on image size to prevent running out of memory. 

Settings used for benchmarking are shown below. With the comparisons, the No. of tiles is the only setting that was altered which is documented above.

<figure><img src="https://460806082-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FUMXPxIZOpwi2uL22W18F%2Fuploads%2FoSm9PAGWHXcvaVcNGDpI%2Ftest_neuron_probability.png?alt=media&#x26;token=52510091-228e-424c-bdb5-8cd48e314509" alt=""><figcaption></figcaption></figure>