ImportGeo-Vol already has a powerful AI-segmentation feature, where you can label a grayscale image to train a neural network to segment the image.

Since GeoDict 2025 it is also possible to train a neural network to Segment Grayscale Image using the GeoDict-AI module. Create training data that match the statistical properties of the scans you want to segment with the trained neural network. Given your training data, a neural network can be trained and applied with Segment Grayscale Image. After producing training data pairs each consisting of one gray value image (input.grw) and one segmented structure (output.gdt), organize the files in the folder structure described here.

There are two possibilities to generate the training data, explained in the following.

Create training data from real scans

Select representative scans having the same number of materials IDs after segmenting and needing the same image processing steps. For each of these scans do the following to create the training data:

1. Import the scan in ImportGeo-Vol.
2. Save the scan as it is as input.grw.
3. Apply image filters as needed.
4. Segment the scan for example using AI-segmentation.
5. Post-process the resulting structure to remove artifacts for example with ProcessGeo-Cleanse.
6. Save the resulting structure as output.gdt.

After all this effort you train a neural network that allows you to skip the time-consuming image processing steps for future scans to segment.

Create artificial training data

Use the GeoDict structure generators and transform the structures to gray value images using the GeoApp Generate Artificial CT-Scan. This app can add noise and many other kinds of image artifacts to the image, generating an image as close as possible to real gray value images. The created structure models in GeoDict should have the same statistical properties as the scans to segment. To ensure having representative training data, have a look at our checklist.

In the following it is described shortly how the User Guide example was created with GrainGeo and the Generate Artificial CT-Scan GeoApp.

Model the structure with GeoDict

Generate an artificial CT-Scan