Segment with AI

The Segment with AI feature lets you segment datasets with any of the deep learning or machine learning semantic segmentation models that you trained (see Training Deep Models for Semantic Segmentation and Machine Learning Segmentation for information about training segmentation models).

The Segment with AI panel, shown below, is available by default on the Segment tab.

Segment with AI

Segment with AI panel

Post-processing is applied automatically to improve the results of the pixel-wise predictions of deep models. To mitigate any jagged predictions on the borders of the output patches, the final 'prediction' or segmentation of a deep model is a mosaic of blended overlapped patches using a second order spline window function.

The following options are available on the Segment with AI panel.

Segment with AI options
  Description
Segmentation model Lets you choose a segmentation model. You should note that models are filtered based on the current selection and are sorted by the number of classes in the model.

Open Remote Library… Click the Open Remote Library button to import a ready-to-use deep model for denoising or super resolution (see Ready-to-Use Deep Models).

Inputs Indicates the currently selected input(s).
  • To change the selection, choose another item(s) in the Data Properties and Setting panel.
  • For multi-modality models, you change the order of the selected inputs by dragging an input, as shown below.

    Reorder inputs

    Note Input order must be consistent with model training.

  • For models with an input dimension of 2.5D or 3D, you can choose the reference slice and spacing.

    Note You can also import the settings from computed previews, if required.

  • Click the Edit Calibration button to normalize calibrated and uncalibrated datasets prior to inference (see Normalizing Data Ranges Prior to Training or Inference).
Classes Lists the classes in the selected model, as well as the color and name associated with each class.

Visible… If selected, the highlight applied to classified voxels will be shown on segmentation previews.

Color… Determines the highlight color applied to classified voxels. If required, you can change the highlight color (see Choosing Colors).

Name… Is the name assigned to the class label and the name that will appear in the output multi-ROI. If required, you can change the assigned name.
Preview Lets you generate previews using the currently selected segmentation model, as shown in the following illustration. You should note that previews are limited to the visible voxels in the selected 2D view.

Compute preview

The following options are available for segmentation previews:

  • You can choose different settings for the input(s). For example, input order for multi-modality models.
  • You can choose to make the highlight applied to segmented classes visible or not visible in the Classes box.
  • The default highlight color can be changed in the Classes box.
  • The opacity of the preview can be adjusted with the Preview opacity slider.

Note You can click the Delete button to remove the preview from the current view.
Segment Lets you automatically segment dataset(s) with the currently selected segmentation model, as shown below. The resulting multi-ROI will added to the Data Properties and Settings panel when processing is complete.

Full Dataset… If selected, the full dataset will be segmented.

Marked Slices… If selected, only marked slices will be segmented (see Marking Image Slices).

With ROI as Mask… If selected, only data within the region of interest selected as the mask will be processed (see Creating Mask ROIs).

With Shape as Mask… If selected, only data within the shape selected as the mask will be processed (see Shapes).
  1. Do the following, as required:
    • If you plan to process data only within a mask, generate the required ROI mask (see Creating Mask ROIs)
  2. Do one of the following:
    • Select the required input(s) in the Data Properties and Settings panel, right-click, and then choose Segment with AI in the pop-up menu.

      The selected input(s) appear in the Segment with AI panel.

    • Choose Artificial Intelligence > Segment with AI on the menu bar.

      You can then select the required input(s) in the Data Properties and Settings panel.

    Note Your selection must match the requirements for the segmentation model that you intend to use. For example, if the number of input datasets is two, then you must select two datasets to access the segmentation model.

    Note By default, the Segment with AI panel is available on the Segment tab on the left sidebar.

  3. Choose a segmentation model in the Segmentation Model drop-down menu.

    The Classes box is updated with information about the model output.

  4. Adjust the input settings, if required.

    For multi-modality models, you change the order of the selected inputs by dragging an input.

  5. Do the following to generate a preview, recommended:
    • Make any of the classes visible or invisible with the associated Visibility icon.

      Note You can also change the color assigned to the applied highlight (see Choosing Colors).

    • Click the Preview button.

      You can adjust the opacity of the preview results with the Preview Opacity slider.

  6. Evaluate the preview of the selected segmentation model.

    If the results are not satisfactory, you can change the segmentation model or retrain the model.

  7. Click the Segment button and then choose the required option in the drop-down menu.

    If you selected a masked option, choose the required mask in the Choose an ROI as Mask or Choose a Shape as Mask dialog, as shown below.

    When processing is complete, the resulting multi-ROI is added to the Data Properties and Settings panel (see Multi-ROI Properties and Settings).