Skip to content

Image Generation

Multiscale

The Multiscale Image Generation module offers an interface with various parameters for manipulating and configuring the MPSlib library. MPSlib features a set of algorithms based on multiple point statistical (MPS) models inferred from a training image. Currently, only the Generalized ENESIM algorithm with direct sampling (DS) mode is available.

Panels and their use

Figure 1
Figure 1: Multiscale Image Generation Module.

Input Data

  • Training Image: Volume that acts as a training image.

  • Hard Data Image: Volume that acts as "Hard Data", where values and points are fixed during the simulation.

  • Mask Image: Volume that acts as a mask for selecting the simulation area. Unselected segments will not be included in the simulation.

Simulation Options

  • Wrap HD into cylinder: If the "Hard Data" is a well image (2D), this option causes the image to be considered a cylinder and performs simulations as 3D.

  • Data type: Determines whether the data type is continuous or categorical. Segmentations and Labelmaps can be used for discrete and continuous simulations, but scalar volumes can only be used for continuous.

  • Categorical: Segmentations control regions and determine the class value of the Hard Data and Training Image volume. Unselected segments will be disregarded.

  • Continuous: Segmentations control which regions and volume values will be used as Hard Data or training data. Unselected segments will be disregarded.

  • Grid Resolution: Voxel resolution of the resulting image (mm).

  • Grid Dimensions: Dimensions of the resulting image.

  • Automatic resampling: Activates the automatic resizing functionality of the input data to the simulation grid's dimension and resolution. If the image is an imagelog, Y-axis resizing is disabled.

    • Spacing: New axis resolution after resizing.
    • Size: New axis dimension after resizing.
    • Ratio: Ratio of the new voxel resolution to the initial resolution.

  • Number of patches: Breaks the simulation grid in several parts to simulate separetaly, improving performance and reducing overall simulation time for big images. Only available for 2D images.

Parameters

  • Number of Conditioning points: Number of conditioning points to be used in each iteration.

  • Number of realizations: Number of simulations and images to be generated.

  • Number of iterations: Maximum number of iterations when searching the training image. Use value -1 to run a full scan on the training image.

  • Random seed: Initial value used to start the simulation. The same seed with the same parameters always generates the same result. Value 0 will generate a random seed for the simulation.

  • Colocate dimensions: For a 3D simulation, ensure that the order in the last dimensions is important, allowing a 2D co-simulation with conditional data in the third dimension.

  • Max search radius: Only conditional data within a defined radius is used as conditioning data.

  • Max distance (normalized): The maximum distance that will lead to the acceptance of a conditional model match. If the value is 0, a perfect match will be sought.

  • Distance power: Weights the conditioning data based on the distance from the central values. A value of 0 configures no weighting.

Output Options

  • Output prefix: Name of the generated volume or file. In case of multiple realizations, a number is added to indicate the realization.

  • Save: Options for saving the results.

  • First realization: Saves only the first realization as an individual volume.
  • All realizations: Saves all realizations as individual volumes.
  • As sequence: Saves the realizations in a sequence set. The "_proxy" output volume indicates it is a sequence and has the controllers for visualizing the realizations.

  • TIF files: Saves all realizations as tiff files.

  • Export directory: Directory where the tiff files will be saved. It is only enabled if the "TIF files" option is selected.

Buttons

  • Run: Executes the MPS. If number of patches is 1, the parallel MPS code will be executed. Otherwise, the sequential MPS code will be executed for each patch and realization."
  • Cancel: Interrupts the simulation execution.

SinGAN Module

The SinGAN Module provides an interface for using SinGAN models within GeoSlicer...

Panels and their usage

Figure 1
Figure 1: SinGAN module presentation.

Parameters

Model Configuration:

  • Select Model: Select the model to be used. Only valid models are listed, within the SinGAN model directory configured by the user.

  • Create TI from model: Reconstructs the training image as a volume within GeoSlicer

Conditioning Image

  • Hard Data Image: Selects the conditioning image to be used during generation.

  • Choose scale to use HD: If selected, the conditioning image is injected only at the chosen scale. If disabled, the image is injected at the first available scale and then resized for subsequent scales.

  • Create Hard Data preview: Creates a preview of the hard data image at the injection scales.

BIG IMAGE:

  • Selected method: Method used for generating large images.

    1. Generation patch on gpu: All processing is performed entirely in RAM and on the GPU, without splitting into patches or writing anything to disk.
    2. Pacth Inference: Generates the image by processing it in small pieces (patches), using the disk to store intermediate data. It is ideal for very large images that do not fit in RAM.
    3. Early crop: Generates the image recursively, resizing the input image for each scale and extracting the corresponding patch.
    4. By chunks: Divides the image into larger blocks (chunks) defined by the user, processing each one individually. It is an approach to handle large images, offering a balance between memory usage and processing time.

  • Set number of chunk: Defines the number of blocks in each dimension in the By chunks method.

Output

  • Number of realizations: Number of images to be generated.

  • Output Prefix: Name of the generated volumes

  • Save: How the simulated image will be saved in GeoSlicer.

    1. As Large Image node: An large image node will be added automaticaly to geoslicer. Refer to Big Image for more information about this node.
    2. As volume: An individual volume is generated for each realization. Available when there is enough memory to load output image.
    3. As sequence: A sequence of volumes containing all realizations is generated. This option is only available for more than one realization and when the option to save as volume is enabled.
    4. As NetCDF files: The generated images are automatically saved as .nc files. Obligatory for big images.
  • Export directory: Directory where the NetCDF images will be saved.