TABLE 1.
List of all filters/features, shallow and deep learning, and label splitter/analysis options within SuRVoS2.
| Filters and features | Machine learning | Label splitter/Analysis |
|---|---|---|
| Produce output images for use in shallow or deep learning pipelines | Uses manual or shallow learning created training annotations to predict classes for other voxels | Inherent characteristics of data or segmented objects used to separate objects into groups |
| Basic Features: | Shallow Learning: | Mean Intensity |
| Simple Invert | Random Forest | Standard Deviation of Intensity |
| Invert Threshold | Extra Random Forest | Variation of Intensity |
| Threshold | Gradient Boosting | Volume |
| Rescale | Support Vector Machine (SVM) | Bounding Box Volume |
| Gamma Correct | Active Contour without Edges (ACWE) | Log Bounding Box Volume |
| Blob: | Watershed | Position X |
| Structure Tensor Determinant | Deep Learning: | Position Y |
| Frangi | 2D U-Net | Position Z |
| Hessian Eigenvalues | 3D U-Net a | Bounding Box Depth |
| Denoising: | FPN a | Bounding Box Height |
| Total Variation Denoise | — | Bounding Box Width |
| Gaussian Blur | — | Oriented Bounding Box Volume |
| Median | — | Log Oriented Bounding Box Volume |
| Wavelet | — | Oriented Bounding Box Depth |
| Edges: | — | Oriented Bounding Box Height |
| Spatial Gradient 3D | — | Oriented Bounding Box Width |
| Difference of Gaussians | — | — |
| Laplacian | — | — |
| Morphology: | — | — |
| Dilation | — | — |
| Erosion | — | — |
| Closing | — | — |
| Euclidean Distance Transform | — | — |
| Skeletonize | — | — |
| Neighborhood: | — | — |
| Gaussian Norm | — | — |
| Gaussian Centre | — | — |
a
Indicates options where training is currently done using the SuRVoS2 API externally to the graphical user interface (GUI), with the deep learning module for prediction available within the SuRVoS2 GUI.