MFSE (MicroFossil Segmentation Editor): A GUI for Editing 3D Segmentations of Foraminifera Derived from X-Ray µCT

Input requirements and options

The software requires a binary shell segmentation as minimal input. A preliminary chamber lumen segmentation can be performed upon project creation. Alternatively, a chamber lumen segmentation acquired with another software can be imported into GIPL (Guy’s Image Processing Lab) or binary format.

GUI

The editing of generated or imported chamber lumen segmentation label fields is the main purpose of the software, and consequently, it was realized as a GUI. It was built in MATLAB (R2017b 9.3.0.713579) using the GUIDE GUI design environment. The architecture follows the classical Model-View-Controller pattern.

Model state is stored as an array of unsigned 16-bit integers (uint16) matching the input data, augmented by a trailing dimension of size four. Theoretically, up to 65k different labels could be stored across four layers. The underlying shell segmentation data is stored as a binary array and remains immutable to the software. Default processing and display parameters of the GUI can be loaded and saved into a user-specific configuration file. Basic input/output functionality, such as loading and saving of projects as well as macros and links to documentation and tutorials, is available via the menu bar. Callbacks validate user input and delegate computation to external utility functions. Modifications of the model state is updated to the view state after the completion of required calculations.

The GUI (Figure 1) layout is structured into eight panels: the label visualization or project view, the project information, view and select, camera controls, 3D environment visualization, basic shell segmentation, lumen operations, and label data. A toolbar menu provides the options for file import, export, and project creation. All GUI elements provide a short explanatory tooltip when the pointer is hovered over them. A user manual, providing concise explanations of all GUI components and available label operations, is accessible as a PDF via the Help option in the toolbar menu.

Figure 1

On the left side of the GUI, the segmentation label field is rendered in 3D and can be rotated and zoomed using the pointer or camera control tools. Labels may be selected individually or as groups using the pointer. The visualization style for each label can be chosen in the label data table (normal, smoothed, or voxel-based). It should be noted that the vectorized visualizations (normal and smoothed) represent interpretative surface renderings rather than exact representations of the underlying binary voxel data.

The project information panel displays essential details about the segmentation project, including data source file names, segmentation data dimensions and resolution, as well as a basic label overview and user comments. The view and select panel provides tools for controlling label visibility and selection through a multi-part label filter, along with buttons for commonly used selections. The 3D environment panel contains options for displaying labels within a 3D reference frame, including labelled axes and a grid. The basic shell segmentation panel allows the underlying imported shell segmentation to be visualized alongside the label field.

The lumen label operations panel consists of three alternatively active panels: general, lumen, and shell label operations. The general label panel includes operations applicable to all label types, while the lumen and shell panels provide functions restricted to their respective label classes. In general, labels can be split, merged, deleted, reduced to a single connected component (singularized), or morphologically simplified. Special label operations specific to lumen and shell classes are described in the following section.

The label data table contains comprehensive information for all labels within the current segmentation project. Users can modify label attributes such as name, morphounit, class, group, type, visibility, and visualization mode. The table also displays computed label properties, including the number of connected components (patches), volume, principal axis length, convex volume, and others. The displayed properties can be customized through the preferences menu in the toolbar.

Chamber lumen delimitation

Distinguishing between the interior and exterior of a chamber of a foraminifera shell with a typically asymmetrical shape and aperture, i.e., determining the chamber biovolume, the volume that is or could be occupied by the living cytoplasm, is a non-trivial problem with no single objectively correct solution. Here, we propose a mathematically reproducible approach based on the ambient occlusion algorithm. Ambient occlusion values (an in-depth explanation of the algorithm can be found in Titschack et al. [2]) are computed for all voxels of the selected label, using the basic shell segmentation as the occluding boundary. The threshold separating the interior from the exterior of a chamber is defined as the ambient occlusion value that maximizes the volume-to-surface-area ratio of the label within a specified range of threshold values. The resulting volume typically represents the chamber interior, bounded by a concave surface whose geometry is governed by the size and shape of the chamber aperture. In addition to in situ delimitation of chamber lumina, the software provides functionality to delineate chamber lumina in a hypothetical in vivo state of the specimen. Once all chamber lumen labels have been correctly enumerated and classified, and the corresponding primary shell labels have been generated (see below), the Delimit in vivo operation allows the user to delimit any chamber lumen as if that lumen were the ultimate, terminal chamber.

Individual chamber wall tracing

After completion of the editing and classification of the chamber lumina, the original shell segmentation can be subdivided into individual chamber shell walls. The underlying algorithm is inspired by the ambient occlusion approach and operates on label data in both voxel-based and vectorized representations. Similar to ambient occlusion, rays are projected from all face centers and vertices of a vectorized chamber lumen patch, following the direction of the corresponding face or vertex normal. Each ray is traced through the basic shell segmentation, and the labels of the intersected shell voxels are recorded. Applying this procedure to all face and vertex normals of a chamber lumen label yields a collection of voxel sequences that collectively represent the shell wall enclosing the corresponding chamber lumen. Factors that may interfere with accurate segmentation of a chamber wall, such as chamber walls merging at sutures, variations in shell thickness, shell porosity, and secondary calcification, can be partially mitigated by adjusting cutoff parameters for the extracted sequence lengths and by allowing interruptions within voxel sequences. In addition, this method enables chamber-specific quantification of secondary calcification.

Typical workflow

The typical complete workflow consists of seven steps (see also Figure 2).

Figure 2

• Project generation (import or generation of chamber lumen labels)

• Editing existing chamber lumen labels

• Classification of the chamber lumen labels

• Tracing of the primary chamber walls

• Editing of the primary chamber walls

• Delimitation of the in-vivo chamber lumina

• Finalization of the individual chamber walls

Several short video tutorials about the GUI, general project creation, and the overall recommended segmentation procedure are available in a YouTube channel (@MFSE-tutorials-123). In addition, the CT scan data and shell segmentation of a planktic foraminifera of the species Globigerinita glutinata are provided along the installation files.

Output

The generated segmentations can be exported in raw binary or GIPL format.