An intensity-based post-processing tool for 3D instance segmentation of organelles in soft X-ray tomograms.

Area of Science:

• Cell Biology
• Biophysics
• Imaging Science

Background:

• Understanding 3D organelle structures and dynamics is crucial for cellular function analysis.
• Soft X-ray tomography reveals complex subcellular organization and inter-organelle interactions.
• Accurate 3D instance segmentation of organelles remains a significant challenge.

Purpose of the Study:

• To develop and validate an intensity-based post-processing tool for accurate 3D organelle instance segmentation.
• To differentiate between sphere-like (insulin vesicles) and columnar (mitochondria) organelle instances.
• To analyze changes in organelle morphology and function in pancreatic beta-cells.

Main Methods:

• Developed an intensity-based post-processing tool utilizing raw tomogram intensity, semantic segmentation masks, and organelle morphology.
• Validated the tool on synthetic tomograms and experimental tomograms of pancreatic beta-cells.
• Compared the tool's performance against connected regions labeling, watershed, and watershed + Gaussian filter methods.

Main Results:

• The proposed tool demonstrated improved accuracy in identifying organelle instances in synthetic tomograms compared to existing methods.
• Enhanced description of organelle structures was achieved in beta-cell tomograms.
• Significant changes in insulin vesicle and mitochondrion volumes and intensities were observed under different experimental conditions.

Conclusions:

• The developed tool effectively separates and identifies individual organelle instances, particularly insulin vesicles and mitochondria.
• The findings reveal potential roles of insulin vesicles and mitochondria in maintaining normal beta-cell function.
• The tool is adaptable for instance segmentation in various cell types and imaging modalities.